Dose Distribution Prediction for Hepatocellular Carcinoma Using Convolutional Neural Networks from Diagnostic CT and MRI: Focus on Passive and Intensity-Modulated Proton Therapy

Dose Distribution Prediction for Hepatocellular Carcinoma Using Convolutional Neural Networks from Diagnostic CT and MRI: Focus on Passive and Intensity-Modulated Proton Therapy | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Dose Distribution Prediction for Hepatocellular Carcinoma Using Convolutional Neural Networks from Diagnostic CT and MRI: Focus on Passive and Intensity-Modulated Proton Therapy Toshiya Rachi, Taku Tochinai This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6397967/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 04 Aug, 2025 Read the published version in Radiation Oncology → Version 1 posted 9 You are reading this latest preprint version Abstract Background Proton therapy is commonly used for hepatocellular carcinoma (HCC). However, its feasibility can be challenging to assess large tumors or those adjacent to critical organs at risk (OARs), as these factors are typically evaluated only after treatment planning. This study aimed to predict proton dose distributions utilizing diagnostic computed tomography (dCT) and magnetic resonance (MR) images, leveraging a convolutional neural network (CNN), to enable early treatment feasibility assessments before planning CT (pCT) acquisition. Methods This study calculated dose distributions and dose-volume histograms (DVH) for 118 patients with HCC using intensity-modulated proton therapy (IMPT) and passive proton irradiation. The CNN model predicted the DVH and dose distributions, which were evaluated using mean absolute error (MAE), mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM). Results The predicted DVHs closely matched the actual DVHs. MAE was consistently below 0.03, with passive proton therapy achieving values between 0.012 and 0.018, indicating high consistency. MSE remained below 0.004 for all cases, confirming clinically acceptable accuracy. PSNR ranged from 24 to 28 dB. SSIM was above 0.94 for both IMPT and passive proton therapy, with the lowest value being 0.841, indicating high structural similarity. Conclusions This study demonstrates the potential of diagnostic imaging in optimizing the workflow for HCC proton therapy planning. The proposed CNN-based model enables early dose distribution predictions, reducing the need for pCT acquisition and improving treatment decision-making. Hepatocellular carcinoma proton therapy convolutional neural network dose distribution diagnostic imaging intensity-modulated proton therapy passive methods Figures Figure 1 Figure 2 Figure 3 Background Advancements in high-precision radiation techniques for hepatocellular carcinoma (HCC) have enabled precise delivery to target volumes while minimizing exposure to healthy tissues. Stereotactic body radiation therapy (SBRT) has been shown to achieve local control rates of over 85% in patients with early-stage HCC [ ]. Similarly, proton therapy has revealed superiority in sparing normal liver tissues while delivering a high dose to the tumor, reducing the risk of radiation-induced liver disease (RILD) in patients with poor liver function [ ]. Radiation therapy is often considered when other standard treatments, such as resection, are not feasible. For example, portal vein tumor thrombosis (PVTT) often limits surgical options, as it can significantly impair liver function and increase the risk of postoperative liver failure. Similarly, patients with advanced comorbidities, such as cirrhosis or severe cardiac conditions, may not tolerate the stress of surgery. These limitations highlight the importance of radiation therapy as a less invasive yet effective alternative for achieving local control in HCC [ , ]. Radiation therapy for HCC is viewed as an adaptive modality for achieving local control. Effective balance between liver damage and dose distribution is critical for successful treatment. Intensity-modulated proton therapy (IMPT) or passive irradiation with a bolus and collimator formation are used in proton therapy [ , , ] and allow precise targeting of the tumor while minimizing the dose to surrounding liver tissues and critical structures. This flexibility helps reduce the RILD. IMPT allows for highly localized dose delivery to the tumor by modulating the intensity of proton beams, reducing the radiation dose to healthy liver tissue. The use of bolus and collimator formations in passive proton therapy further enhances dose conformity and sparing of healthy tissues, making it particularly beneficial for tumors near critical organs. However, radiation therapy may not be feasible when large tumors or critical organs at risk (OARs) are near the target, making it impossible to meet dose constraints for the normal liver or other OARs, even with planning CT imaging. In such cases, patient radiation exposure from planning CT imaging and the staff effort required for treatment planning may become unjustified. Recent advancements in automated treatment planning using convolutional neural networks (CNN) have shown promise in addressing the challenges of dose distribution prediction [ , ]. Algorithms like U-Net and its derivatives have been widely used to predict dose distributions. However, they often require expensive equipment with Graphics Processing Unit (GPU) capabilities [ , ], which has limited their widespread adoption in many clinical settings. This study aimed to predict dose distributions using CNNs with a general-purpose central processing unit (CPU) from existing diagnostic CT (dCT) and magnetic resonance image (MRI) taken before the implementation of planning CT. This enables the possibility of executing dose distribution predictions on commonly available terminals, such as those used for electronic medical records, allowing radiation treatment dose distributions to be presented to patients and facilitating decision-making without relying on planning CT. This approach can help reduce unnecessary patient exposure and alleviate the workload of treatment planning staff, positioning it as an application of diagnostic imaging in proton therapy. Methods Data acquisition The study included 118 patients diagnosed with HCC and fewer than three metastases who underwent passive proton therapy between 2021 and 2024. The prescribed doses were categorized into three groups: 66 Gy/10 fractions for the peripheral type, 72.6–76 Gy/20–22 fractions for the hilar type, and 74–76 Gy/37–38 fractions for the gastrointestinal proximity type [ , , ]. For all patients, IMPT plans were created with the same prescribed doses as for passive irradiation. All plans were designed to ensure that the D90% of the Clinical Target Volume (CTV) was covered by 100% of the prescribed dose. Different treatment planning systems (TPSs) were employed for each irradiation type: Eclipse (ver. 16.0, Varian Medical Systems, Inc.) for IMPT and an in-house developed SGI_TPS (ver. 2.0, Sumitomo Heavy Industries, Ltd.) for passive proton irradiation plans. Next, contours were drawn on the dCT and MRI for each patient. The target structures, Gross Tumor Volume (GTV) and CTV were automatically transferred from the targets delineated on the planning CT images used in clinical practice to the dCT and MRI using deformable image registration (DIR) technology with MIM Maestro (version 7.2.9, MIM Software, Inc.). OARs were contoured using an atlas-based model with DIR. All contours were subsequently reviewed by a radiation oncologist. The contours used in this study included GTV, CTV, normal liver, duodenum, and bile duct. These patient data were divided into 100 training sets and 18 test sets. Prediction of Dose Volume Histogram and Dose Distribution using CNN Overview of the CNN Model for DVH As the initial step in predicting the dose distribution, a 1-dimensional convolutional neural network (1D CNN) was used to predict the DVH. This approach improved the accuracy of the predicted dose distribution by first predicting the DVH and incorporating it into the prediction flow of the dose distribution. For each patient, the prescribed dose [in Gy] and the ROIs delineated on the planning CT were used to extract the following 10 geometric parameters: prescribed dose, liver segment number, major axis axis [in cm] of the GTV, volume [in cm³] of the GTV, the proportion of the liver occupied by the GTV [in %], dice coefficient between the liver and GTV, Hausdorff distance [in cm] and mean distance to an agreement [in cm] between the liver and GTV, shortest distance [in cm] between the GTV and the duodenum, and shortest distance [in cm] between the GTV and the bile duct. These parameters were used to create explanatory variables as Train_item. Additionally, the DVH data for each OAR, calculated from the dose distributions generated by different irradiation techniques on the planning CT, were gathered and added as Train_DVH. By using both Train_item and Train_DVH, a comprehensive input was constructed to train the model. Similarly, nine geometric parameters were derived from the contours on the dCT and MR images, and these were used as Test_item. By inputting Test_item, a model, CNN1, was developed to predict the DVH (Output_DVH) for the test patients. The architecture is shown in Fig. 1A. In a clinical setting, this approach would allow physicians to contour the target and OARs on dCT or MRI, enabling the display of the corresponding DVH. CNN Architecture for DVH The architecture of the CNN1 model, designed for DVH prediction, comprises an input layer followed by several convolutional layers and dense layers. The input layer accepts a one-dimensional vector representation of the explanatory variables, which is reshaped into a 2D format suitable for convolutional operations. The convolutional layers utilize a kernel size of 1, allowing the model to focus on element-wise transformations and preserve spatial resolution. Each convolutional layer employs the Rectified Linear Unit (ReLU) activation function to introduce non-linearity, followed by max-pooling layers to reduce the dimensionality of the feature maps and improve computational efficiency. Flattening is applied to transform the extracted feature maps into a fully connected format, which is then passed through multiple dense layers with gradually decreasing numbers of neurons to refine the learned features progressively. The final output layer comprises a linear activation function to predict the DVH values directly. The model's optimizer, Adam, is used with a learning rate of 0.001, balancing convergence speed and stability during training. The loss function is a mean squared error (MSE), as it is well-suited for regression tasks, and accuracy is employed as an additional metric to evaluate performance. Additionally, the model was designed to leverage commonly available computational resources by utilizing a 1D CNN architecture, which is computationally less demanding compared to 3D architectures. This approach ensures accessibility for clinical implementation without requiring specialized hardware. The training process and predictions were conducted using Python libraries such as TensorFlow/Keras. Evaluate the predicted DVH To evaluate the predicted DVHs for the normal liver, duodenum, and bile duct, dose differences between the predicted and actual DVHs (calculated using the TPS) were measured at 10% intervals across the range from 10–100% of the prescription dose. Additionally, clinically important metrics derived from the DVHs, including V30 [%] for the normal liver and D1cc [%] for the duodenum and bile duct, were compared between the predicted and actual values. The V30 metric represents the proportion of liver volume receiving at least 30 Gy and is critical for assessing the risk of RILD. The D1cc metrics for the duodenum and bile duct indicate the maximum dose delivered to the most exposed 1 cm³ volume, providing insight into the potential for radiation-induced toxicity. To determine whether statistically significant differences exist between predicted and actual values, paired two-tailed t-tests were performed for each dose point at 10% intervals and for clinical indices. The t-test was selected because it is a standard statistical method for comparing means of paired samples under the assumption that the differences follow a normal distribution. A p-value threshold of ≥ 0.05 was considered statistically significant. Overview of the CNN Model for Dose Distribution In the next step, Train_sum was created by concatenating the Train_item from step 2.2.1 with the Train_DVH for the three ROIs. Using the RT_dose data in the Digital Imaging and Communications in Medicine (DICOM) format from the treatment plans derived by the TPS, the intensity and position information of the dose distribution were obtained. These were used as Train_dose, and both Train_sum and Train_dose were used as explanatory variables to construct CNN2. By inputting the Test_item from the test group and the Output_DVH predicted by CNN1, a 3D dose distribution (Output_Dose) that corresponds to the predicted DVH was generated. The architecture of CNN2 is shown in Fig. 1B. Using CNN1 and CNN2, this study derives the dose distribution from the positional information of contours obtained from images. Therefore, it enables the prediction of dose distribution and DVH from dCT and MRI, regardless of image signal values, allowing for treatment feasibility assessment. CNN Architecture for Dose Distribution The architecture of CNN2 was designed to integrate both geometric and dosimetric information effectively. The input layer for Train_sum, derived from the DVH and geometric parameters, is reshaped and flattened before being concatenated with the geometric features from Train_item. The convolutional layers use ReLU as the activation function due to its computational efficiency and effectiveness in handling non-linearity. The output layer consists of three separate dense layers corresponding to the x, y, and z components of the 3D dose distribution. Each dense layer uses a linear activation function to produce continuous predictions for each dimension. The model is trained using the Adam optimizer, selected for its adaptive learning rate and computational efficiency, with a learning rate of 0.001. The loss function is the MSE, chosen to penalize larger errors more heavily, and accuracy metrics are calculated for each output dimension (x, y, z). Evaluate the predicted Dose Distribution We predicted the dose distributions for the patients in the test group using CNN1 and CNN2. To evaluate the accuracy of these predictions, the predicted and actual dose distributions were compared using the cumulative dose projection (CDP). CDP represents the sum of dose distributions across the axial, sagittal, and coronal planes, providing a comprehensive visualization of the dose distribution. The mean absolute error (MAE), MSE, peak signal-to-noise ratio (PSNR) [ , ], and structural similarity index (SSIM) [ ] were calculated for each CDP. The formulas for MAE and MSE are defined as follows: \(\:MAE=\:\frac{1}{N}{\sum\:}_{i=1}^{N}\left|{y}_{i}-{\widehat{y}}_{i}\right|\) ・・・・・ (1) \(\:MSE=\:\frac{1}{N}{\sum\:}_{i=1}^{N}{\left({y}_{i}-{\widehat{y}}_{i}\right)}^{2}\) ・・・・・ (2) where N is the total number of pixels in the dose distribution map, \(\:{y}_{i}\) is the actual dose distribution and \(\:{\widehat{y}}_{i}\) is the predicted dose distribution. The formulas for calculating PSNR and SSIM are shown in Equations (3) and (4): \(\:PSNR=10*{\text{log}}_{10}\left(\frac{{MAX}^{2}}{MSE}\right)\) ・・・・・ (3) \(\:SSIM(x,y)=\:\frac{\left(2{\mu\:}_{x}{\mu\:}_{y}+{C}_{1}\right)\left(2{\sigma\:}_{xy}+{C}_{2}\right)}{\left({\mu\:}_{x}^{2}+{\mu\:}_{y}^{2}+{C}_{1}\right)\left({\sigma\:}_{x}^{2}+{\sigma\:}_{y}^{2}+{C}_{2}\right)}\) ・・・・・ (4) Here, “MAX” represents the maximum value of the dose distribution (normalized to 1). \(\:{\mu\:}_{x}\:\text{a}\text{n}\text{d}\:{\mu\:}_{y}\:\) are the mean values of the actual dose distribution image arrayམand the predicted dose distribution image array \(\:y\) , respectively. \(\:{\sigma\:}_{x}^{2}\:\text{a}\text{n}\text{d}\:{\sigma\:}_{y}^{2}\) are the variances of \(\:x\) and \(\:y\) , respectively, and \(\:{\sigma\:}_{xy}\) is the covariance. C1 and C2 are stabilization constants. By combining these metrics, we provided a robust and transparent evaluation of the predicted dose distributions. MAE and MSE quantify the absolute and squared differences, respectively, offering straightforward error measures. PSNR emphasizes the relative magnitude of the errors, while SSIM captures perceptual similarity by considering the structural and statistical properties of the dose distribution. Results Predicted DVH Figure 2 presents examples of DVHs predicted using IMPT and passive proton techniques for two patients. It compares the DVHs of the normal liver, duodenum, and bile duct predicted from dCT and MRI contour information with the actual calculated DVHs. The predicted DVHs visually closely matched the actual measurements. Tables 1 and 2 compare the predicted and calculated DVHs for all test patients. They describe the volume differences at 10% dose intervals for everyone, dose differences at D1cc [%], and normal liver V30 [%]. The tables summarize the mean values, standard deviations, and p-values for the predicted indicators. IMPT and Passive predicted from dCT are denoted as dCT_IMPT and dCT_Passive, respectively, while those predicted from MRI are referred to as MRI_IMPT and MRI_Passive. For dCT_IMPT for the normal liver, the maximum difference was 1.84 ± 2.75% at the 100% dose, with an average difference of 2% and a standard deviation of 3%. For the duodenum, the maximum difference was 0.38 ± 2.60% at the 10% dose, with an average difference within 1% and a standard deviation within 3%. For the bile duct, the maximum difference was 0.95 ± 3.31% at a 30% dose, with an average difference within 1% and a standard deviation within 3% above 40%. The D1cc [%] ranged from − 1.28% to -4.51% (standard deviation, 4.29–5.76%). For dCT_Passive, the maximum difference in the normal liver was 0.17 ± 2.40% at the 40% dose, with an average difference within 1% and a standard deviation within 3%. For the duodenum, the maximum difference was 1.21 ± 4.24% at the 10% dose, with an average difference within 2% and a standard deviation within 2% above 20%. For the bile duct, the maximum difference was 0.69 ± 5.02% at the 20% dose, with an average difference within 1% and a standard deviation within 2% above 30%. The D1cc [%] ranged from 1.05% to -2.21% (standard deviation, 2.97–9.84%). For MRI_IMPT, the normal liver showed a maximum difference of 0.38 ± 2.25% at the 100% dose, with an average difference within 1% and a standard deviation within 3%. For the duodenum, the maximum difference was 0.31 ± 1.15% at the 60% dose, with an average difference within 1% and a standard deviation within 2%. For the bile duct, the maximum difference was 1.04 ± 4.89% at the 20% dose, with an average difference within 2% and a standard deviation within 3% above 70%. The D1cc [%] ranged from 0.12% to -4.58% (standard deviation, 2.26–5.20%). For MRI_Passive, the normal liver exhibited a maximum difference of 0.16 ± 2.47% at the 100% dose, with an average difference within 1% across all doses and a standard deviation within 3%. For the duodenum, the maximum difference was 1.38 ± 4.39% at the 10% dose, with an average difference within 2% and a standard deviation greater than 2% above 20%. For the bile duct, the maximum difference was 1.13 ± 4.36% at the 20% dose, with an average difference under 2% and a standard deviation above 30%. The D1cc [%] ranged from − 0.12% to -1.37% (standard deviation, 2.18–2.90%). The DVHs predicted based on contours delineated on dCT and MRI showed no significant differences in the average DVH differences at 10% dose intervals or in the normal liver V30 [%] when compared to those derived from planning CT for both IMPT and passive proton therapy techniques. However, significant p-values (< 0.05) were observed in the D1cc [%] dose differences for some OARs. Predicted dose The comparison of dose distributions was performed using CDP, calculated in the axial, sagittal, and coronal planes, by comparing the predicted and calculated values. Figure 3 displays the three-dimensional dose distributions of IMPT and passive proton therapy predicted from contours on both dCT and MRI, along with the CDP in the axial plane. The fused images of the predicted dose distributions demonstrate that the targets are well covered across all combinations of dCT and MRI, as well as IMPT and passive proton therapy. In the axial CDP calculated from both the predicted and calculated dose distributions, signal values along the same lines on both the X- and Y-axes were observed to be generally consistent. The mean values and standard deviations of MAE, MSE, PSNR, and SSIM derived from the CDPs for all test patients are summarized in Table 3. The MAE remained below 0.03 across all results. In particular, for passive proton therapy, the MAE ranged from 0.012 to 0.018, demonstrating highly consistent reproducibility. The MSE was also generally below 0.004, indicating an error level that is clinically acceptable. For passive proton therapy, the MSE was below 0.003 across all views, confirming high accuracy. For IMPT, the highest MSE was observed in the coronal direction for both MRI and dCT. The standard deviation in the sagittal direction for MRI_IMPT was slightly higher, highlighting the need for caution due to variability between data sets. The PSNR ranged from 24 to 28 dB, indicating clinically acceptable reproducibility. Passive proton therapy showed high PSNR values across all directions, suggesting a high similarity to the original dose distributions. On the other hand, for IMPT, the dose distributions predicted from both dCT and MRI tended to show slightly lower PSNR values in the coronal direction, indicating a slight decline in reproducibility. Based on the SSIM results, passive proton therapy demonstrated high structural similarity across all directions, with values exceeding 0.94, indicating excellent structural reproducibility. In contrast, IMPT showed a slightly lower trend overall, with the coronal direction suggesting slightly reduced structural reproducibility. However, SSIM values remained above 0.89 under all conditions, which can be considered clinically acceptable. Overall, these results demonstrate that the proposed prediction method can achieve highly accurate and reproducible dose distributions for both IMPT and passive proton therapy. The findings particularly highlight the robustness of passive proton therapy predictions. Additionally, the entire process—from automatic contouring to DVH prediction and dose distribution completion—can be completed in approximately 30 minutes. This means that by delineating contours on existing images, the predicted dose distribution can be presented to the patient within 30 minutes, allowing for a timely assessment of treatment feasibility. Discussion In this study, a 1D CNN model in two steps was developed based on contours from two modality images, existing DVH, and dose distributions. The process involved predicting the DVH and dose distribution. We envisioned this model to assist oncologists during initial consultations, using a standard CPU-based PC for displaying medical records. While 2D or higher-dimensional CNNs may enhance prediction accuracy, they require longer computation times and expensive PCs equipped with GPUs [ ]. This was the reason for adopting the 1D CNN model. However, handling large-volume 3D images such as CT and MRI can potentially degrade PC performance, so it is essential to minimize the dataset size as much as possible [ ]. In this study, we successfully predicted dose distributions without using the full 3D image data by relying solely on geometric values—such as target size and the distances between the target and OARs—extracted from contour information. Additionally, the RT_Dose, a 3D array used as the ground truth, was reduced to one-fourth of its original size in each dimension. While this downscaling posed a risk of reducing prediction accuracy, our results demonstrated that high accuracy was still achieved. Furthermore, in predicting dose distributions from contour data, it was found that predicting the DVH first and then using it to guide dose distribution prediction resulted in smaller errors for both the DVH and the final dose distribution compared to directly predicting the dose distribution and calculating the DVH afterward. Based on Tables 2 and 3, comparisons between the ground truth and the DVHs of OARs predicted from contours delineated on dCT and MRI for both IMPT and passive methods showed no statistically significant differences across the dose range from 10–100% of the prescription dose. However, for D1cc [%], the mean error ranged from − 4.5% to + 1.0% across all cases, with large variations in standard deviation, resulting in statistically significant differences. These discrepancies are thought to arise in small high-dose regions exceeding the prescription dose, which are known to be random in nature and difficult to predict accurately [ ]. Nguyen et al. conducted a study using a U-Net-based deep learning model to predict dose distributions for prostate cancer IMRT plans from organ contour information. They reported that the mean absolute differences for Dmax and Dmean were less than 5% of the prescribed dose for both PTV and OARs. Our similar results suggest that our findings are comparable to those from GPU-based studies [ ]. Moreover, an error of < 5% in D1cc is unlikely to pose a significant issue when making clinical decisions about whether to proceed with treatment. However, achieving closer agreement may be possible by incorporating information that strongly influences low-dose distributions—such as beam angles from the accelerator side—and by considering patient-specific hyperparameters, which could further improve prediction accuracy [ ][ ]. Next, we discuss the prediction accuracy of dose distributions using MAE, MSE, PSNR, and SSIM metrics across irradiation techniques (IMPT and passive proton therapy) and imaging modalities (MRI and dCT). The MAE values ranged from 0.012 to 0.030 across all directions, with the smallest error observed in the sagittal direction for dCT_Passive (0.012) and the largest in the coronal direction for MRI_IMPT (0.030). Considering that the acceptable range of dose measurement error in radiotherapy quality assurance is generally within ± 3%, these values suggest favorable prediction accuracy [ ]. For MSE, values ranged from 0.002 ± 0.001 to 0.004 ± 0.002, with the best results observed for dCT_IMPT and dCT_Passive in the axial and sagittal directions. These ranges are considered highly favorable for similarity evaluations within the imaging field, indicating high prediction accuracy across all irradiation types. Huai-Wen Zhang et al. conducted dose distribution predictions for liver SBRT, reporting MSE values ranging from 0.0004 to 0.008 compared to the ground truth, which is nearly equivalent to the results obtained in this study [ ]. PSNR is a quantitative metric used to evaluate image reproducibility, with a clinically acceptable accuracy generally considered to fall within the range of 25–30 dB [ ]. The PSNR results in this study ranged from 24 to 28 dB, confirming that the predicted dose distributions achieved a clinically sufficient level of reproducibility. Passive proton therapy exhibited the highest PSNR across all directions, indicating the most accurate prediction performance. In contrast, IMPT showed a slight decrease in PSNR in the coronal direction, which may be attributed to the characteristics of the spot-scanning technique. IMPT generates complex dose distributions through energy modulation, leading to steeper dose gradients and increased local variations. Consequently, minor reductions in reproducibility between the predicted and reference dose distributions may occur. Conversely, passive proton therapy employs fixed-port irradiation, resulting in fewer low-dose regions and a more uniform dose distribution. This characteristic likely contributes to consistently higher PSNR values. SSIM values were also high, ranging from 0.898 to 0.959. Passive techniques achieved the highest SSIM scores (0.943–0.959), indicating excellent structural similarity to the reference dose distribution. IMPT followed with values ranging from 0.898 to 0.940, which also indicates a high degree of similarity. While SSIM values between 0.85 and 0.90 are considered to reflect good agreement, some localized visual deviations may still be observed. Overall, the SSIM values exceeded 0.9 in most cases, indicating good structural similarity [ ]. Our study demonstrates that both IMPT and passive techniques achieve high prediction accuracy for dose distributions based on diagnostic imaging, with passive techniques consistently exhibiting superior performance. Furthermore, MRI-based predictions showed accuracy comparable to those derived from dCT and the ground truth calculated from planning CT images, highlighting the feasibility of MRI-only workflows. Additionally, our CNN implementation is CPU-based and designed for deployment on widely available PCs, such as those used for electronic medical records. Given the limited number of studies on proton dose distribution predictions from existing dCT and MRI images, our system enables real-time dose distribution estimation during a patient's initial consultation, allowing for immediate treatment feasibility assessments. This is particularly beneficial for proton therapy, where treatment facilities are fewer than those for X-ray therapy, making it advantageous for patients who must travel long distances. Furthermore, predicting dose distributions from existing diagnostic images can assist in determining whether planning CT should be performed for patients with large hepatocellular carcinoma, where the feasibility of proton therapy is uncertain due to trade-offs with OARs. This approach may also help reduce the workload of treatment planning staff. While this study primarily focuses on generating predicted dose distributions for proton therapy, future research aims to store these distributions in DICOM format for integration into treatment planning systems. This advancement could enable the direct replacement of predicted dose distributions with actual irradiation machine outputs. Our previous research has demonstrated the feasibility of implementing this approach as a novel inverse dose calculation algorithm for Volumetric modulated arc therapy, and we aim to extend its application to proton therapy [ ]. Conclusions Our study evaluated the feasibility of proton therapy for HCC before acquiring planning CT by providing predicted dose distributions during the initial consultation. The proposed prediction model demonstrated high accuracy and reproducibility in both DVH and dose distribution predictions in proton therapy. The high SSIM and PSNR values, as well as the favorable MSE results, confirm the strong agreement between predicted and reference dose distributions. This approach has the potential to reduce unnecessary radiation exposure and support the selection of appropriate treatment strategies, particularly for patients with advanced tumors where the trade-off with OAR constraints is critical. By integrating irradiation angles and patient-specific anatomical features into the training process and optimizing the loss function, further improvements in prediction accuracy are expected. Additionally, our findings highlight the clinical utility of diagnostic imaging in radiation therapy, demonstrating its potential to streamline the workflow from initial consultation to treatment planning. This could reduce the time required for CT-based planning, enabling oncologists to assess treatment feasibility earlier in the process and improving efficiency for both patients and staff. Moreover, implementing our CNN-based prediction model on widely available CPU-based systems, such as those used in electronic medical records, allows real-time dose estimation at the point of care. This could be particularly beneficial in proton therapy, where limited treatment facilities often require patients to travel long distances, potentially reducing the number of hospital visits. Finally, expanding this approach to other cancer types, leveraging advancements in diagnostic imaging, and incorporating emerging machine-learning techniques could further enhance its clinical utility. These prospects underscore the broad applicability of our research and provide a clear direction for advancing radiation therapy practices. Abbreviations HCC: Hepatocellular carcinoma, CNN: Convolutional neural networks, CT: Computed tomography, MRI: Magnetic resonance imaging, VMAT: Volumetric modulated arc therapy, IMPT: Intensity-modulated proton therapy, DVH: Dose-volume histogram, MAE: Mean absolute error, MSE: Mean squared error, PSNR: Peak signal-to-noise ratio, SSIM: Structural similarity index, DICOM: Digital imaging and communications in Medicine Declarations Ethical approval and consent to participate This study is a retrospective analysis of radiotherapy outcomes using existing patient data. Ethical approval for the use of these data was granted by the Ethics Committee of the National Cancer Center Hospital, Japan (Approval No. 2020-272, dated 12 October 2020). All procedures were carried out in accordance with the Declaration of Helsinki and relevant institutional guidelines." Consent for publication Written informed consent was obtained from the patient(s) for the use of personal data and images in this publication. The patient(s) were fully informed about the purpose of the study, the intended use of their data and images, and their rights to privacy. The consent form is retained by the corresponding author and is available upon request, should it be necessary for legal or ethical purposes. Additionally, this information can be referenced on the National Cancer Center Hospital East website: https://www.ncc.go.jp/jp/ncce/index.html. Availability of data and materials Research data are stored in an institutional repository and will be made available upon request to the corresponding author. Competing interest The authors declare that they have no competing interests. Funding Statement This study was supported by JSPS KAKENHI (Proposal No. 21K07741). Authors contributions TR: Writing, Review & Editing, Software Development, Data curation Contouring, Formal analysis, and Visualization. TT: Review & Editing, Patient Selection, Contouring, and Dose Distribution Creation. Acknowledgments No acknowledgments. References Takeda A, Sanuki N, Tsurugi Y, Iwabuchi S, Matsunaga K, Ebinuma H, et al. Phase 2 study of stereotactic body radiotherapy and optional transarterial chemoembolization for solitary hepatocellular carcinoma not amenable to resection and radiofrequency ablation. Cancer. 2016;13:2041-49. Available from: https://doi.org/10.1002/cncr.30008. Yu-Lun T, Takei H, Lizumi T, Okumura T, Sekino Y, Numajiri H, et al. Capacity of proton beams in preserving normal liver tissue during proton beam therapy for hepatocellular carcinoma. J Radiat Res. 2020;62:133–41. Available from: https://doi.org/ 10.1093/jrr/rraa098. Korean Liver Cancer Study Group, National Cancer Center Korea. Practice guidelines for management of hepatocellular carcinoma 2009. Korean J Hepatol. 2009;15:391-423. Benson AB III, D'Angelica MI, Abbott DE, Abrams RP, Alberts MJ, Saenz AS, et al. NCCN guidelines insights: hepatobiliary cancers. J Natl Compr Canc Netw. 2017;15:563-73. Kalogeridi MA, Zygogianni A, Kyrgias G, Kouvaris J, Chatziioannou S, Kelekis N, et al. Role of radiotherapy in the management of hepatocellular carcinoma: a systematic review. World J Hepatol. 2015;7:101-12. Available from: https://doi.org/10.4254/wjh.v7.i1.101. Nakayama H, Sugahara S, Tokita M, Fukuda K, Mizumoto M, Abei M, et al. Proton beam therapy for hepatocellular carcinoma: the University of Tsukuba experience. Cancer. 2009;115:5499-506. Available from: https://doi.org/10.1002/cncr.24619. Fukumitsu N, Sugahara S, Nakayama H, Fukuda K, Mizumoto M, Abei M, et al. A prospective study of hypofractionated proton beam therapy for patients with hepatocellular carcinoma. Int J Radiat Oncol Biol Phys. 2009;74:831-6. Available from: https://doi.org/10.1016/j.ijrobp.2008.10.073. Ronneberger O, Fischer P, Brox T. U-net: convolutional networks for biomedical image segmentation. In: Lecture Notes in Computer Science. Springer; 2015. p. 234-41. Available from: https://doi.org/10.1007/978-3-319-24574-4_28. He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2016. p. 770-8. Available from: https://doi.org/10.1109/CVPR.2016.90. Liu Y, Chen Z, Wang J, Wang X, Qu B, Ma L, et al. Dose prediction using a three-dimensional convolutional neural network for nasopharyngeal carcinoma with tomotherapy. Front Oncol. 2021;11:752007. Available from: https://doi.org/10.3389/fonc.2021.752007. Kearney V, Chan JW, Haaf S, Descovich M, Solberg TD. DoseNet: a volumetric dose prediction algorithm using 3D fully-convolutional neural networks. Phys Med Biol. 2018;63:235022. Available from: https://doi.org/10.1088/1361-6560/aaef74. Mizumoto M, Okumura T, Hashimoto T, Fukuda K, Oshiro Y, Fukumitsu N, et al. Proton beam therapy for hepatocellular carcinoma: a comparison of three treatment protocols. Int J Radiat Oncol Biol Phys. 2011;81:1039-45. Available from: https://doi.org/10.1016/j.ijrobp.2010.07.015. Mizumoto M, Tokuuye K, Sugahara S, Nakayama H, Fukumitsu N, Ohara K, et al. Proton beam therapy for hepatocellular carcinoma adjacent to the porta hepatis. Int J Radiat Oncol Biol Phys. 2008;71:462-7. Available from: https://doi.org/10.1016/j.ijrobp.2007.09.056. Kawashima M, Furuse J, Nishio T, Konishi M, Ishii H, Kinoshita T, et al. Phase II study of radiotherapy employing proton beam for hepatocellular carcinoma. J Clin Oncol. 2005;23:1839-46. Available from: https://doi.org/10.1200/JCO.2005.00.620. Robeson SM, Willmott CJ. Decomposition of the mean absolute error (MAE) into systematic and unsystematic components. PLOS ONE. 2023;18:e0279774. Available from: https://doi.org/10.1371/journal.pone.0279774. Tanabe Y, Ishida T. Quantification of the accuracy limits of image registration using peak signal-to-noise ratio. Radiol Phys Technol. 2017;10:91-4. Available from: https://doi.org/10.1007/s12194-016-0372-3. Ma C, Wang R, Zhou S, Wang M, Yue H, Zhang Y, et al. The structural similarity index for IMRT quality assurance: radiomics-based error classification. Med Phys. 2021;48:80-93. Available from: https://doi.org/10.1002/mp.14559. Sahiner B, Pezeshk A, Hadjiiski LM, Wang X, Drukker K, Cha KH, et al. Deep learning in medical imaging and radiation therapy. Med Phys. 2019;46:e1-e36. Available from: https://doi.org/10.1002/mp.13264. Lin Y, Liu Y, Chen H, Yang X, Ma K, Zheng Y, et al. LENAS: learning-based neural architecture search and ensemble for 3-D radiotherapy dose prediction. IEEE Trans Cybern. 2024;PP. Available from: https://doi.org/10.1109/TCYB.2024.3390769. Ahn SH, Kim E, Kim C, Cheon W, Kim M, Lee SB, et al. Deep learning method for prediction of patient-specific dose distribution in breast cancer. Radiat Oncol. 2021;16:154. Available from: https://doi.org/10.1186/s13014-021-01864-9. Nguyen D, Long T, Jia X, Lu W, Gu X, Iqbal Z et al. A feasibility study for predicting optimal radiation therapy dose distributions of prostate cancer patients from patient anatomy using deep learning. Sci. Rep. 2019;9:1–10. Available from: https://doi.org/10.1038/s41598-018-37741-x. Bai X, Zhang J, Wang B, Wang S, Xiang Y, Hou Q. Sharp loss: a new loss function for radiotherapy dose prediction based on fully convolutional networks. Biomed Eng Online. 2021;20:101. Available from: https://doi.org/10.1186/s12938-021-00937-w. Shamsi A, Asgharnezhad H, Mohammadi P, Soofi R. An uncertainty-aware loss function for training neural networks with calibrated predictions. Mach Learn Comput Vis Pattern Recognit. Available from: https://doi.org/10.48550/arXiv.2110.03260. Fraass B, Doppke K, Hunt M, Kutcher G, Starkschall G, Stern R, et al. American Association of Physicists in Medicine Radiation Therapy Committee Task Group 53: quality assurance for clinical radiotherapy treatment planning. Med Phys. 1998;25:1773-829. Available from: https://doi.org/10.1118/1.598373. Zhang HW, Wang YH, Hu B, Pang HW. Uninvolved liver dose prediction in stereotactic body radiation therapy for liver cancer based on the neural network method. World J Gastrointest Oncol. 2024;16:4146–56. Available from: https://doi.org/10.4251/wjgo.v16.i10.4146. Kobako M. Image compression guidelines for digital documents. JIIMA Standardization Committee, Vice Chair (JIS). Available from: https://www.jiima.or.jp/pdf/5_JIIMA_guideline.pdf. Wang Z, Bovik AC, Sheikh HR, Simoncelli EP. Image quality assessment: from error visibility to structural similarity. IEEE Trans Image Process. 2004;13:600-12. Available from: https://ieeexplore.ieee.org/document/1284395. Rachi T, Parshuram RV, Tanaka Y, Togo H. Examination of conversion method of dose distribution of lung cancer IMRT using fluence reversible calculation function in O-ring type linac and C-type linac. Phys Eng Sci Med. 2022;45:559-67. Available from: https://doi.org/10.1007/s13246-022-01122-6. Tables Table 1. Predicted Dose Volume Histogram (DVH) Metrics for dCT Across Irradiation Techniques: Mean, Standard Deviation, and p-Values for ROIs dCT IMPT 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% D1cc V30 Liver 0.17 0.07 0.23 0.32 0.49 0.60 0.66 0.78 1.24 1.84 -4.54 0.58 SD 2.65 2.33 2.04 1.89 2.05 2.13 2.13 2.27 2.78 2.75 4.67 1.71 p-value 0.79 0.90 0.65 0.50 0.34 0.26 0.22 0.17 0.08 0.06 0.00 0.18 Duodenum 0.38 0.00 -0.32 -0.05 0.18 0.37 0.34 0.16 -0.04 -0.23 -1.64 SD 2.60 1.95 1.18 0.74 0.80 1.79 1.35 0.68 0.16 0.95 2.97 p-value 0.54 0.99 0.28 0.79 0.34 0.40 0.30 0.33 0.33 0.33 0.03 Bile duct 0.28 0.00 0.95 1.17 0.34 0.00 0.10 0.13 0.50 0.42 -1.01 SD 3.05 4.16 3.31 2.32 2.31 1.76 1.73 0.92 1.88 1.56 5.57 p-value 0.71 1.00 0.25 0.05 0.56 1.00 0.81 0.55 0.28 0.28 0.45 Passive 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% D1cc V30 Liver 0.17 0.25 0.27 0.17 0.21 -0.04 -0.20 -0.25 -0.29 -0.08 -2.21 1.05 SD 1.72 1.96 2.01 2.40 1.62 1.39 0.95 1.00 1.13 1.68 2.97 1.56 p-value 0.68 0.60 0.58 0.78 0.61 0.92 0.40 0.32 0.30 0.85 0.01 0.01 Duodenum 1.21 0.14 -0.25 -0.49 -0.66 -0.55 -0.51 -0.24 -0.10 -0.01 1.05 SD 4.24 1.66 1.12 1.41 1.87 1.79 1.68 0.70 0.39 0.02 9.84 p-value 0.26 0.74 0.37 0.17 0.16 0.23 0.23 0.18 0.29 0.33 0.48 Bile duct 0.04 0.69 0.24 -0.39 -0.62 -0.37 -0.21 -0.10 -0.11 0.24 -1.98 SD 3.16 5.02 1.23 2.17 1.84 1.38 1.22 0.94 1.02 0.65 3.29 p-value 0.96 0.58 0.44 0.47 0.18 0.29 0.48 0.68 0.66 0.15 0.03 Table 2. Predicted Dose Volume Histogram (DVH) Metrics for MRI Across Irradiation Techniques: Mean, Standard Deviation, and p-Values for ROIs MRI IMPT 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% D1cc V30 Liver -0.33 -0.04 0.68 0.60 0.40 0.35 0.26 0.18 0.26 0.38 -4.58 0.57 SD 1.75 1.91 1.78 1.39 1.61 1.86 1.76 1.70 1.68 2.25 5.20 2.09 p-value 0.47 0.93 0.16 0.12 0.35 0.48 0.57 0.69 0.56 0.52 0.00 0.31 Duodenum 0.04 -0.50 -0.45 -0.19 0.03 0.31 0.26 0.04 -0.11 -0.39 -0.67 SD 0.54 1.36 0.97 0.71 0.10 1.15 0.96 0.16 0.41 1.47 2.26 p-value 0.79 0.20 0.11 0.34 0.34 0.34 0.34 0.34 0.34 0.34 0.29 Bile duct -0.78 1.04 1.59 1.07 -0.27 -1.00 -0.80 -0.42 0.31 0.52 0.12 SD 1.87 4.89 3.45 3.94 3.70 3.53 2.31 1.11 0.90 1.13 3.93 p-value 0.14 0.44 0.11 0.33 0.79 0.31 0.22 0.18 0.23 0.11 0.92 Passive 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% D1cc V30 Liver -0.12 0.04 0.26 -0.13 0.24 0.19 0.01 -0.25 -0.34 0.16 -1.37 -0.12 SD 1.35 1.52 1.72 1.07 0.68 0.59 0.69 1.09 1.41 2.47 2.18 2.16 p-value 0.74 0.92 0.57 0.65 0.19 0.24 0.97 0.39 0.37 0.80 0.03 0.83 Duodenum 1.38 0.10 -0.18 -0.25 -0.37 -0.39 -0.09 0.00 0.00 0.00 -0.82 SD 4.39 1.35 0.43 0.67 1.18 1.40 0.35 0.01 0.01 0.01 2.69 p-value 0.24 0.79 0.12 0.17 0.25 0.30 0.32 0.34 0.34 0.34 0.22 Bile duct 0.52 1.13 -0.04 -0.33 -0.53 -0.50 -0.38 -0.12 -0.06 0.18 -0.12 SD 3.97 4.36 1.63 2.45 2.08 1.64 0.95 0.75 0.42 0.77 2.90 p-value 0.62 0.33 0.92 0.61 0.34 0.25 0.14 0.54 0.62 0.38 0.87 The table presents the mean, standard deviation, and p-values of volume differences at each 10% dose interval for the true and predicted DVHs of test patients for dCT (Table 1) and MRI (Table 2). It also includes the dose differences for D1cc [%] and normal liver V30 [%]. Table 3. Quantitative Evaluation Metrics for Predicted Dose Distribution Maps MRI_IMPT dCT_IMPT MRI_Passive dCT_Passive MAE axial average 0.024 0.024 0.012 0.013 SD 0.008 0.007 0.004 0.004 sagittal average 0.018 0.016 0.014 0.012 SD 0.007 0.007 0.006 0.006 coronal average 0.030 0.028 0.017 0.018 SD 0.009 0.008 0.006 0.008 MSE axial average 0.003 0.003 0.002 0.002 SD 0.001 0.001 0.001 0.001 sagittal average 0.003 0.002 0.003 0.002 SD 0.002 0.001 0.001 0.001 coronal average 0.004 0.004 0.003 0.003 SD 0.002 0.001 0.001 0.002 PSNR axial average 25.798 25.744 27.265 26.972 SD 2.055 2.086 2.003 1.925 sagittal average 26.346 26.814 26.200 27.742 SD 2.644 1.946 2.409 2.547 coronal average 24.190 24.870 25.812 25.978 SD 1.657 1.592 1.920 2.499 SSIM axial average 0.920 0.919 0.957 0.955 SD 0.026 0.021 0.014 0.012 sagittal average 0.937 0.940 0.954 0.959 SD 0.020 0.018 0.016 0.017 coronal average 0.898 0.902 0.945 0.943 SD 0.025 0.021 0.020 0.024 CDPs in the axial, sagittal, and coronal directions were calculated from the dose distributions predicted for each irradiation type based on the dCT and MR images of all patients. To evaluate consistency from the CDPs, MAE, MSE, PSNR, and SSIM were calculated, and their mean values and standard deviations are shown. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 04 Aug, 2025 Read the published version in Radiation Oncology → Version 1 posted Editorial decision: Revision requested 03 May, 2025 Reviews received at journal 03 May, 2025 Reviews received at journal 02 May, 2025 Reviewers agreed at journal 16 Apr, 2025 Reviewers agreed at journal 15 Apr, 2025 Reviewers invited by journal 09 Apr, 2025 Editor assigned by journal 08 Apr, 2025 Submission checks completed at journal 08 Apr, 2025 First submitted to journal 07 Apr, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6397967","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":440611985,"identity":"44c3b6a3-0e46-4002-ba92-5521798b44e4","order_by":0,"name":"Toshiya Rachi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3klEQVRIie3QvQrCMBDA8ZMMXU5dhULzClcCTlJHX6MS6FxwFQwIugiu3XyGvkFKoZPYtdBJfISCk4MfnbpE3UTyWwLh/hwJgGX9KAY06d6QaRzbJHod3ySQdxOjqZufmzguveG+1CxeBsBnChaxacsgEm5CtRhVElhSSPCPGkRiShDGDKmeq4oB6ysNfhKCQGPiXB/JaXUo848TfG7RIWnZJnz0LjniwkWSflpJyrGQSJgp41ucnZM2eAu4V2bnCy4Dj2/XhTD9WId+riXd24hPixZXwC7fJZZlWX/uDlOkPjkafcOAAAAAAElFTkSuQmCC","orcid":"","institution":"National Cancer Center Hospital East","correspondingAuthor":true,"prefix":"","firstName":"Toshiya","middleName":"","lastName":"Rachi","suffix":""},{"id":440611986,"identity":"5e03b87d-237f-4968-b2a6-f0a2c9dc4726","order_by":1,"name":"Taku Tochinai","email":"","orcid":"","institution":"Japan Agency for Medical Research and Development (AMED)","correspondingAuthor":false,"prefix":"","firstName":"Taku","middleName":"","lastName":"Tochinai","suffix":""}],"badges":[],"createdAt":"2025-04-08 01:08:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6397967/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6397967/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s13014-025-02708-6","type":"published","date":"2025-08-04T15:57:15+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":80712243,"identity":"81fd4c9b-c66a-4a89-bd2c-b4a17b8073a5","added_by":"auto","created_at":"2025-04-16 09:14:00","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":870084,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow of CNN Models for Predicting DVH and Dose Distribution\u003c/p\u003e\n\u003cp\u003e(A) Flow diagram illustrating CNN1 for predicting the DVH. The model uses one-dimensional input data, Train_item, as the input layer. In the intermediate layer, Train_item is reshaped into a single-column and passed through Conv1D layers followed by MaxPooling1D layers to extract features. These features are then flattened into a vector. This flattened vector is passed through seven fully connected layers to refine the representation and predict the DVH. The output layer generates a one-dimensional prediction, which is reshaped into a two-dimensional data structure to match the original DVH format. For testing, Test_item is used to generate DVH predictions. (B) Flow diagram illustrating CNN2, which predicts the dose distribution using outputs from CNN1. The input layer uses two data sources: the two-dimensional input data, Train_item. In the intermediate layer, Train_DVH is flattened and concatenated with Train_item to form a combined feature vector. The combined data is reshaped and processed through Conv1D and MaxPooling1D layers to extract features. Dose distribution patterns are refined through six fully connected layers. The final output is reshaped to reconstruct the spatial dose distribution. By inputting Test_item from the test group and the Output_DVH predicted by CNN1 , CNN2 predicts the three-dimensional dose distribution (Output_Dose) for the test group patients.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6397967/v1/faf4e337115d713815e67b95.jpeg"},{"id":80712241,"identity":"e0b606c5-abf4-4ced-976d-820c1d3a33cb","added_by":"auto","created_at":"2025-04-16 09:14:00","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":866866,"visible":true,"origin":"","legend":"\u003cp\u003eActual (orange) and predicted (blue) DVHs for the normal liver, duodenum, and bile duct. These were calculated using two irradiation techniques on the dCT and MR images of two representative patients. The DVHs show good agreement, confirming that the predicted values closely match the true values visually.\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6397967/v1/f09899cce83ce463c0d1f727.jpeg"},{"id":80712233,"identity":"44431908-62fd-4b5b-9321-39a1c8212658","added_by":"auto","created_at":"2025-04-16 09:14:00","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1987738,"visible":true,"origin":"","legend":"\u003cp\u003ePredicted dose distribution maps for the same patients in figure 2. These three-dimensional dose distributions were predicted using IMPT and passive irradiation techniques on dCT and MRI. The CDPs were calculated by integrating dose distributions across the axial, sagittal, and coronal planes. The true and predicted values are compared on the X and Y axes at the marked lines in the images.\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6397967/v1/5c318eb24123387e7f1505ca.jpeg"},{"id":88814226,"identity":"c124d702-bc41-4ba0-80b4-951fd5303bd1","added_by":"auto","created_at":"2025-08-11 16:08:44","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6222950,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6397967/v1/768cf3ce-edcf-4d29-b89e-d14bf48e9ff5.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Dose Distribution Prediction for Hepatocellular Carcinoma Using Convolutional Neural Networks from Diagnostic CT and MRI: Focus on Passive and Intensity-Modulated Proton Therapy","fulltext":[{"header":"Background","content":"\u003cp\u003eAdvancements in high-precision radiation techniques for hepatocellular carcinoma (HCC) have enabled precise delivery to target volumes while minimizing exposure to healthy tissues. Stereotactic body radiation therapy (SBRT) has been shown to achieve local control rates of over 85% in patients with early-stage HCC [\u003ca class=\"FNLink\" href=\"#Fn1\" id=\"#FNLinkFn1\"\u003e\u003c/a\u003e]. Similarly, proton therapy has revealed superiority in sparing normal liver tissues while delivering a high dose to the tumor, reducing the risk of radiation-induced liver disease (RILD) in patients with poor liver function [\u003ca class=\"FNLink\" href=\"#Fn2\" id=\"#FNLinkFn2\"\u003e\u003c/a\u003e]. Radiation therapy is often considered when other standard treatments, such as resection, are not feasible. For example, portal vein tumor thrombosis (PVTT) often limits surgical options, as it can significantly impair liver function and increase the risk of postoperative liver failure. Similarly, patients with advanced comorbidities, such as cirrhosis or severe cardiac conditions, may not tolerate the stress of surgery. These limitations highlight the importance of radiation therapy as a less invasive yet effective alternative for achieving local control in HCC [\u003ca class=\"FNLink\" href=\"#Fn3\" id=\"#FNLinkFn3\"\u003e\u003c/a\u003e,\u003ca class=\"FNLink\" href=\"#Fn4\" id=\"#FNLinkFn4\"\u003e\u003c/a\u003e].\u003c/p\u003e \u003cp\u003eRadiation therapy for HCC is viewed as an adaptive modality for achieving local control. Effective balance between liver damage and dose distribution is critical for successful treatment. Intensity-modulated proton therapy (IMPT) or passive irradiation with a bolus and collimator formation are used in proton therapy [\u003ca class=\"FNLink\" href=\"#Fn5\" id=\"#FNLinkFn5\"\u003e\u003c/a\u003e, \u003ca class=\"FNLink\" href=\"#Fn6\" id=\"#FNLinkFn6\"\u003e\u003c/a\u003e, \u003ca class=\"FNLink\" href=\"#Fn7\" id=\"#FNLinkFn7\"\u003e\u003c/a\u003e] and allow precise targeting of the tumor while minimizing the dose to surrounding liver tissues and critical structures. This flexibility helps reduce the RILD. IMPT allows for highly localized dose delivery to the tumor by modulating the intensity of proton beams, reducing the radiation dose to healthy liver tissue. The use of bolus and collimator formations in passive proton therapy further enhances dose conformity and sparing of healthy tissues, making it particularly beneficial for tumors near critical organs.\u003c/p\u003e \u003cp\u003eHowever, radiation therapy may not be feasible when large tumors or critical organs at risk (OARs) are near the target, making it impossible to meet dose constraints for the normal liver or other OARs, even with planning CT imaging. In such cases, patient radiation exposure from planning CT imaging and the staff effort required for treatment planning may become unjustified.\u003c/p\u003e \u003cp\u003eRecent advancements in automated treatment planning using convolutional neural networks (CNN) have shown promise in addressing the challenges of dose distribution prediction [\u003ca class=\"FNLink\" href=\"#Fn8\" id=\"#FNLinkFn8\"\u003e\u003c/a\u003e,\u003ca class=\"FNLink\" href=\"#Fn9\" id=\"#FNLinkFn9\"\u003e\u003c/a\u003e]. Algorithms like U-Net and its derivatives have been widely used to predict dose distributions. However, they often require expensive equipment with Graphics Processing Unit (GPU) capabilities [\u003ca class=\"FNLink\" href=\"#Fn10\" id=\"#FNLinkFn10\"\u003e\u003c/a\u003e,\u003ca class=\"FNLink\" href=\"#Fn11\" id=\"#FNLinkFn11\"\u003e\u003c/a\u003e], which has limited their widespread adoption in many clinical settings.\u003c/p\u003e \u003cp\u003eThis study aimed to predict dose distributions using CNNs with a general-purpose central processing unit (CPU) from existing diagnostic CT (dCT) and magnetic resonance image (MRI) taken before the implementation of planning CT. This enables the possibility of executing dose distribution predictions on commonly available terminals, such as those used for electronic medical records, allowing radiation treatment dose distributions to be presented to patients and facilitating decision-making without relying on planning CT. This approach can help reduce unnecessary patient exposure and alleviate the workload of treatment planning staff, positioning it as an application of diagnostic imaging in proton therapy.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData acquisition\u003c/h2\u003e \u003cp\u003eThe study included 118 patients diagnosed with HCC and fewer than three metastases who underwent passive proton therapy between 2021 and 2024. The prescribed doses were categorized into three groups: 66 Gy/10 fractions for the peripheral type, 72.6\u0026ndash;76 Gy/20\u0026ndash;22 fractions for the hilar type, and 74\u0026ndash;76 Gy/37\u0026ndash;38 fractions for the gastrointestinal proximity type [\u003ca class=\"FNLink\" href=\"#Fn12\" id=\"#FNLinkFn12\"\u003e\u003c/a\u003e, \u003ca class=\"FNLink\" href=\"#Fn13\" id=\"#FNLinkFn13\"\u003e\u003c/a\u003e, \u003ca class=\"FNLink\" href=\"#Fn14\" id=\"#FNLinkFn14\"\u003e\u003c/a\u003e].\u003c/p\u003e \u003cp\u003eFor all patients, IMPT plans were created with the same prescribed doses as for passive irradiation. All plans were designed to ensure that the D90% of the Clinical Target Volume (CTV) was covered by 100% of the prescribed dose. Different treatment planning systems (TPSs) were employed for each irradiation type: Eclipse (ver. 16.0, Varian Medical Systems, Inc.) for IMPT and an in-house developed SGI_TPS (ver. 2.0, Sumitomo Heavy Industries, Ltd.) for passive proton irradiation plans.\u003c/p\u003e \u003cp\u003eNext, contours were drawn on the dCT and MRI for each patient. The target structures, Gross Tumor Volume (GTV) and CTV were automatically transferred from the targets delineated on the planning CT images used in clinical practice to the dCT and MRI using deformable image registration (DIR) technology with MIM Maestro (version 7.2.9, MIM Software, Inc.). OARs were contoured using an atlas-based model with DIR. All contours were subsequently reviewed by a radiation oncologist. The contours used in this study included GTV, CTV, normal liver, duodenum, and bile duct. These patient data were divided into 100 training sets and 18 test sets.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePrediction of Dose Volume Histogram and Dose Distribution using CNN\u003c/h3\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eOverview of the CNN Model for DVH\u003c/h2\u003e \u003cp\u003eAs the initial step in predicting the dose distribution, a 1-dimensional convolutional neural network (1D CNN) was used to predict the DVH. This approach improved the accuracy of the predicted dose distribution by first predicting the DVH and incorporating it into the prediction flow of the dose distribution.\u003c/p\u003e \u003cp\u003eFor each patient, the prescribed dose [in Gy] and the ROIs delineated on the planning CT were used to extract the following 10 geometric parameters: prescribed dose, liver segment number, major axis axis [in cm] of the GTV, volume [in cm\u0026sup3;] of the GTV, the proportion of the liver occupied by the GTV [in %], dice coefficient between the liver and GTV, Hausdorff distance [in cm] and mean distance to an agreement [in cm] between the liver and GTV, shortest distance [in cm] between the GTV and the duodenum, and shortest distance [in cm] between the GTV and the bile duct. These parameters were used to create explanatory variables as Train_item. Additionally, the DVH data for each OAR, calculated from the dose distributions generated by different irradiation techniques on the planning CT, were gathered and added as Train_DVH. By using both Train_item and Train_DVH, a comprehensive input was constructed to train the model.\u003c/p\u003e \u003cp\u003eSimilarly, nine geometric parameters were derived from the contours on the dCT and MR images, and these were used as Test_item. By inputting Test_item, a model, CNN1, was developed to predict the DVH (Output_DVH) for the test patients. The architecture is shown in Fig.\u0026nbsp;1A. In a clinical setting, this approach would allow physicians to contour the target and OARs on dCT or MRI, enabling the display of the corresponding DVH.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eCNN Architecture for DVH\u003c/h3\u003e\n\u003cp\u003eThe architecture of the CNN1 model, designed for DVH prediction, comprises an input layer followed by several convolutional layers and dense layers. The input layer accepts a one-dimensional vector representation of the explanatory variables, which is reshaped into a 2D format suitable for convolutional operations. The convolutional layers utilize a kernel size of 1, allowing the model to focus on element-wise transformations and preserve spatial resolution. Each convolutional layer employs the Rectified Linear Unit (ReLU) activation function to introduce non-linearity, followed by max-pooling layers to reduce the dimensionality of the feature maps and improve computational efficiency. Flattening is applied to transform the extracted feature maps into a fully connected format, which is then passed through multiple dense layers with gradually decreasing numbers of neurons to refine the learned features progressively. The final output layer comprises a linear activation function to predict the DVH values directly. The model's optimizer, Adam, is used with a learning rate of 0.001, balancing convergence speed and stability during training. The loss function is a mean squared error (MSE), as it is well-suited for regression tasks, and accuracy is employed as an additional metric to evaluate performance.\u003c/p\u003e \u003cp\u003eAdditionally, the model was designed to leverage commonly available computational resources by utilizing a 1D CNN architecture, which is computationally less demanding compared to 3D architectures. This approach ensures accessibility for clinical implementation without requiring specialized hardware. The training process and predictions were conducted using Python libraries such as TensorFlow/Keras.\u003c/p\u003e\n\u003ch3\u003eEvaluate the predicted DVH\u003c/h3\u003e\n\u003cp\u003eTo evaluate the predicted DVHs for the normal liver, duodenum, and bile duct, dose differences between the predicted and actual DVHs (calculated using the TPS) were measured at 10% intervals across the range from 10\u0026ndash;100% of the prescription dose. Additionally, clinically important metrics derived from the DVHs, including V30 [%] for the normal liver and D1cc [%] for the duodenum and bile duct, were compared between the predicted and actual values. The V30 metric represents the proportion of liver volume receiving at least 30 Gy and is critical for assessing the risk of RILD. The D1cc metrics for the duodenum and bile duct indicate the maximum dose delivered to the most exposed 1 cm\u0026sup3; volume, providing insight into the potential for radiation-induced toxicity.\u003c/p\u003e \u003cp\u003eTo determine whether statistically significant differences exist between predicted and actual values, paired two-tailed t-tests were performed for each dose point at 10% intervals and for clinical indices. The t-test was selected because it is a standard statistical method for comparing means of paired samples under the assumption that the differences follow a normal distribution. A p-value threshold of \u0026ge;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eOverview of the CNN Model for Dose Distribution\u003c/h2\u003e \u003cp\u003eIn the next step, Train_sum was created by concatenating the Train_item from step 2.2.1 with the Train_DVH for the three ROIs. Using the RT_dose data in the Digital Imaging and Communications in Medicine (DICOM) format from the treatment plans derived by the TPS, the intensity and position information of the dose distribution were obtained. These were used as Train_dose, and both Train_sum and Train_dose were used as explanatory variables to construct CNN2. By inputting the Test_item from the test group and the Output_DVH predicted by CNN1, a 3D dose distribution (Output_Dose) that corresponds to the predicted DVH was generated. The architecture of CNN2 is shown in Fig.\u0026nbsp;1B.\u003c/p\u003e \u003cp\u003eUsing CNN1 and CNN2, this study derives the dose distribution from the positional information of contours obtained from images. Therefore, it enables the prediction of dose distribution and DVH from dCT and MRI, regardless of image signal values, allowing for treatment feasibility assessment.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eCNN Architecture for Dose Distribution\u003c/h3\u003e\n\u003cp\u003eThe architecture of CNN2 was designed to integrate both geometric and dosimetric information effectively. The input layer for Train_sum, derived from the DVH and geometric parameters, is reshaped and flattened before being concatenated with the geometric features from Train_item. The convolutional layers use ReLU as the activation function due to its computational efficiency and effectiveness in handling non-linearity.\u003c/p\u003e \u003cp\u003eThe output layer consists of three separate dense layers corresponding to the x, y, and z components of the 3D dose distribution. Each dense layer uses a linear activation function to produce continuous predictions for each dimension.\u003c/p\u003e \u003cp\u003eThe model is trained using the Adam optimizer, selected for its adaptive learning rate and computational efficiency, with a learning rate of 0.001. The loss function is the MSE, chosen to penalize larger errors more heavily, and accuracy metrics are calculated for each output dimension (x, y, z).\u003c/p\u003e\n\u003ch3\u003eEvaluate the predicted Dose Distribution\u003c/h3\u003e\n\u003cp\u003eWe predicted the dose distributions for the patients in the test group using CNN1 and CNN2. To evaluate the accuracy of these predictions, the predicted and actual dose distributions were compared using the cumulative dose projection (CDP). CDP represents the sum of dose distributions across the axial, sagittal, and coronal planes, providing a comprehensive visualization of the dose distribution. The mean absolute error (MAE), MSE, peak signal-to-noise ratio (PSNR) [\u003ca class=\"FNLink\" href=\"#Fn15\" id=\"#FNLinkFn15\"\u003e\u003c/a\u003e,\u003ca class=\"FNLink\" href=\"#Fn16\" id=\"#FNLinkFn16\"\u003e\u003c/a\u003e], and structural similarity index (SSIM) [\u003ca class=\"FNLink\" href=\"#Fn17\" id=\"#FNLinkFn17\"\u003e\u003c/a\u003e] were calculated for each CDP. The formulas for MAE and MSE are defined as follows:\u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:MAE=\\:\\frac{1}{N}{\\sum\\:}_{i=1}^{N}\\left|{y}_{i}-{\\widehat{y}}_{i}\\right|\\)\u003c/span\u003e \u003c/span\u003e ・・・・・ (1)\u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:MSE=\\:\\frac{1}{N}{\\sum\\:}_{i=1}^{N}{\\left({y}_{i}-{\\widehat{y}}_{i}\\right)}^{2}\\)\u003c/span\u003e \u003c/span\u003e ・・・・・ (2)\u003c/p\u003e \u003cp\u003ewhere \u003cem\u003eN\u003c/em\u003e is the total number of pixels in the dose distribution map, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{y}_{i}\\)\u003c/span\u003e\u003c/span\u003e is the actual dose distribution and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\widehat{y}}_{i}\\)\u003c/span\u003e\u003c/span\u003e is the predicted dose distribution.\u003c/p\u003e \u003cp\u003eThe formulas for calculating PSNR and SSIM are shown in Equations (3) and (4):\u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:PSNR=10*{\\text{log}}_{10}\\left(\\frac{{MAX}^{2}}{MSE}\\right)\\)\u003c/span\u003e \u003c/span\u003e ・・・・・ (3)\u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:SSIM(x,y)=\\:\\frac{\\left(2{\\mu\\:}_{x}{\\mu\\:}_{y}+{C}_{1}\\right)\\left(2{\\sigma\\:}_{xy}+{C}_{2}\\right)}{\\left({\\mu\\:}_{x}^{2}+{\\mu\\:}_{y}^{2}+{C}_{1}\\right)\\left({\\sigma\\:}_{x}^{2}+{\\sigma\\:}_{y}^{2}+{C}_{2}\\right)}\\)\u003c/span\u003e \u003c/span\u003e ・・・・・ (4)\u003c/p\u003e \u003cp\u003eHere, \u0026ldquo;MAX\u0026rdquo; represents the maximum value of the dose distribution (normalized to 1). \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\mu\\:}_{x}\\:\\text{a}\\text{n}\\text{d}\\:{\\mu\\:}_{y}\\:\\)\u003c/span\u003e\u003c/span\u003eare the mean values of the actual dose distribution image arrayམand the predicted dose distribution image array \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y\\)\u003c/span\u003e\u003c/span\u003e, respectively. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\sigma\\:}_{x}^{2}\\:\\text{a}\\text{n}\\text{d}\\:{\\sigma\\:}_{y}^{2}\\)\u003c/span\u003e\u003c/span\u003e are the variances of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:x\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y\\)\u003c/span\u003e\u003c/span\u003e, respectively, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\sigma\\:}_{xy}\\)\u003c/span\u003e\u003c/span\u003e is the covariance. C1 and C2 are stabilization constants.\u003c/p\u003e \u003cp\u003eBy combining these metrics, we provided a robust and transparent evaluation of the predicted dose distributions. MAE and MSE quantify the absolute and squared differences, respectively, offering straightforward error measures. PSNR emphasizes the relative magnitude of the errors, while SSIM captures perceptual similarity by considering the structural and statistical properties of the dose distribution.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003ePredicted DVH\u003c/h2\u003e \u003cp\u003eFigure 2 presents examples of DVHs predicted using IMPT and passive proton techniques for two patients. It compares the DVHs of the normal liver, duodenum, and bile duct predicted from dCT and MRI contour information with the actual calculated DVHs. The predicted DVHs visually closely matched the actual measurements. Tables\u0026nbsp;1 and 2 compare the predicted and calculated DVHs for all test patients. They describe the volume differences at 10% dose intervals for everyone, dose differences at D1cc [%], and normal liver V30 [%]. The tables summarize the mean values, standard deviations, and p-values for the predicted indicators. IMPT and Passive predicted from dCT are denoted as dCT_IMPT and dCT_Passive, respectively, while those predicted from MRI are referred to as MRI_IMPT and MRI_Passive.\u003c/p\u003e \u003cp\u003eFor dCT_IMPT for the normal liver, the maximum difference was 1.84\u0026thinsp;\u0026plusmn;\u0026thinsp;2.75% at the 100% dose, with an average difference of 2% and a standard deviation of 3%. For the duodenum, the maximum difference was 0.38\u0026thinsp;\u0026plusmn;\u0026thinsp;2.60% at the 10% dose, with an average difference within 1% and a standard deviation within 3%. For the bile duct, the maximum difference was 0.95\u0026thinsp;\u0026plusmn;\u0026thinsp;3.31% at a 30% dose, with an average difference within 1% and a standard deviation within 3% above 40%. The D1cc [%] ranged from \u0026minus;\u0026thinsp;1.28% to -4.51% (standard deviation, 4.29\u0026ndash;5.76%).\u003c/p\u003e \u003cp\u003eFor dCT_Passive, the maximum difference in the normal liver was 0.17\u0026thinsp;\u0026plusmn;\u0026thinsp;2.40% at the 40% dose, with an average difference within 1% and a standard deviation within 3%. For the duodenum, the maximum difference was 1.21\u0026thinsp;\u0026plusmn;\u0026thinsp;4.24% at the 10% dose, with an average difference within 2% and a standard deviation within 2% above 20%. For the bile duct, the maximum difference was 0.69\u0026thinsp;\u0026plusmn;\u0026thinsp;5.02% at the 20% dose, with an average difference within 1% and a standard deviation within 2% above 30%. The D1cc [%] ranged from 1.05% to -2.21% (standard deviation, 2.97\u0026ndash;9.84%).\u003c/p\u003e \u003cp\u003eFor MRI_IMPT, the normal liver showed a maximum difference of 0.38\u0026thinsp;\u0026plusmn;\u0026thinsp;2.25% at the 100% dose, with an average difference within 1% and a standard deviation within 3%. For the duodenum, the maximum difference was 0.31\u0026thinsp;\u0026plusmn;\u0026thinsp;1.15% at the 60% dose, with an average difference within 1% and a standard deviation within 2%. For the bile duct, the maximum difference was 1.04\u0026thinsp;\u0026plusmn;\u0026thinsp;4.89% at the 20% dose, with an average difference within 2% and a standard deviation within 3% above 70%. The D1cc [%] ranged from 0.12% to -4.58% (standard deviation, 2.26\u0026ndash;5.20%).\u003c/p\u003e \u003cp\u003eFor MRI_Passive, the normal liver exhibited a maximum difference of 0.16\u0026thinsp;\u0026plusmn;\u0026thinsp;2.47% at the 100% dose, with an average difference within 1% across all doses and a standard deviation within 3%. For the duodenum, the maximum difference was 1.38\u0026thinsp;\u0026plusmn;\u0026thinsp;4.39% at the 10% dose, with an average difference within 2% and a standard deviation greater than 2% above 20%. For the bile duct, the maximum difference was 1.13\u0026thinsp;\u0026plusmn;\u0026thinsp;4.36% at the 20% dose, with an average difference under 2% and a standard deviation above 30%. The D1cc [%] ranged from \u0026minus;\u0026thinsp;0.12% to -1.37% (standard deviation, 2.18\u0026ndash;2.90%).\u003c/p\u003e \u003cp\u003eThe DVHs predicted based on contours delineated on dCT and MRI showed no significant differences in the average DVH differences at 10% dose intervals or in the normal liver V30 [%] when compared to those derived from planning CT for both IMPT and passive proton therapy techniques. However, significant p-values (\u0026lt;\u0026thinsp;0.05) were observed in the D1cc [%] dose differences for some OARs.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003ePredicted dose\u003c/h2\u003e \u003cp\u003eThe comparison of dose distributions was performed using CDP, calculated in the axial, sagittal, and coronal planes, by comparing the predicted and calculated values. Figure\u0026nbsp;3 displays the three-dimensional dose distributions of IMPT and passive proton therapy predicted from contours on both dCT and MRI, along with the CDP in the axial plane. The fused images of the predicted dose distributions demonstrate that the targets are well covered across all combinations of dCT and MRI, as well as IMPT and passive proton therapy. In the axial CDP calculated from both the predicted and calculated dose distributions, signal values along the same lines on both the X- and Y-axes were observed to be generally consistent.\u003c/p\u003e \u003cp\u003eThe mean values and standard deviations of MAE, MSE, PSNR, and SSIM derived from the CDPs for all test patients are summarized in Table\u0026nbsp;3.\u003c/p\u003e \u003cp\u003eThe MAE remained below 0.03 across all results. In particular, for passive proton therapy, the MAE ranged from 0.012 to 0.018, demonstrating highly consistent reproducibility.\u003c/p\u003e \u003cp\u003eThe MSE was also generally below 0.004, indicating an error level that is clinically acceptable. For passive proton therapy, the MSE was below 0.003 across all views, confirming high accuracy. For IMPT, the highest MSE was observed in the coronal direction for both MRI and dCT. The standard deviation in the sagittal direction for MRI_IMPT was slightly higher, highlighting the need for caution due to variability between data sets.\u003c/p\u003e \u003cp\u003eThe PSNR ranged from 24 to 28 dB, indicating clinically acceptable reproducibility. Passive proton therapy showed high PSNR values across all directions, suggesting a high similarity to the original dose distributions. On the other hand, for IMPT, the dose distributions predicted from both dCT and MRI tended to show slightly lower PSNR values in the coronal direction, indicating a slight decline in reproducibility.\u003c/p\u003e \u003cp\u003eBased on the SSIM results, passive proton therapy demonstrated high structural similarity across all directions, with values exceeding 0.94, indicating excellent structural reproducibility. In contrast, IMPT showed a slightly lower trend overall, with the coronal direction suggesting slightly reduced structural reproducibility. However, SSIM values remained above 0.89 under all conditions, which can be considered clinically acceptable.\u003c/p\u003e \u003cp\u003eOverall, these results demonstrate that the proposed prediction method can achieve highly accurate and reproducible dose distributions for both IMPT and passive proton therapy. The findings particularly highlight the robustness of passive proton therapy predictions.\u003c/p\u003e \u003cp\u003eAdditionally, the entire process\u0026mdash;from automatic contouring to DVH prediction and dose distribution completion\u0026mdash;can be completed in approximately 30 minutes. This means that by delineating contours on existing images, the predicted dose distribution can be presented to the patient within 30 minutes, allowing for a timely assessment of treatment feasibility.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, a 1D CNN model in two steps was developed based on contours from two modality images, existing DVH, and dose distributions. The process involved predicting the DVH and dose distribution. We envisioned this model to assist oncologists during initial consultations, using a standard CPU-based PC for displaying medical records. While 2D or higher-dimensional CNNs may enhance prediction accuracy, they require longer computation times and expensive PCs equipped with GPUs [\u003ca class=\"FNLink\" href=\"#Fn18\" id=\"#FNLinkFn18\"\u003e\u003c/a\u003e]. This was the reason for adopting the 1D CNN model.\u003c/p\u003e \u003cp\u003eHowever, handling large-volume 3D images such as CT and MRI can potentially degrade PC performance, so it is essential to minimize the dataset size as much as possible [\u003ca class=\"FNLink\" href=\"#Fn19\" id=\"#FNLinkFn19\"\u003e\u003c/a\u003e]. In this study, we successfully predicted dose distributions without using the full 3D image data by relying solely on geometric values\u0026mdash;such as target size and the distances between the target and OARs\u0026mdash;extracted from contour information. Additionally, the RT_Dose, a 3D array used as the ground truth, was reduced to one-fourth of its original size in each dimension. While this downscaling posed a risk of reducing prediction accuracy, our results demonstrated that high accuracy was still achieved. Furthermore, in predicting dose distributions from contour data, it was found that predicting the DVH first and then using it to guide dose distribution prediction resulted in smaller errors for both the DVH and the final dose distribution compared to directly predicting the dose distribution and calculating the DVH afterward.\u003c/p\u003e \u003cp\u003eBased on Tables\u0026nbsp;2 and 3, comparisons between the ground truth and the DVHs of OARs predicted from contours delineated on dCT and MRI for both IMPT and passive methods showed no statistically significant differences across the dose range from 10\u0026ndash;100% of the prescription dose. However, for D1cc [%], the mean error ranged from \u0026minus;\u0026thinsp;4.5% to +\u0026thinsp;1.0% across all cases, with large variations in standard deviation, resulting in statistically significant differences. These discrepancies are thought to arise in small high-dose regions exceeding the prescription dose, which are known to be random in nature and difficult to predict accurately [\u003ca class=\"FNLink\" href=\"#Fn20\" id=\"#FNLinkFn20\"\u003e\u003c/a\u003e]. Nguyen et al. conducted a study using a U-Net-based deep learning model to predict dose distributions for prostate cancer IMRT plans from organ contour information. They reported that the mean absolute differences for Dmax and Dmean were less than 5% of the prescribed dose for both PTV and OARs. Our similar results suggest that our findings are comparable to those from GPU-based studies [\u003ca class=\"FNLink\" href=\"#Fn21\" id=\"#FNLinkFn21\"\u003e\u003c/a\u003e].\u003c/p\u003e \u003cp\u003eMoreover, an error of \u0026lt;\u0026thinsp;5% in D1cc is unlikely to pose a significant issue when making clinical decisions about whether to proceed with treatment. However, achieving closer agreement may be possible by incorporating information that strongly influences low-dose distributions\u0026mdash;such as beam angles from the accelerator side\u0026mdash;and by considering patient-specific hyperparameters, which could further improve prediction accuracy [\u003ca class=\"FNLink\" href=\"#Fn22\" id=\"#FNLinkFn22\"\u003e\u003c/a\u003e][\u003ca class=\"FNLink\" href=\"#Fn23\" id=\"#FNLinkFn23\"\u003e\u003c/a\u003e].\u003c/p\u003e \u003cp\u003eNext, we discuss the prediction accuracy of dose distributions using MAE, MSE, PSNR, and SSIM metrics across irradiation techniques (IMPT and passive proton therapy) and imaging modalities (MRI and dCT).\u003c/p\u003e \u003cp\u003eThe MAE values ranged from 0.012 to 0.030 across all directions, with the smallest error observed in the sagittal direction for dCT_Passive (0.012) and the largest in the coronal direction for MRI_IMPT (0.030). Considering that the acceptable range of dose measurement error in radiotherapy quality assurance is generally within \u0026plusmn;\u0026thinsp;3%, these values suggest favorable prediction accuracy [\u003ca class=\"FNLink\" href=\"#Fn24\" id=\"#FNLinkFn24\"\u003e\u003c/a\u003e].\u003c/p\u003e \u003cp\u003eFor MSE, values ranged from 0.002\u0026thinsp;\u0026plusmn;\u0026thinsp;0.001 to 0.004\u0026thinsp;\u0026plusmn;\u0026thinsp;0.002, with the best results observed for dCT_IMPT and dCT_Passive in the axial and sagittal directions. These ranges are considered highly favorable for similarity evaluations within the imaging field, indicating high prediction accuracy across all irradiation types. Huai-Wen Zhang et al. conducted dose distribution predictions for liver SBRT, reporting MSE values ranging from 0.0004 to 0.008 compared to the ground truth, which is nearly equivalent to the results obtained in this study [\u003ca class=\"FNLink\" href=\"#Fn25\" id=\"#FNLinkFn25\"\u003e\u003c/a\u003e].\u003c/p\u003e \u003cp\u003ePSNR is a quantitative metric used to evaluate image reproducibility, with a clinically acceptable accuracy generally considered to fall within the range of 25\u0026ndash;30 dB [\u003ca class=\"FNLink\" href=\"#Fn26\" id=\"#FNLinkFn26\"\u003e\u003c/a\u003e]. The PSNR results in this study ranged from 24 to 28 dB, confirming that the predicted dose distributions achieved a clinically sufficient level of reproducibility. Passive proton therapy exhibited the highest PSNR across all directions, indicating the most accurate prediction performance. In contrast, IMPT showed a slight decrease in PSNR in the coronal direction, which may be attributed to the characteristics of the spot-scanning technique. IMPT generates complex dose distributions through energy modulation, leading to steeper dose gradients and increased local variations. Consequently, minor reductions in reproducibility between the predicted and reference dose distributions may occur. Conversely, passive proton therapy employs fixed-port irradiation, resulting in fewer low-dose regions and a more uniform dose distribution. This characteristic likely contributes to consistently higher PSNR values.\u003c/p\u003e \u003cp\u003eSSIM values were also high, ranging from 0.898 to 0.959. Passive techniques achieved the highest SSIM scores (0.943\u0026ndash;0.959), indicating excellent structural similarity to the reference dose distribution. IMPT followed with values ranging from 0.898 to 0.940, which also indicates a high degree of similarity. While SSIM values between 0.85 and 0.90 are considered to reflect good agreement, some localized visual deviations may still be observed. Overall, the SSIM values exceeded 0.9 in most cases, indicating good structural similarity [\u003ca class=\"FNLink\" href=\"#Fn27\" id=\"#FNLinkFn27\"\u003e\u003c/a\u003e].\u003c/p\u003e \u003cp\u003eOur study demonstrates that both IMPT and passive techniques achieve high prediction accuracy for dose distributions based on diagnostic imaging, with passive techniques consistently exhibiting superior performance. Furthermore, MRI-based predictions showed accuracy comparable to those derived from dCT and the ground truth calculated from planning CT images, highlighting the feasibility of MRI-only workflows.\u003c/p\u003e \u003cp\u003eAdditionally, our CNN implementation is CPU-based and designed for deployment on widely available PCs, such as those used for electronic medical records. Given the limited number of studies on proton dose distribution predictions from existing dCT and MRI images, our system enables real-time dose distribution estimation during a patient's initial consultation, allowing for immediate treatment feasibility assessments. This is particularly beneficial for proton therapy, where treatment facilities are fewer than those for X-ray therapy, making it advantageous for patients who must travel long distances.\u003c/p\u003e \u003cp\u003eFurthermore, predicting dose distributions from existing diagnostic images can assist in determining whether planning CT should be performed for patients with large hepatocellular carcinoma, where the feasibility of proton therapy is uncertain due to trade-offs with OARs. This approach may also help reduce the workload of treatment planning staff.\u003c/p\u003e \u003cp\u003eWhile this study primarily focuses on generating predicted dose distributions for proton therapy, future research aims to store these distributions in DICOM format for integration into treatment planning systems. This advancement could enable the direct replacement of predicted dose distributions with actual irradiation machine outputs. Our previous research has demonstrated the feasibility of implementing this approach as a novel inverse dose calculation algorithm for Volumetric modulated arc therapy, and we aim to extend its application to proton therapy [\u003ca class=\"FNLink\" href=\"#Fn28\" id=\"#FNLinkFn28\"\u003e\u003c/a\u003e].\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eOur study evaluated the feasibility of proton therapy for HCC before acquiring planning CT by providing predicted dose distributions during the initial consultation. The proposed prediction model demonstrated high accuracy and reproducibility in both DVH and dose distribution predictions in proton therapy. The high SSIM and PSNR values, as well as the favorable MSE results, confirm the strong agreement between predicted and reference dose distributions. This approach has the potential to reduce unnecessary radiation exposure and support the selection of appropriate treatment strategies, particularly for patients with advanced tumors where the trade-off with OAR constraints is critical.\u003c/p\u003e \u003cp\u003eBy integrating irradiation angles and patient-specific anatomical features into the training process and optimizing the loss function, further improvements in prediction accuracy are expected. Additionally, our findings highlight the clinical utility of diagnostic imaging in radiation therapy, demonstrating its potential to streamline the workflow from initial consultation to treatment planning. This could reduce the time required for CT-based planning, enabling oncologists to assess treatment feasibility earlier in the process and improving efficiency for both patients and staff.\u003c/p\u003e \u003cp\u003eMoreover, implementing our CNN-based prediction model on widely available CPU-based systems, such as those used in electronic medical records, allows real-time dose estimation at the point of care. This could be particularly beneficial in proton therapy, where limited treatment facilities often require patients to travel long distances, potentially reducing the number of hospital visits.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eFinally, expanding this approach to other cancer types, leveraging advancements in diagnostic imaging, and incorporating emerging machine-learning techniques could further enhance its clinical utility. These prospects underscore the broad applicability of our research and provide a clear direction for advancing radiation therapy practices.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eHCC: Hepatocellular carcinoma, CNN: Convolutional neural networks, CT: Computed tomography, MRI: Magnetic resonance imaging, VMAT: Volumetric modulated arc therapy, IMPT: Intensity-modulated proton therapy, DVH: Dose-volume histogram, MAE: Mean absolute error, MSE: Mean squared error, PSNR: Peak signal-to-noise ratio, SSIM: Structural similarity index, DICOM: Digital imaging and communications in Medicine\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthical approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study is a retrospective analysis of radiotherapy outcomes using existing patient data. Ethical approval for the use of these data was granted by the Ethics Committee of the National Cancer Center Hospital, Japan (Approval No. 2020-272, dated 12 October 2020). All procedures were carried out in accordance with the Declaration of Helsinki and relevant institutional guidelines.\u0026quot;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWritten informed consent was obtained from the patient(s) for the use of personal data and images in this publication. The patient(s) were fully informed about the purpose of the study, the intended use of their data and images, and their rights to privacy. The consent form is retained by the corresponding author and is available upon request, should it be necessary for legal or ethical purposes. Additionally, this information can be referenced on the National Cancer Center Hospital East website: https://www.ncc.go.jp/jp/ncce/index.html.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eResearch data are stored in an institutional repository and will be made available upon request to the corresponding author.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interest\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;The authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by JSPS KAKENHI (Proposal No. 21K07741).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTR: Writing, Review \u0026amp; Editing, Software Development, Data curation Contouring, Formal analysis, and Visualization.\u003c/p\u003e\n\u003cp\u003eTT: Review \u0026amp; Editing, Patient Selection, Contouring, and Dose Distribution Creation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;No acknowledgments.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eTakeda A, Sanuki N, Tsurugi Y, Iwabuchi S, Matsunaga K, Ebinuma H, et al. Phase 2 study of stereotactic body radiotherapy and optional transarterial chemoembolization for solitary hepatocellular carcinoma not amenable to resection and radiofrequency ablation. Cancer. 2016;13:2041-49. Available from: https://doi.org/10.1002/cncr.30008.\u003c/li\u003e\n \u003cli\u003eYu-Lun T, Takei H, Lizumi T, Okumura T, Sekino Y, Numajiri H, et al. Capacity of proton beams in preserving normal liver tissue during proton beam therapy for hepatocellular carcinoma. J Radiat Res. 2020;62:133\u0026ndash;41. Available from: https://doi.org/ 10.1093/jrr/rraa098.\u003c/li\u003e\n \u003cli\u003eKorean Liver Cancer Study Group, National Cancer Center Korea. Practice guidelines for management of hepatocellular carcinoma 2009. Korean J Hepatol. 2009;15:391-423.\u003c/li\u003e\n \u003cli\u003eBenson AB III, D\u0026apos;Angelica MI, Abbott DE, Abrams RP, Alberts MJ, Saenz AS, et al. NCCN guidelines insights: hepatobiliary cancers. J Natl Compr Canc Netw. 2017;15:563-73.\u003c/li\u003e\n \u003cli\u003eKalogeridi MA, Zygogianni A, Kyrgias G, Kouvaris J, Chatziioannou S, Kelekis N, et al. Role of radiotherapy in the management of hepatocellular carcinoma: a systematic review. World J Hepatol. 2015;7:101-12. Available from: https://doi.org/10.4254/wjh.v7.i1.101.\u003c/li\u003e\n \u003cli\u003eNakayama H, Sugahara S, Tokita M, Fukuda K, Mizumoto M, Abei M, et al. Proton beam therapy for hepatocellular carcinoma: the University of Tsukuba experience. Cancer. 2009;115:5499-506. Available from: https://doi.org/10.1002/cncr.24619.\u003c/li\u003e\n \u003cli\u003eFukumitsu N, Sugahara S, Nakayama H, Fukuda K, Mizumoto M, Abei M, et al. A prospective study of hypofractionated proton beam therapy for patients with hepatocellular carcinoma. Int J Radiat Oncol Biol Phys. 2009;74:831-6. Available from: https://doi.org/10.1016/j.ijrobp.2008.10.073.\u003c/li\u003e\n \u003cli\u003eRonneberger O, Fischer P, Brox T. U-net: convolutional networks for biomedical image segmentation. In: Lecture Notes in Computer Science. Springer; 2015. p. 234-41. Available from: https://doi.org/10.1007/978-3-319-24574-4_28.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eHe K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2016. p. 770-8. Available from: https://doi.org/10.1109/CVPR.2016.90.\u003c/li\u003e\n \u003cli\u003eLiu Y, Chen Z, Wang J, Wang X, Qu B, Ma L, et al. Dose prediction using a three-dimensional convolutional neural network for nasopharyngeal carcinoma with tomotherapy. Front Oncol. 2021;11:752007. Available from: https://doi.org/10.3389/fonc.2021.752007.\u003c/li\u003e\n \u003cli\u003eKearney V, Chan JW, Haaf S, Descovich M, Solberg TD. DoseNet: a volumetric dose prediction algorithm using 3D fully-convolutional neural networks. Phys Med Biol. 2018;63:235022. Available from: https://doi.org/10.1088/1361-6560/aaef74.\u003c/li\u003e\n \u003cli\u003eMizumoto M, Okumura T, Hashimoto T, Fukuda K, Oshiro Y, Fukumitsu N, et al. Proton beam therapy for hepatocellular carcinoma: a comparison of three treatment protocols. Int J Radiat Oncol Biol Phys. 2011;81:1039-45. Available from: https://doi.org/10.1016/j.ijrobp.2010.07.015.\u003c/li\u003e\n \u003cli\u003eMizumoto M, Tokuuye K, Sugahara S, Nakayama H, Fukumitsu N, Ohara K, et al. Proton beam therapy for hepatocellular carcinoma adjacent to the porta hepatis. Int J Radiat Oncol Biol Phys. 2008;71:462-7. Available from: https://doi.org/10.1016/j.ijrobp.2007.09.056.\u003c/li\u003e\n \u003cli\u003eKawashima M, Furuse J, Nishio T, Konishi M, Ishii H, Kinoshita T, et al. Phase II study of radiotherapy employing proton beam for hepatocellular carcinoma. J Clin Oncol. 2005;23:1839-46. Available from: https://doi.org/10.1200/JCO.2005.00.620.\u003c/li\u003e\n \u003cli\u003eRobeson SM, Willmott CJ. Decomposition of the mean absolute error (MAE) into systematic and unsystematic components. PLOS ONE. 2023;18:e0279774. Available from: https://doi.org/10.1371/journal.pone.0279774.\u003c/li\u003e\n \u003cli\u003eTanabe Y, Ishida T. Quantification of the accuracy limits of image registration using peak signal-to-noise ratio. Radiol Phys Technol. 2017;10:91-4. Available from: https://doi.org/10.1007/s12194-016-0372-3.\u003c/li\u003e\n \u003cli\u003eMa C, Wang R, Zhou S, Wang M, Yue H, Zhang Y, et al. The structural similarity index for IMRT quality assurance: radiomics-based error classification. Med Phys. 2021;48:80-93. Available from: https://doi.org/10.1002/mp.14559.\u003c/li\u003e\n \u003cli\u003eSahiner B, Pezeshk A, Hadjiiski LM, Wang X, Drukker K, Cha KH, et al. Deep learning in medical imaging and radiation therapy. Med Phys. 2019;46:e1-e36. Available from: https://doi.org/10.1002/mp.13264.\u003c/li\u003e\n \u003cli\u003eLin Y, Liu Y, Chen H, Yang X, Ma K, Zheng Y, et al. LENAS: learning-based neural architecture search and ensemble for 3-D radiotherapy dose prediction. IEEE Trans Cybern. 2024;PP. Available from: https://doi.org/10.1109/TCYB.2024.3390769.\u003c/li\u003e\n \u003cli\u003eAhn SH, Kim E, Kim C, Cheon W, Kim M, Lee SB, et al. Deep learning method for prediction of patient-specific dose distribution in breast cancer. Radiat Oncol. 2021;16:154. Available from: https://doi.org/10.1186/s13014-021-01864-9.\u003c/li\u003e\n \u003cli\u003eNguyen D, Long T, Jia X, Lu W, Gu X, Iqbal Z et al. A feasibility study for predicting optimal radiation therapy dose distributions of prostate cancer patients from patient anatomy using deep learning. Sci. Rep. 2019;9:1\u0026ndash;10. Available from: https://doi.org/10.1038/s41598-018-37741-x.\u003c/li\u003e\n \u003cli\u003eBai X, Zhang J, Wang B, Wang S, Xiang Y, Hou Q. Sharp loss: a new loss function for radiotherapy dose prediction based on fully convolutional networks. Biomed Eng Online. 2021;20:101. Available from: https://doi.org/10.1186/s12938-021-00937-w.\u003c/li\u003e\n \u003cli\u003eShamsi A, Asgharnezhad H, Mohammadi P, Soofi R. An uncertainty-aware loss function for training neural networks with calibrated predictions. Mach Learn Comput Vis Pattern Recognit. Available from: https://doi.org/10.48550/arXiv.2110.03260.\u003c/li\u003e\n \u003cli\u003eFraass B, Doppke K, Hunt M, Kutcher G, Starkschall G, Stern R, et al. American Association of Physicists in Medicine Radiation Therapy Committee Task Group 53: quality assurance for clinical radiotherapy treatment planning. Med Phys. 1998;25:1773-829. Available from: https://doi.org/10.1118/1.598373.\u003c/li\u003e\n \u003cli\u003eZhang HW, Wang YH, Hu B, Pang HW. Uninvolved liver dose prediction in stereotactic body radiation therapy for liver cancer based on the neural network method. World J Gastrointest Oncol. 2024;16:4146\u0026ndash;56. Available from: https://doi.org/10.4251/wjgo.v16.i10.4146.\u003c/li\u003e\n \u003cli\u003eKobako M. Image compression guidelines for digital documents. JIIMA Standardization Committee, Vice Chair (JIS). Available from: https://www.jiima.or.jp/pdf/5_JIIMA_guideline.pdf.\u003c/li\u003e\n \u003cli\u003eWang Z, Bovik AC, Sheikh HR, Simoncelli EP. Image quality assessment: from error visibility to structural similarity. IEEE Trans Image Process. 2004;13:600-12. Available from: https://ieeexplore.ieee.org/document/1284395.\u003c/li\u003e\n \u003cli\u003eRachi T, Parshuram RV, Tanaka Y, Togo H. Examination of conversion method of dose distribution of lung cancer IMRT using fluence reversible calculation function in O-ring type linac and C-type linac. Phys Eng Sci Med. 2022;45:559-67. Available from: https://doi.org/10.1007/s13246-022-01122-6.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1.\u003c/strong\u003e\u003cstrong\u003e Predicted Dose Volume Histogram (DVH) Metrics for dCT Across Irradiation Techniques: Mean, Standard Deviation, and p-Values for ROIs\u003c/strong\u003e\u003c/p\u003e\n\u003ctable width=\"609\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003edCT\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eIMPT\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e10%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e20%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e30%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e40%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e50%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e60%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e70%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e80%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e90%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e100%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003eD1cc\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003eV30\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eLiver\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.17\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.07\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.23\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.32\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.49\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.60\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.66\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.78\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.24\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.84\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-4.54\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.58\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.65\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.33\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.89\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.05\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.13\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.13\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.27\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.78\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.75\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e4.67\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.71\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.79\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.90\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.65\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.50\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.26\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.22\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.17\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.08\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.06\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eDuodenum\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.38\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.32\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.05\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.37\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.16\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.23\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-1.64\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.60\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.95\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.74\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.80\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.79\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.35\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.68\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.16\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.95\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.97\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.54\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.99\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.28\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.79\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.40\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.30\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.33\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.33\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.33\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.03\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eBile duct\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.28\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.95\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.17\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.10\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.13\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.50\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.42\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-1.01\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.05\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e4.16\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.31\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.32\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.31\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.76\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.73\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.92\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.88\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.56\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e5.57\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.71\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.25\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.05\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.56\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.81\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.55\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.28\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.28\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.45\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003ePassive\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e10%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e20%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e30%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e40%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e50%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e60%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e70%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e80%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e90%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e100%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003eD1cc\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003eV30\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eLiver\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.17\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.25\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.27\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.17\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.21\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.20\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.25\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.29\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.08\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-2.21\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.05\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.72\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.96\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.01\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.40\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.62\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.39\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.95\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.13\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.68\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.97\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.56\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.68\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.60\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.58\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.78\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.61\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.92\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.40\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.32\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.30\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.85\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.01\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.01\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eDuodenum\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.21\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.14\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.25\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.49\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.66\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.55\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.51\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.24\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.10\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.01\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.05\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e4.24\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.66\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.12\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.41\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.87\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.79\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.68\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.70\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.39\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.02\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e9.84\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.26\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.74\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.37\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.17\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.16\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.23\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.23\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.29\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.33\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.48\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eBile duct\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.69\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.24\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.39\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.62\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.37\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.21\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.10\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.11\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.24\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e-1.98\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.16\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e5.02\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.23\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.17\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.84\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.38\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.22\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.94\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.02\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.65\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.29\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"60\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.96\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.58\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.44\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.47\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.29\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"42\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.48\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"50\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.68\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.66\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.15\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.03\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"46\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2. Predicted Dose Volume Histogram (DVH) Metrics for MRI Across Irradiation Techniques: Mean, Standard Deviation, and p-Values for ROIs\u003c/strong\u003e\u003c/p\u003e\n\u003ctable width=\"603\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eMRI\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eIMPT\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e10%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e20%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e30%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e40%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e50%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e60%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e70%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e80%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e90%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e100%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003eD1cc\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003eV30\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eLiver\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.33\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.68\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.60\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.40\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.35\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.26\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.26\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.38\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-4.58\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.57\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.75\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.91\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.78\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.39\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.61\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.86\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.76\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.70\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.68\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.25\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e5.20\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.09\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.47\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.93\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.16\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.12\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.35\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.48\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.57\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.69\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.56\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.52\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.31\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eDuodenum\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.50\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.45\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.19\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.03\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.31\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.26\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.11\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.39\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.67\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.54\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.36\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.97\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.71\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.10\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.15\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.96\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.16\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.41\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.47\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.26\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.79\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.20\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.11\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.29\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eBile duct\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.78\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.59\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.07\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.27\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-1.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.80\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.42\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.31\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.52\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.12\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.87\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e4.89\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.45\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.94\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.70\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.53\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.31\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.11\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.90\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.13\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.93\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.14\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.44\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.11\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.33\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.79\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.31\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.22\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.23\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.11\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.92\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003ePassive\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e10%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e20%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e30%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e40%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e50%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e60%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e70%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e80%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e90%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e100%\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003eD1cc\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003eV30\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eLiver\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.12\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.26\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.13\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.24\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.19\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.01\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.25\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.16\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-1.37\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.12\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.35\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.52\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.72\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.07\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.68\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.59\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.69\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.09\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.41\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.47\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.16\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.74\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.92\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.57\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.65\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.19\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.24\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.97\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.39\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.37\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.80\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.03\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.83\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eDuodenum\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.38\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.10\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.25\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.37\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.39\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.09\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.00\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.82\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e4.39\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.35\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.43\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.67\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.40\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.35\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.01\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.01\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.01\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.69\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.24\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.79\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.12\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.17\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.25\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.30\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.32\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.22\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eBile duct\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.52\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.13\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.04\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.33\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.53\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.50\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.38\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.12\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.06\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.18\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e-0.12\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e3.97\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e4.36\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.63\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.45\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.08\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.64\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.95\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.75\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.42\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.77\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.90\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"66\"\u003e\n\u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.62\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.33\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.92\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.61\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.25\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"41\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.14\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"49\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.54\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.62\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.38\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.87\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"45\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThe table presents the mean, standard deviation, and p-values of volume differences at each 10% dose interval for the true and predicted DVHs of test patients for dCT (Table 1) and MRI (Table 2). It also includes the dose differences for D1cc [%] and normal liver V30 [%].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3.\u003c/strong\u003e\u003cstrong\u003e Quantitative Evaluation Metrics for Predicted Dose Distribution Maps\u003c/strong\u003e\u003c/p\u003e\n\u003ctable width=\"491\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd width=\"51\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003eMRI_IMPT\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003edCT_IMPT\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003eMRI_Passive\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003edCT_Passive\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"6\" width=\"51\"\u003e\n\u003cp\u003e\u003cstrong\u003eMAE\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003eaxial\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.024\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.024\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.012\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.013\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.008\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.007\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.004\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.004\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003esagittal\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.018\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.016\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.014\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.012\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.007\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.007\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.006\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.006\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003ecoronal\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.030\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.028\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.017\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.018\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.009\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.008\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.006\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.008\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"6\" width=\"51\"\u003e\n\u003cp\u003e\u003cstrong\u003eMSE\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003eaxial\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.003\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.003\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.002\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.002\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003esagittal\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.003\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.002\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.003\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.002\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.002\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003ecoronal\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.004\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.004\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.003\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.003\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.002\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.002\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"6\" width=\"51\"\u003e\n\u003cp\u003e\u003cstrong\u003ePSNR\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003eaxial\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e25.798\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e25.744\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e27.265\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e26.972\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.055\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.086\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.003\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.925\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003esagittal\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e26.346\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e26.814\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e26.200\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e27.742\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.644\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.946\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.409\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.547\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003ecoronal\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e24.190\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e24.870\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e25.812\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e25.978\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.657\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.592\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e1.920\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e2.499\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"6\" width=\"51\"\u003e\n\u003cp\u003e\u003cstrong\u003eSSIM\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003eaxial\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.920\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.919\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.957\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.955\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.026\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.021\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.014\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.012\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003esagittal\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.937\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.940\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.954\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.959\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.020\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.018\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.016\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.017\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"53\"\u003e\n\u003cp\u003e\u003cstrong\u003ecoronal\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eaverage\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.898\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.902\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.945\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.943\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"69\"\u003e\n\u003cp\u003e\u003cstrong\u003eSD\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.025\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"77\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.021\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"82\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.020\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"81\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.024\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eCDPs in the axial, sagittal, and coronal directions were calculated from the dose distributions predicted for each irradiation type based on the dCT and MR images of all patients. To evaluate consistency from the CDPs, MAE, MSE, PSNR, and SSIM were calculated, and their mean values and standard deviations are shown.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"radiation-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"raon","sideBox":"Learn more about [Radiation Oncology](http://ro-journal.biomedcentral.com/)","snPcode":"13014","submissionUrl":"https://submission.nature.com/new-submission/13014/3","title":"Radiation Oncology","twitterHandle":"@OncoBioMed","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Hepatocellular carcinoma, proton therapy, convolutional neural network, dose distribution, diagnostic imaging, intensity-modulated proton therapy, passive methods","lastPublishedDoi":"10.21203/rs.3.rs-6397967/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6397967/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eProton therapy is commonly used for hepatocellular carcinoma (HCC). However, its feasibility can be challenging to assess large tumors or those adjacent to critical organs at risk (OARs), as these factors are typically evaluated only after treatment planning. This study aimed to predict proton dose distributions utilizing diagnostic computed tomography (dCT) and magnetic resonance (MR) images, leveraging a convolutional neural network (CNN), to enable early treatment feasibility assessments before planning CT (pCT) acquisition.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis study calculated dose distributions and dose-volume histograms (DVH) for 118 patients with HCC using intensity-modulated proton therapy (IMPT) and passive proton irradiation. The CNN model predicted the DVH and dose distributions, which were evaluated using mean absolute error (MAE), mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe predicted DVHs closely matched the actual DVHs. MAE was consistently below 0.03, with passive proton therapy achieving values between 0.012 and 0.018, indicating high consistency. MSE remained below 0.004 for all cases, confirming clinically acceptable accuracy. PSNR ranged from 24 to 28 dB. SSIM was above 0.94 for both IMPT and passive proton therapy, with the lowest value being 0.841, indicating high structural similarity.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThis study demonstrates the potential of diagnostic imaging in optimizing the workflow for HCC proton therapy planning. The proposed CNN-based model enables early dose distribution predictions, reducing the need for pCT acquisition and improving treatment decision-making.\u003c/p\u003e","manuscriptTitle":"Dose Distribution Prediction for Hepatocellular Carcinoma Using Convolutional Neural Networks from Diagnostic CT and MRI: Focus on Passive and Intensity-Modulated Proton Therapy","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-16 09:13:55","doi":"10.21203/rs.3.rs-6397967/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-05-03T18:49:12+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-03T17:49:01+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-03T01:02:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"204541632562899267711857057635009589227","date":"2025-04-16T10:33:01+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"145615795498609500365080620965147209703","date":"2025-04-15T14:15:36+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-09T04:28:42+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-04-08T13:32:55+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-08T13:26:52+00:00","index":"","fulltext":""},{"type":"submitted","content":"Radiation Oncology","date":"2025-04-08T01:03:38+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"radiation-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"raon","sideBox":"Learn more about [Radiation Oncology](http://ro-journal.biomedcentral.com/)","snPcode":"13014","submissionUrl":"https://submission.nature.com/new-submission/13014/3","title":"Radiation Oncology","twitterHandle":"@OncoBioMed","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"003744f2-3048-4b85-8198-729ef56bfab3","owner":[],"postedDate":"April 16th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-08-11T16:04:32+00:00","versionOfRecord":{"articleIdentity":"rs-6397967","link":"https://doi.org/10.1186/s13014-025-02708-6","journal":{"identity":"radiation-oncology","isVorOnly":false,"title":"Radiation Oncology"},"publishedOn":"2025-08-04 15:57:15","publishedOnDateReadable":"August 4th, 2025"},"versionCreatedAt":"2025-04-16 09:13:55","video":"","vorDoi":"10.1186/s13014-025-02708-6","vorDoiUrl":"https://doi.org/10.1186/s13014-025-02708-6","workflowStages":[]},"version":"v1","identity":"rs-6397967","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6397967","identity":"rs-6397967","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}