Investigate the quantification accuracy of small lesions in oncological 18F-FDG PET/CT using a deep progressive learning reconstruction method

Investigate the quantification accuracy of small lesions in oncological 18F-FDG PET/CT using a deep progressive learning reconstruction method | 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 Investigate the quantification accuracy of small lesions in oncological 18F-FDG PET/CT using a deep progressive learning reconstruction method Lei Xu, Rui Yang, Ru-shuai Li, Ren-cong Liu, Qing-le Meng, Feng Wang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6366594/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 17 Jan, 2026 Read the published version in BMC Medical Imaging → Version 1 posted 11 You are reading this latest preprint version Abstract Background To investigate the impact of deep progressive learning reconstruction (DPR) on small lesion detection and image quality compared to ordered subset expectation maximum (OSEM) and regularized OSEM (ROSEM) in 18 F-FDG PET/CT imaging. Methods The NEMA phantom was filled with 18 F-FDG solution, with a hot sphere-to-background ratio of 4:1. Twenty-six patients with 18 F-FDG-avid lung lesions (diameter < 2.0 cm) were enrolled in the study. The PET images were reconstructed by seven groups: routine OSEM, ROSEM with a penalization factor of 0.8 (ROSEM), and DPR reconstructions with five different filter strength factors ranging from smooth to sharp:1–5 (DPR1, DPR2, DPR3, DPR4, and DPR5). The contrast recovery (CR), background variability (BV), contrast-to-noise ratio (CNR), and radioactivity concentration ratio (RCR) were measured in the phantom study. The maximum standardized uptake values (SUV max ), target-to-background ratio (TBR), CNR, the volume of the lesions, and the coefficient of variation (COV) of the liver were calculated and compared between these methods in the patient study. Two radiologists evaluated the image quality using a five-point Likert scale. Results In the phantom study, the DPR2 to DPR5 and ROSEM groups achieved higher CR, CNR, RCR, and lower BV than the OSEM group. For the smallest 10-mm hot sphere, the DPR3 achieved a CNR of 24.46, while the ROSEM and OSEM groups achieved the corresponding values of 22.25 and 10.39, respectively. In the patient study, the DPR1 to DPR4 groups exhibited significantly lower liver COVs than the OSEM group (all p < 0.05). Nevertheless, no significant difference was observed between the DPR3 and ROSEM groups (p = 0.65). The lesion SUVs, TBRs, and CNRs of the DPR2 to DPR4 groups were found to be significantly higher than those of the OSEM group (all p < 0.05). However, the lesion SUVs and TBRs of the DPR4 and DPR5 groups were found to be equivalent to those of the ROSEM group (SUVs: p = 0.19–0.61, and TBRs: p = 0.26–0.70). Moreover, the CNRs of the DPR3 and ROSEM groups were found to be comparable (p = 0.75). The volume of the lesion obtained from the ROSEM group was statistically equivalent to that measured on CT images (p = 0.48), but those were overestimated by all DPR groups (all p < 0.05). The overall image quality scores for DPR2, DPR3, and DPR4 were found to be superior to those obtained with OSEM (p < 0.01), while those for DPR2 and DPR3 groups were not statistically different from the ROSEM group (p = 0.56, and p = 0.85). Conclusions The DPR method demonstrated a significant improvement in lesion contrast, TBR, and volumetric quantification accuracy for small lesions compared to the OSEM method in the 18 F-FDG PET/CT imaging. The DPR method, with a filter strength factor of 3, demonstrated comparable enhancement of PET image quality to the ROSEM method. PET/CT 18F-FDG Deep progressive learning Small lesions Image quality Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Background 18 F-FDG PET/CT is a highly effective functional imaging technology that has been widely adopted in clinical practice [ 1 – 3 ]. In order to achieve an accurate diagnosis, staging/restaging, and treatment monitoring, it is essential to have precise quantification of the radiotracer and high-quality images. However, image noise and contrast ratio can limit the image quality and quantitative accuracy. Clinical practice has demonstrated that the PET reconstruction algorithm has a substantial effect impact on both image quality and the accuracy of measuring the standardized uptake value (SUV) [ 4 , 5 ]. The most commonly used reconstruction algorithm at present is the ordered subset expectation maximization (OSEM) approach. However, the OSEM approach requires a considerable number of iterations to attain optimal quantitative accuracy in the presence of elevated image noise levels. Accordingly, the iterative process must be terminated prematurely. To address this issue, an additional post-processing technique, such as Gaussian filtering, has been employed to reduce the image noise. Nevertheless, this results in a certain degree of compromising with regard to image resolution and lesion detectability. To address this limitation, a regularized image reconstruction algorithm (ROSEM) that permits complete iterative convergence was developed, demonstrating enhanced performance compared to OSEM [ 6 ]. ROSEM incorporates the pixel-to-pixel total variation, global noise equivalent counts, and local sensitivity profile into a smooth penalty term in the iterative reconstruction process. This process serves to suppress background noise in images and enhance contrast between lesions, thereby obviating the necessity for post-processing filters. In recent years, convolutional neural networks (CNNs) have yielded state-of-the-art results in PET image denoising largely due to the rapid advancement of deep learning. Xing et al. [ 7 ] trained a deep learning-based denoising model using 90-second acquisition duration OSEM images with the objective of achieving target 180-second acquisition duration OSEM images. The model demonstrated the capacity to reduce the acquisition duration by 25% − 50% without compromising lesion contrast. In a related study, Mehranian et al. [ 8 ] developed a deep learning enhancement model using full-duration regularized OSEM images as the training target images. The model was subsequently demonstrated to be capable of reducing acquisition time and injected dose by 50%. Nevertheless, it is difficult to directly learn from input images to target images when there is a considerable disparity between them. Furthermore, the majority of studies utilize deep learning as a post-processing technique, rather than integrating the neural network model into the iterative process. Therefore, incorporating deep learning into the conventional iterative optimization model may prove an effective means of enhancing PET image quality. In a study inspired by a progressive learning strategy, Lv et al. [ 9 ] proposed a deep progressive learning (DPL) method for PET image reconstruction. This method serves to bridge the gap between low-quality images and high-quality images through the implementation of denoising and then enhancement steps. The results demonstrate the feasibility of this method for reducing noise and improving lesion contrast in PET images. In light of the superior performance of DPL, a revolutionary new deep learning iterative PET reconstruction technique, the Deep Progressive Learning Reconstruction (DPR) algorithm, is currently commercially available on the Chinese market (HYPER DPR, United Imaging Healthcare, China). The DPR algorithm effectively integrates the strengths of multiple CNNs with the exceptional performance of a comprehensive total-body PET/CT training dataset (uEXPLORER, United Imaging Healthcare, China). This integration results in the generation of high-contrast, low-noise PET images, while simultaneously minimizing the associated side effects such as low count statistics, decay effect and partial-volume effect. Preliminary studies have demonstrated that the DPR method can reduce the administered activity of 18 F-FDG by up to two-thirds in a clinical setting while maintaining image quality and allowing for more accurate quantification of lesions in overweight or obese patients [ 10 , 11 ]. Nevertheless, the performance of DPR on small lesions remain to be fully elucidated. The aim of this study was to examine the accuracy of the depiction and quantification of 18 F-FDG PET/CT images reconstructed using the DPR algorithm for small lung lesions with a diameter of less than 2.0 cm. A compared quantitative and qualitative comparison of image quality metrics was conducted between DPR reconstruction with different filter strengths and OSEM and ROSEM, using data from a NEMA phantom and patient data. The image quality and the accuracy of quantification of small lesions were evaluated in order to determine the optimal filter strength factor for DPR reconstruction. Methods DPR algorithm DPR is an iterative image reconstruction technique that integrates a denoising network (CNN-DE) and an enhancement network (CNN-EH) into the iterative part of the OSEM algorithm, thereby producing low-noise and high-contrast 18 F-FDG PET images. The CNN-DE was previously trained to differentiate between noise and signal without influencing the metabolic distribution. The CNN-EH was subjected to a pre-training process with the objective of enhancing image contrast. Furthermore, the PET scan reconstruction also integrates the time-of-flight (TOF) and point-spread function (PSF) methods, which can be combined with conventional image filters. The deep learning CNNs were trained using data from uEXPLORER, a total body PET scanner, in which the PET images provided high contrast and minimal noise [ 12 ]. The dimensions of the training image were 249 × 249 × 671 with a voxel size of 2.4 × 2.4 × 2.68 mm 3 and 499 × 499 × 1342 with a voxel size of 1.2 × 1.2 × 1.34 mm 3 for the 2.4 mm and 1.2 mm networks, respectively. In order to train the CNN-DE, target images are utilized, which are 15-minute scanning low noise uEXPLORER images. The input images are introduced at a rate of 10% of the total counts. To train the CNN-EH, images with two OSEM iterations and a higher number of OSEM iterations were employed as the target images and input images, respectively. The CNN-DE and CNN-EH are both designed based on the feedback network [ 13 ]. The training dataset was constructed using a total of 161,040 image slice pairs from 80 patients. This comprised 53,680 and 107,360 slice pairs for the 2.4 mm and 1.2 mm networks, respectively. The test dataset was constructed using a total of 40,260 image slice pairs from 20 patients, including 13,420 and 26,840 slice pairs for the 2.4 mm and 1.2 mm models, respectively. Further details on the design of the DPR algorithm design, as well as the network training and testing procedures, can be found in the reference literature [ 9 ]. Phantom study In this study, a National Electrical Manufacturers Association (NEMA) International Electrotechnical Commission (IEC) body phantom was utilized, comprising six fillable spheres with inner diameters of 10 mm, 13 mm, 17 mm, 22 mm, 28 mm, and 37 mm, respectively. Furthemore, a cylinder lung insert with a diameter of 5 cm and a length of 16 cm was positioned at the center of the phantom. The phantom background was filled with an 18 F-FDG solution at an activity concentration of 8.26 kBq/ml, while the four smallest spheres were filled with an 18 F-FDG solution at 33.04 kBq/ml. This yielded a sphere-to-background ratio of 4:1. Patients study The study population consisted of 30 consecutive patients who were referred to the Nanjing First Hospital between July 20 and December 20, 2022, and underwent 18 F-FDG PET/CT examinations for cancer staging or restaging. The study’s inclusion criteria were as follows: the presence of 18 F-FDG-avid lesions was identified in the lungs, the target lesion was able to be segmented on CT images with a diameter of less than 2.0 cm, and the availability of list-mode raw data for additional PET reconstructions. The exclusion criteria were as follows: the 18 F-FDG uptake time was found to be longer than 100 minutes or less than 40 minutes in three patients. Furthermore, it was observed that respiratory motion had a considerable impact on the accuracy of lesion segmentation on PET images in a single patient. In conclusion, the study cohort comprised 26 patients, with a female-to-male ratio of 11:15, and an age range of 33 to 78 years. This retrospective study was approved by the institutional review board of Nanjing First Hospital and did not require to written informed consent. PET/CT procedure for the phantom and patient study The PET/CT scans were conducted using a digital PET/CT scanner (uMI780, United Imaging Healthcare, Shanghai, China), with a 30 cm axial length and a system sensitivity of 15 kcps/MBq. Prior to the administration of 18 F-FDG, patients were required to fast for a minimum of six hours and their blood glucose levels were confirmed to be less than 10 mmol/mL. A weight-based dose of 5.0 MBq/kg 18 F-FDG was administered to the patients as an intravenous bolus. Prior to the commencement of the scanning procedure, patients were instructed to consume a volume of water between 0.5 and 1.0 liters. The patients and phantom were subjected to CT scanning with a fixed tube voltage of 120 kV and an automated milliampere-second (mAs) technique for dose modulation. This provided the necessary anatomical information and attenuation correction for the PET images. Subsequently, a whole-body PET scan was acquired in 3D list mode for a period of two minutes per bed position. This provided coverage of the area from the skull base to the mid-thigh of the patient, or approximately 30 cm to encompass the entire phantom. PET image reconstruction The PET images were reconstructed into seven groups, comprising the routine OSEM, regularized OSEM, and DPR algorithms, with five filter strengths ranging from 1 to 5, indicating smooth to sharp. In the following sections, the aforementioned groups will henceforth be referred to as OSEM, ROSEM, DPR1, DPR2, DPR3, DPR4, and DPR5, respectively. The OSEM group was implemented using two iterations and 20 subsets. Additionally, a Gaussian filter with a full width at half maximum of 3 mm, a 256 × 256 matrix, and a 600 mm field of view (FOV) was employed. Additionally, a slice thickness of 3 mm, TOF, and a PSF model were employed. The necessary corrections, including those for scatter, random, dead time, decay, attenuation, and normalization, were incorporated into the reconstruction. The ROSEM group employed HYPER Iterative, a commercially available implementation of Bayesian penalized likelihood (BPL) algorithms that incorporates a total variation regulator and the sensitivity profile of PET scanners into the penalization term. The penalty factor is a hyperparameter that regulates the image contrast and smoothness. The operator can adjust this value between 0 and 1. In this study, a penalty factor of 0.8 was selected based on the findings of a preliminary study [ 14 ], which demonstrated that this factor yielded optimal accuracy in depicting and quantifying small lesions in oncological 18 F-FDG PET/CT. Furthermore, the DPR algorithm was employed for PET reconstruction without the application of any additional post-processing methods. The FOV, matrix, and slice thickness utilized in the ROSEM and DPR groups were identical to those employed in the OSEM group. The CT images were reconstructed with a FOV of 600 mm, a matrix of 512 × 512, and a slice thickness of 3 mm, with 1.5 mm increments. Quantitative evaluation of the phantom and patient images The quantitative analyses were conducted by a senior nuclear radiologist on a dedicated workstation (uWI-MI, United Imaging Healthcare, China). For the IEC body phantom study, the quality of the PET images was evaluated in accordance with the NEMA NU-2012 protocol. The region of interest (ROI) was delineated at the center of each hot sphere with a diameter matching that of the sphere. The background ROI was delineated in the peripheral area of the phantom background at the central slice of the spheres, as well as ± 1cm and ± 2cm from the central axis. A total of 60 background ROIs of varying size, with 12 ROIs on each of the five slices, are to be drawn. It is imperative that the locations of all ROIs remain fixed between successive measurements, and that the mean counts in each background ROI are duly recorded. The percent contrast recovery (CR) and background variation (BV) were calculated using the appropriate equations, namely (1) and (2). The contrast-to-noise ratio (CNR), a measure of the signal level in the presence of noise, was calculated using equations (3). The residual lung error (LE) was calculated using Eq. (4). The radioactivity concentration ratio (RCR) was determined by dividing the measured activity by the injected activity of the hot sphere. Furthermore, the normalized activity of four hot spheres was calculated as the mean activity concentration of all reconstruction groups in relation to that of the OSEM group. This demonstrates the comparative change resulting from the various reconstructions, with the OSEM serving as the reference. In the context of the patient study, a volume of interest (VOI) with a diameter of 3 cm was delineated manually at the same position on a homogeneous area of the right liver lobe for each image. The mean and standard deviation (SUV mean and SUV sd ) were recorded, respectively. The coefficient of variation (COV) in the liver, which serves as a measure of background noise, was calculated by dividing the SUV sd by the SUV mean . A semi-automatic segmentation tool (MI-Oncology, United Imaging Healthcare, Shanghai, China) was utilized to delineate each small 18 F-FDG-avid lung lesion on PET and CT images. The volume and maximum of standardized uptake value (SUV max ) of the lesion were quantified on the PET image using a 41% SUV max threshold. Only lesions with a diameter of less than 2.0 cm, as determined on the corresponding CT image, were included in the analysis. Furthermore, the lesion was segmented using CT images, and the resulting volume was employed as a standard reference for comparison with PET- and CT-derived volumes. To evaluate the image contrast, the tumor-to-background ratio (TBR) was calculated by dividing the lesion’s SUV max by the liver’s SUV mean . The contrast-to-noise ratio (CNR) of the lesion was calculated by dividing the lesion SUV max by the liver’s SUV sd . $$\:{CR}_{H,j}=\frac{{C}_{H,j}/{C}_{B,j}-1}{A-1}\times\:100\%\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(1\right)$$ $$\:{BV}_{j}=\frac{\sqrt{\sum\:_{k=1}^{K}{\left({C}_{B,j,k}-{C}_{B,j}\right)}^{2}/\left(K-1\right)}}{{C}_{B,j}}\times\:100\%\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(2\right)$$ $$\:{CNR}_{H,j}=\frac{{CR}_{H,j}}{{BV}_{j}}\times\:100\%\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(3\right)$$ $$\:{LE}_{i}=\frac{{C}_{lung,i}}{{C}_{B,i}}\times\:100\%\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(4\right)$$ The notation C H,j represents the average counts within a ROI for hot sphere j . The symbol BV j signifies the background variability for sphere j , and A is the standard hot sphere to background concentration ratio. The symbol C lung,i denotes the average counts for an ROI drawn in the lung insert on slice i , while C lung,i represents the average background counts for slice i . Finally, the symbol K = 60 signifies the number of selected background ROIs. Visual image quality assessment for the patient images The quality of the PET image was evaluated by two nuclear radiologists, each with over 10 years of experience, on a dedicated workstation (uWS-MI R004, United Imaging Healthcare, China). To mitigate potential bias, the order in which the seven series of PET images were evaluated was randomized for each patient. Furthermore, the readers were blinded to the patient’s clinical information and reconstruction settings. A five-point Likert scale was employed to evaluate three perspectives: image noise, the lesion depiction, and the overall image quality. The rating scale employed was as follows: 1 = The image quality is poor, with an excessive amount of noise or unfavorable lesion contrast, and insufficient lesion depiction. 2 = The image quality is deemed unacceptable, with suboptimal noise and blurring of lesions, which impairs diagnostic confidence. 3 = Acceptable image quality, comparable to that of routine oncological 18 F-FDG PET/CT images, with an appropriate level of noise and average lesion delineation that enables a clinical diagnosis. 4 = The image quality is optimal, with the desired level of noise and the capacity to detect small lesions with precision, thereby fostering robust diagnostic confidence. A score of 5 indicates an excellent image quality, with optimal noise, clear contrast of lesions, and accurate depiction and quantification of small lesions, thereby instilling complete confidence in the diagnostic process. The scores for image quality, which were independently assessed by the two raters, were recorded for the purposes of evaluating inter-reader agreement test. In the event of a discrepancy between the two radiologists, a consensus meeting was convened with a third physician proficient in nuclear medicine to reach a final determination. Statistical analysis The mean ± standard deviation is employed to present the data. A two-tailed paired t-test was employed to assess quantitative data that exhibited a normal distribution, as determined by the Shapiro-Wilk normality test. The p -value was adjusted using the Benjamini & Hochberg correction to account for the false discovery rate resulting from multiple comparisons. A simple linear regression was employed to quantify the degree of agreement between liver COV and lesion SUV max between OSEM and the other reconstruction groups. To assess the inter-reader agreement, a Cohen’s kappa test was employed. The third reader provided qualitative assessments of image quality scores, which were subsequently evaluated using a matched-pairs Wilcoxon signed-rank test. A p-value of less than 0.05 was considered to indicate statistical significance. All data were processed using GraphPad Prism 9.0 and Microsoft Excel 2022 software. Results Phantom study The CR, BV, CNR, LE, RCR, and normalized activity of the NEMA phantom with various reconstruction methods are presented in Fig. 1 . For all reconstruction groups, the trend of CRs exhibited an initial increase, followed by a decrease with the increase in the diameter of the hot sphere. The highest values for the CRs were observed for DPR4 and DPR5, followed by DPR2, DPR3, and ROSEM, and reaching their lowest values for DPR1 and OSEM across all spheres. With the exception of the 22 mm hot sphere, the CRs exhibited by DPR1 were equivalent to those exhibited by OSEM for each hot sphere. Moreover, the CRs exhibited a near equivalence between DPR4 and DPR5, with the discrepancies confined to a narrow range (-0.4–2.1%). As the diameter of the hot sphere increased, the discrepancy in CRs responsible for DPR reconstruction groups diminished. The CRs of the DPR2 and DPR3 were observed to be higher than those of ROSEM at the diameters of the 17-mm and 22-mm hot spheres. It is noteworthy that ROSEM achieved a CR of 71.2% for the 10 mm hot sphere, which was higher than the corresponding values for DPR2 and DPR3 of 64.4% and 68.5%, respectively. Upon increasing the diameter of the sphere to 13 mm, the CRs exhibited a near-equivalence between DPR3 (75.1%) and ROSEM (75.0%) (Fig. 1 a). The BVs of each reconstruction group exhibited a declining trend as the diameter of the hot sphere increased, with the gap progressively narrowing to a narrow range. Additionally, the lowest BVs were observed in DPR2, followed by DPR3, DPR1, ROSEM, DPR4, and DPR5, while the highest BVs were observed in OSEM at the same diameter of the hot sphere. Furthermore, the BVs of DPR1 were found to be comparable to those of ROSEM. Notably, the BVs for DPR2 to DPR3 were observed to be 2.2–2.8% for the smallest 10 mm hot sphere. The corresponding values for OSEM and ROSEM were 5.1% and 3.2%, respectively (Fig. 1 b). The CNRs of each reconstruction group demonstrated an increase as the diameter of the spheres increased. The CNRs of each sphere exhibited the highest values for DPR2, followed closely by DPR3, and the lowest values for OSEM. In particular, the CNR of DPR2 was 29.27 for the smallest sphere with a diameter of 10 mm. Moreover, the CNRs of DPR1 were observed to be comparable to those of ROSEM, with the exception of the 10 mm hot sphere (Fig. 1 c). In all reconstruction groups, the highest residual lung error was observed for DPR1 (7.6%), followed by OSEM (7.0%) and ROSEM (5.6%). The discrepancies between DPR2 and DPR5 were minimal, spanning a narrow range of 4.3–3.6% (Fig. 1 d). All reconstruction groups exhibited a mean normalized activity value exceeding 1.0, demonstrating a slight increase with the enhancement filter strength. The mean normalized activity of DPR1 was found to be highly comparable to that of OSEM, while that of ROSEM was found to be similar to that of DPR2 and DPR3 (Fig. 1 e). A positive correlation was observed between the RCRs and the diameter of the hot sphere for all reconstruction groups. Furthermore, an increase in the strength of the filter factor was accompanied by a corresponding raise in the RCRs of each sphere. The DPR4 and DPR5 groups exhibited the highest RCRs, while the OSEM and DPR1 groups demonstrated the lowest RCRs. The RCRs of DPR1 were found to be equivalent to those of OSEM, with the exception of the 22 mm hot sphere. Furthermore, the RCRs of the ROSEM group were observed to be higher than those of the DPR2 and DPR3 groups in the 10 mm and 13 mm hot spheres, while being lower than those of the DPR2 and DPR3 groups in the 17 mm and 22 mm hot spheres (Fig. 1 f). Patient characteristics The study cohort comprised 26 lung lesions. The diameter of these lesions was 0.94 ± 0.18 cm (range 0.7–1.4 cm), which corresponds to a volume of 0.26 ± 0.14 cm 3 (range 0.10–0.74 cm 3 ) as determined by CT imaging. The mean body weight of the patients was 65.7 ± 11.6 kg (range 41–93 kg), while the mean height was 1.65 ± 0.07 m (range 1.52–1.78 m). The administered activity of 18 F-FDG was 327.12 ± 48.4 MBq (range 218.3–411.14 MBq), with an average uptake time of 62.81 ± 14.82 minutes (range 46–97 minutes). The primary cancer type and other clinical characteristics are presented in Table 1 for reference. Table 1 Patient clinical characteristics Parameter Value Number of patients 26 Age (years) 58.3 ± 14.8 [33, 78] Gender Female 11 Male 15 Height (m) 1.65 ± 0.07 [1.52, 1.78] Weight (kg) 65.7 ± 11.6 [41, 93] Body mass index (kg/m 2 ) 24.14 ± 3.56 [17.06, 32.18] Injected activity (MBq) 327.12 ± 48.4 [218.3, 411.14] Injected activity per weight (MBq/kg) 5.04 ± 0.66 [3.95, 6.50] Uptake time (minutes) 62.81 ± 14.82 [46, 97] Primary cancer type Lung cancer 10 Paraganglioma 1 Lymphoma 3 Rectal cancer 3 Pancreatic cancer 1 Thyroid cancer 3 Neuroendocrine tumor 1 Vestibular adenocarcinoma 1 Breast cancer 1 Gastric cancer 1 Liver cancer 1 Quantitative image evaluation for the patient study The mean SUV mean for the liver was approximately 2.55 across all reconstruction groups (Table 2 and Fig. 2 a). The liver SUV sd demonstrated an increase with the enhancement filter strength in the DPR groups, with the exception of DPR1. The lowest liver SUV sd were observed in the DPR2 reconstruction group. The mean liver SUV sd of DPR5 was slightly higher than that of the OSEM group, while the mean liver SUV sd of DPR3 were comparable to those of the ROSEM group (Fig. 2 b). The liver COVs for all reconstruction groups were less than 15%, with an average of less than 11% (Table 2 and Fig. 2 c). The DPR1 to DPR4 groups exhibited significantly lower liver COVs than the OSEM group (all p < 0.05), whereas no significant difference was observed between the DPR5 and OSEM groups (p = 0.48) (Fig. 2 d). The liver COVs for the ROSEM group were observed to be higher than those for the DPR1 group (p = 0.09), yet lower than those for the DPR3 group (p = 0.65). The liver COVs for DPR2 were found to be significantly lower than those for ROSEM (p < 0.01), while the COVs for DPR4 and DPR5 were found to be significantly higher than those for ROSEM (all p < 0.01). Moreover, the mean lesion SUV max exhibited a positive correlation with increasing filter strength factor (Table 2 , Fig. 3 a, and Fig. 3 e). The lesion SUV max of the ROSEM and all DPR groups was significantly higher than that of the OSEM group, with the exception of the DPR1 group (all p < 0.05 and p = 0.40, respectively). The lesion SUV max of the ROSEM group was found to be significantly higher than that of the DPR1, DPR2, and DPR3 groups, respectively (all p < 0.01). Meanwhile, the DPR4 and DPR5 groups demonstrated results comparable to the ROSEM group (p = 0.61and p = 0.18, respectively). The mean normalized lesion SUV max increased by 53–76% for the DPR2 to DPR5 groups, and by 71% for the ROSEM groups in comparison to the OSEM group (Fig. 3 e). Table 2 Quantitative results of the patient study (mean ± SD) Methods Liver SUV mean Liver COV (%) Lesion SUV max Lesion TBR Lesion CNR Lesion volume (cm 3 ) OSEM 2.55 ± 0.41 9.80 ± 1.53 4.02 ± 0.81 1.63 ± 0.47 16.68 ± 4.13 0.64 ± 0.29 ROSEM 2.55 ± 0.40 5.71 ± 1.07 6.82 ± 1.36 2.73 ± 0.70 49.38 ± 15.18 0.31 ± 0.16 DPR1 2.55 ± 0.41 4.91 ± 1.35 3.88 ± 0.91 1.57 ± 0.54 32.98 ± 10.88 0.65 ± 0.33 DPR2 2.55 ± 0.40 4.21 ± 0.94 6.11 ± 1.16 2.46 ± 0.68 59.92 ± 15.29 0.37 ± 0.18 DPR3 2.55 ± 0.40 5.84 ± 1.23 6.38 ± 1.16 2.57 ± 0.68 45.09 ± 11.29 0.35 ± 0.16 DPR4 2.55 ± 0.40 8.25 ± 1.43 6.73 ± 1.16 2.70 ± 0.68 33.28 ± 7.89 0.33 ± 0.15 DPR5 2.55 ± 0.41 10.3 ± 1.65 6.96 ± 1.17 2.79 ± 0.67 27.46 ± 6.34 0.34 ± 0.19 CT 0.26 ± 0.14 *Data are presented as the mean ± standard deviation The DPR5 group exhibited the highest mean lesion TBR of all the reconstruction groups. As the filter strength increased, the lesion TBR also increased (Table 2 and Fig. 3 b). The TBRs of the DPR2 to DPR5 groups were found to be significantly higher than those of the OSEM group (all p < 0.05), whereas the DPR1 and OSEM exhibited comparable TBRs (p = 0.37). The lesion TBR of the DPR1, DPR2, and DPR3 groups was significantly lower than that of the ROSEM group (all p < 0.01), while the lesion TBR of the DPR4 and DPR5 groups was comparable to that of the ROSEM group (p = 0.70 and p = 0.26, respectively). The mean normalized lesion TBR increased by 1.53 to 1.75 times for the DPR2 to DPR5 groups, and by 1.71 times for the ROSEM group in comparison to the OSEM group (Fig. 3 f). The mean lesion CNR of all the DPR groups and the ROSEM group were found to be significantly higher than that of the OSEM group (all p < 0.01). The DPR2 group exhibited the highest CNRs among all reconstruction groups (Table 2 , and Fig. 3 c). The CNR of the ROSEM group was found to be significantly lower than that of the DPR2 group (p < 0.01), but significantly higher than that of the DPR1, DPR4, and DPR5 groups (all p < 0.01). The mean CNR of the DPR3 groups was observed to be smaller than that of the ROSEM group, although the difference was not statistically significant (p = 0.75). The mean normalized lesion CNR exhibited a 1.67-to-3.63-fold increase for the DPR groups and a 2.98-fold increase for the ROSEM group in comparison to the OSEM group (Fig. 3 g). The volume of the lesions was found to be significantly higher in all the DPR groups compared to that measured on CT images (all p < 0.05). Nevertheless, the lesion volume derived from the ROSEM group was statistically equivalent to that measured on CT images (p = 0.48) (Table 2 , and Fig. 3 d). The lesion volume of the OSEM group was found to be significantly larger than that of the ROSEM group and all DPR groups, with the exception of DPR1 (all p 0.99, respectively). The mean lesion volume was observed to be 2.66 times that of the OSEM and DPR1 groups, and 1.49 to 1.33 times that of the DPR2 to DPR5 groups, and 1.22 times that of the ROSEM group in comparison to the CT volumetric measurement (Fig. 3 h). Therefore, the volume derived from the ROSEM group was found to be more accurate than those derived from the OSEM group and all the DPR groups when CT volumetric measurement was employed as the standard reference. In a subsequent subgroup analysis, the small lesions were divided into two categories based on their diameters: sub-centimeter (D < 1.0 cm, n = 16, range 0.7–0.9 cm) and medium-size (1.0 ≤ D < 2.0 cm, n = 10, range 1.0 -1.8 cm). In each category, the lesion SUV max of the DPR groups exhibited an increasing trend with the enhancement of filter strength. The mean normalized SUV max for sub-centimeter lesions was observed to be higher than that for medium-size lesions across the ROSEM and all the DPR groups, with the exception of the DPR1 group at the same filter strength ( Fig. 4 a and Fig. 4 e). The mean normalized SUV max was greater than 1.6 for sub-centimeter lesions (range 1.61–1.85) and 1.4 for medium-sized lesions (range 1.41–1.61) across all the DPR groups, with the exception of the DPR1 group. Nevertheless, the average normalized SUV max exhibited only minor differences between sub-centimeter and medium-sized lesions in the DPR1 group. The average normalized lesion SUV max of the ROSEM group demonstrated a significant improvement, reaching 1.85 times that of sub-centimeter lesions and 1.51 times that of medium-sized lesions, respectively. A comparable pattern was observed in the metric of lesion TBR (Fig. 4 b and Fig. 4 f). The impact of lesion diameter on the lesion CNR for different reconstruction techniques is illustrated in Fig. 4 c and Fig. 4 g. The mean lesion CNR of the DPR groups initially increases with filter strength, reaching a maximum value, and then subsequently declines. This trend reaches a maximum at the DPR2 groups for both the sub-centimeter and medium-size categories. The mean normalized lesion CNR of the sub-centimeter group was observed to be greater than that of the medium-size group for all reconstruction procedures, with the exception of the DPR1 groups. The DPR2 group exhibited the most substantial increase in both groups, with a 3.81-fold rise in the sub-centimeter cohort and a 3.34-fold surge in the medium-size group. When the reconstruction settings were identical, the mean of the normalized lesion volume for the sub-centimeter group was observed to be greater than that of the medium-size group (Fig. 4 d and Fig. 4 h). The mean of the normalized lesion volume for the OSEM group exhibited the most pronounced fluctuations between the sub-centimeter and medium-size categories, with a range of 2.96 to 2.18. In contrast, the mean of the normalized lesion volume for the ROSEM group was approximately 1.2, with a value of 1.25 for the sub-centimeter group and 1.17 for the medium-size group. Moreover, minor discrepancies were identified between the sub-centimeter and medium-size categories within the DPR5 group. A linear regression of liver COV and lesion SUV max revealed a significant degree of agreement between the OSEM method and other reconstruction methods (Fig. 5 ) . The liver COV concordance correlation coefficients for the DPR groups (R = 0.91–0.94) were all greater than 0.91, with the exception of the DPR1 group (R = 0.60). The liver COV concordance correlation coefficients were found to be comparable between the DPR1 and ROSEM groups (R = 0.60 and R = 0.61, respectively). As the filter strength increased, the concordance correlation coefficients of lesion SUV max between the OSEM and DPR groups demonstrated a decline (R = 0.91 − 0.75). The lowest concordance correlation coefficient was observed between the OSEM and ROSEM groups (R = 0.71). Visual image scores for the patient study The inter-reader agreement for visual noise, lesion depiction, and overall image quality was substantial agreement between the readers, with kappa values of 0.695, 0.697, and 0.771, respectively. The results for visual noise, lesion depiction, and overall image quality scores are presented in Fig. 6 . As the filter strength increased, the visual noise score initially increased and subsequently declined (Fig. 6 a). The visual noise score achieved by the DPR2 group was the highest (4.54 ± 0.50), while that achieved by the OSEM group and the DPR5 group was the lowest (2.88 ± 0.32 and 2.88 ± 0.58, respectively). The visual noise scores for the DPR1, DPR2, and DPR3 groups were found to be superior to those of OSEM (all p < 0.001). In contrast, the DPR4 and DPR5 groups exhibited higher visual scores, although the difference was not statistically significant when compared to the OSEM group (p = 0.149 and p = 0.923, respectively). In comparison to the ROSEM group, the DPR2, DPR4, and DPR5 groups exhibited elevated visual noise scores relative to the ROSEM group (p = 0.021, p = 0.0008, and p < 0.0001, respectively), whereas the DRP1 and DPR3 groups exhibited the same visual noise scores as the ROSEM groups (p = 0.748 and p = 0.608, respectively). The lesion depiction score was observed to improve with an elevate filter strength (4.35–4.75) (Fig. 6 b). The DPR4 (4.85 ± 0.36) and DPR5 (4.85 ± 0.36) groups received the highest scores, respectively, while the DPR1 (2.61 ± 0.74) group received the lowest score. The ROSEM and DPR groups exhibited higher lesion depiction scores in comparison to the OSEM groups, with the exception of the DPR1 group (all p < 0.01 and p = 0.52, respectively). Moreover, the lesion depiction scores of the DPR2 to DPR5 groups did not differ significantly from the ROSEM group (p = 0.08–0.87), while the lesion depiction score of the DPR1 group was inferior to that of the ROSEM group (p < 0.001). The overall image quality exhibited an initial improvement, but subsequently demonstrated a decline with an increase in the filter strength factor (Fig. 6 c). The highest overall image quality score was achieved by the DPR3 group (4.04 ± 0.19), while the lowest score was received by the DPR1 group (2.73 ± 0.44). The overall image quality scores for the DPR2, DPR3, and DPR4 groups were found to be superior to those of the OSEM group (p < 0.01). Although the DPR1 group exhibited a lower overall image quality score than the OSEM group, this difference was not found to be statistically significant (p = 0.21). It is noteworthy that the DPR5 group exhibited a marginally higher overall quality score than the OSEM group (p = 0.82), indicating that radiologists may be amenable to the DPR5 group. Moreover, the overall quality scores of the DPR2 and DPR3 groups were not statistically different from the ROSEM group (p = 0.56 and p = 0.85, respectively). In contrast, the overall quality scores of the DPR1, DPR4, and DPR5 groups were significantly higher than that of the ROSEM group (all p < 0.05). Discussion The objective of this study was to evaluate the performance of the DPR algorithm with regard to depiction and quantification accuracy for small lesions in oncological 18 F-FDG PET/CT imaging. Moreover, the optimal filter strength factor of the DPR algorithm was determined using both phantom and patient data. The results of the phantom study indicated that the DPR2 and DPR3 reconstructions yielded the highest hot sphere CNR, in comparison to the OSEM and ROSEM reconstructions. The patient study demonstrated that the DPR2 reconstruction exhibited lower image noise than the OSEM, the ROSEM, and DPR3 reconstructions. Moreover, the DPR3 reconstruction exhibited comparable SUV max , CNR, and TBR for small lesions to the ROSEM reconstruction. However, the lesion volume derived from the ROSEM reconstruction was found to be more accurate than those of the DPR2 and DPR3 reconstructions, with CT measurements serving as the standard reference. In conclusion, it can be posited that the DPR reconstruction with a filter strength factor between 2 and 3 may provide high-quality images that may be beneficial for the detection of small lesions and the increase in quantification accuracy. In the domain of oncology, the capacity to identify and characterize minute, low-intensity/uptake lesions is of paramount importance for the early diagnosis and staging of patients. Nevertheless, it remains a significant challenge to achieve accurate quantification of radiotracer uptake in the small lesions, primarily due to the overwhelming influence of noise in low signal-to-noise ratio (SNR) images. In order to address this challenge, Rep S et al. [ 15 ] proposed a small-voxel OSEM reconstruction (2 mm in-line pixel size) and evaluated the PET image quality using phantoms with a low target-to-background ratio. The findings indicated that the small-voxel reconstruction method can reliably facilitate the precise delineation of small lesions, enhance lesion contrast, and improve image quality. However, the study did not validate its findings in patient data. Furthermore, the image noise increased as a result of the lower counts per voxel, which may necessitate a longer acquisition time to compensate for the lower counts’ statistics. Another potential avenue for improvement is the application of more sophisticated reconstruction algorithms, including those based on BPL and deep learning techniques. The ROSEM reconstruction, a novel BPL method, has been demonstrated to enhance spatial resolution and lesion contrast for small lesions, thereby facilitating more accurate quantification than the OSEM reconstruction. Nevertheless, the efficacy of small lesion detection is contingent upon the penalization factor. Prior research has indicated that the ROSEM reconstruction with a penalization factor of 0.8–0.9 improves the depiction and quantification accuracy of small lung lesions in oncological 18 F-FDG PET/CT in comparison to the OSEM reconstruction [ 14 ]. These findings are consistent with those observed for other tracers, including 68 Ga-PSMA-11 [ 16 ] and 68 Ga-DOTATATE [ 17 ], although the penalty factor would be regularized as 0.1–0.2. The findings of this study align with those of previous investigations, indicating that the ROSEM algorithm can elevate the SUV max of lesions by 85% in sub-centimeter lesions and 51% in medium-sized lesions, respectively. Recent studies have indicated that deep learning-based image denoising methods have the potential to enhance image quality although this may also lead to a reduction in CNR for small lesions [ 18 , 19 ]. The direct learning of high-SNR images from low-SNR images is inherently unstable, resulting in suboptimal outcomes. The DPR algorithm can address the challenges posed by this issue through a progressive learning approach. In this approach, a denoising network (CNN-DE) is immediately followed by the initial expectation maximization iterations, resulting in an image with minimal noise. In the second expectation maximization iteration, the contrast of small lesions is gradually recovered and then enhanced through an enhancement network (CNN-EH). The DPR algorithm has the potential to enhance the detectability and the accuracy of quantification for small pulmonary nodules. Nevertheless, the efficacy of this approach may be contingent on the filter strength factor, lesion size, and lesion contrast or TBR. The quantitative analysis of the IEC body phantom study revealed that the CNR increased by a factor of 2.8 for DPR2 and 2.3 for DPR3 in comparison to the OSEM method when imaging on the 10-mm hot sphere. However, when the diameter of the hot sphere was increased to 13 mm, the gains decreased to 2.44 and 2.01, respectively, for the same filter strength factor. These results were further confirmed in the patient study and are consistent with the findings of previous studies [ 7 , 9 ]. The clinical data indicated that the DPR2 and DPR3 groups increased by over 41% and 48%, respectively, for medium-centimeter lesions SUV max in comparison to the OSEM group. In the case of sub-centimeter lesions, the SUV max increase was observed to be up to 61% and 68%, respectively, which represented a greater gain than that observed in a previous study [ 20 ]. The aforementioned study evaluated the performance of the DPR and OSEM algorithms in quantifying SUV in 63 sub-centimeter lesions with a mean diameter of (0.76 ± 0.15) cm. The results demonstrated that the average SUV max (11.46) of DPR was approximately 30% higher than that of OSEM (8.90). However, the filter strengthening factor was not provided. Furthermore, the study indicated that the DPR2 algorithm can achieve the greatest gain in CNR in both sub-centimeter and medium-centimeter lesions compared to the OSEM algorithm. In contrast, the DPR3 algorithm demonstrated superior TBR in comparison to the DPR2 algorithm, indicating that the DPR3 algorithm is more effective in enhancing the lesion contrast. Furthermore, the results indicated that the DPR3 reconstruction could provide a marginally more precise estimation of lesion volume than the DPR2 reconstruction when a CT-derived volume was employed as the standard reference. However, all the DPR groups demonstrated an overestimation of the volume of small lesions. To the best of our knowledge, no previous studies have been conducted on the impact of the SUV max threshold on lesion quantification accuracy using the DPR algorithm. The present study suggests that a threshold of greater than 41% for SUV max may enhance the accuracy of lesion segmentation. Figure 7 illustrate PET images of a patient diagnosed with thyroid cancer and lung metastasis, obtained through the application of diverse reconstruction techniques. The images demonstrate the presence of numerous lesions in the maximum intensity projection (MIP) images of the patient. While the small lesions appear indistinct and blurred in the images reconstructed by the OSEM and the DPR1 algorithms, they become more pronounced in the images reconstructed by the DPR2 to DPR5 algorithms. This indicates that an inappropriate selection of the filter strength factor may result in an over-or under-estimation of noise, which could lead to images that are either oversmoothed or under-smoothed (Fig. 7 ). An increase in the filter strength factor may result in the generation of enhanced images; however, it may also lead to the introduction of noise (Fig. 7 ). The selection of an optimal filter strength factor is often a challenging process, influenced by a number of factors, including the specific radiotracer used, the preference of the radiologists, the size of the patient, and the assessment of image quality. Accordingly, the filter strength factor is frequently specified as a range, in accordance with the recommendations of the equipment manufacturer. In clinical practice, it is recommended that the filter strength factor be fixed in order to maintain consistency in SUV measurements. In the visual analysis, the DPR2 reconstruction demonstrated the highest noise score, while the DPR3 reconstruction exhibited the highest lesion depiction score and the highest overall image quality score. As radiologists are primarily concerned with diagnosis, their preference may be influenced by images with a lower background noise level or higher lesion contrast. These images may facilitate the diagnosis of small and low-contrast lesions. The study concluded that, based on visual inspection and quantitative measurements, the optimal choice is DPR reconstruction with a filter strength factor of 3, as it provides superior contrast and lower noise for small lesions. A reliable and precise measurement of radiotracer uptake is of paramount importance for differential diagnosis, treatment planning, and therapy response evaluation [ 21 , 22 ]. Nevertheless, the application of advanced reconstruction techniques, such as the ROSEM and DPR algorithms, has the potential to elevate the SUV of minor lesions and enhance contrast recovery. Such findings may consequently influence the criteria for image interpretation and the evaluation of therapy in subsequent studies. Consequently, it is necessary to update the quantitative interpretation criteria in line with the latest developments in PET technology in the clinical practice. To address these issues, the European Association of Nuclear Medicine (EANM) has recommended that at least two images should be reconstructed for a routine PET examination, with one image displaying the optimal image quality and the other adhering to the EANM Research Ltd. (EARL) specification for quantitative measurement [ 23 ]. Teoh EJ et al. [ 24 ] put forth the proposition that elevated SUV max thresholds may be justified when employing semi-quantitative analyses for the purpose of diagnosing malignancy. Wu Z. et al. [ 25 ] measured the recovery coefficient (RC) of each phantom sphere with a diameter of 10–37 mm, subsequently establishing an association between RC values and the sphere diameter. This can be employed in the partial-volume-effect (PVE) correction to enhance the accuracy of SUV measurements of pulmonary nodules. The results of the phantom and the patient studies demonstrated that the DPR reconstruction method enhanced the accuracy of quantification in relation to the true uptake and the test-retest reliability through the PVE correction. This could facilitate the evaluation of treatment response in follow-up studies, particularly in the case of small lesions. Furthermore, it was observed that DPR1 with the highest smooth strength could achieve comparable and agreeable CR and lesion SUV max with the OSEM reconstruction. This could serve as an alternative solution to address the concern regarding the increasing SUV. It is thus recommended that a harmonized DPR protocol should be implemented in clinical practice in order to ensure consistent results across multi-center trials. It is essential to recognize that this study is constrained by a number of limitations. Firstly, it should be noted that this study was conducted at a single center with a limited number of enrolled patients. A large-scale multicenter study is expected to be anticipated, particularly with regard to the quantification accuracy of SUVs, which is of paramount importance in multicenter or cross-machine PET studies. Secondly, further investigation is required to elucidate the relationship between SUV measurement under the DPR reconstruction and pathological results requires. Such information could potentially facilitate the early differential diagnosis of lung cancer. Thirdly, this study focused exclusively on the comparison of depiction and quantification accuracy in the small lesions among the PET images reconstructed by the OSEM, ROSEM, and DPR algorithms. Further investigation is necessary to determine the potential benefits of the DPR algorithm in the detection of large lesions, the improvement of image quality for obese patients, and the reduction of acquisition time. This is a topic that our research team plans to examine in greater depth. Ultimately, the DPR method has only been trained and applied with 18 F-FDG PET image reconstruction. Consequently, in order for the DPR algorithm to be applied to other tracers, such as prostate-specific membrane antigen (PSMA) and fibroblast activation protein inhibitor (FAPI), it is necessary to either retrain the network using a new dataset or retune it via transfer learning. Conclusions The study demonstrated that the DPR method exhibited superior performance in comparison to the OSEM method in terms of small lesion contrast, image noise suppression, contrast recovery, and volumetric quantification accuracy. The application of a filter strength factor of 3 to the DPR method yielded the optimal contrast-to-noise ratio, thereby demonstrating the highest detectability of small lesions. The findings indicated that DPR3 yielded results comparable to those of the ROSEM method in terms of enhancing the quality of PET images. Abbreviations CNN convolutional neural network DPR deep progressive reconstruction PET/CT positron emission tomography/computed tomography OSEM ordered subset expectation maximization BPL Bayesian penalized likelihood ROSEM regularized OSEM NEMA National Electrical Manufacturers Association CR contrast recovery BV background variability CNR contrast-to-noise ratio RCR radioactivity concentration ratio ROI region of interest VOI volume of interest SD standard deviation SUV standard uptake value COV coefficient of variation Declarations Availability for data and materials The datasets used and analyzed during the current study are available from the corresponding author on reasonable request. Ethical approval and consent to participate All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study was approved by the institutional ethical review board of Nanjing First Hospital, Nanjing Medical University (KY20171208-02), and the written informed consent was waived due to the retrospective nature of this study. Consent for publication Patients signed informed consent regarding the publication of their data and photographs. Competing interests The authors declare that there are no conflicts of interest regarding the publication of this paper. Funding This research was supported by funding from a Jiangsu Provincial Frontier Grant (BE2017612), a Nanjing Municipal Health Science and Technology Development Fund (ZKX22036), a Nanjing Municipal Health Science and Technology Development Fund (YKK20104), and Jiangsu Provincial Medical Key Discipline Cultivation Unit (JSDW202247). Author Contribution LX wrote the manuscript and performed the data analysis. LX, RY, QL-M, and RS-L acquired the data. RC-L and FW carried out the image interpretation. LX, FW, and QL-M conceived and designed the study. All authors read and approved the final manuscript. Acknowledgement We thank all members of the research group. Data Availability The datasets used and analyzed during the current study are available from the corresponding author on reasonable request. References Tahari AK, Wahl RL. Quantitative FDG PET/CT in the community: experience from. interpretation of outside oncologic PET/CT exams in referred cancer patients. J Med Imaging Radiat Oncol . 2014;58(2):183-188. Anand SS, Singh H, Dash AK. Clinical applications of PET and PET-CT. Med J Armed Forces. India. 2009;65(4):353–8. Fletcher JW, Djulbegovic B, Soares HP, et al. Recommendations on the use of 18F-FDG PET in oncology. J Nucl Med. 2008;49:480–508. Adams MC, Turkington TG, Wilson JM, et al. A systematic review of the factors affecting. accuracy of SUV measurements. AJR Am J Roentgenol . 2010;195(2):310-320. Tong S, Alessio AM, Kinahan PE. Image reconstruction for PET/CT scanners: past achievements. and future challenges. Imaging Med . 2010;2(5):529-545. Xu L, Cui C, Li R, et al. Phantom and clinical evaluation of the effect of a new Bayesian penalized likelihood reconstruction algorithm (HYPER Iterative) on 68 Ga-DOTA-NOC PET/CT image quality. EJNMMI Res . 2022;12(1):73. Xing Y, Qiao W, Wang T, et al. Deep learning-assisted PET imaging achieves fast scan/low-dose examination. EJNMMI Phys . 2022;9(1):7. Mehranian A, Wollenweber SD, Walker MD, et al. Image enhancement of whole-body oncology [ 18 F]-FDG PET scans using deep neural networks to reduce noise. Eur J Nucl Med Mol Imaging . 2022;49(2):539-549. Lv Y, Xi C. PET image reconstruction with deep progressive learning. Phys Med Biol . 2021;66(10):10. Wang T, Qiao W, Wang Y, et al. Deep progressive learning achieves whole-body low-dose 18 F-FDG PET imaging. EJNMMI Phys . 2022;9(1):82. Li J, Xi C, Dai H, et al. Enhanced PET imaging using progressive conditional deep image prior. Phys Med Biol . 2023;68(17):10.1088/1361-6560/acf091. Published 2023 Sep 1. Badawi RD, Shi H, Hu P, et al. First Human Imaging Studies with the EXPLORER Total-Body. PET Scanner. J Nucl Med . 2019;60(3):299-303. Li Z, Yang J, Liu Z, et al. Feedback network for image super-resolution[C]//Proceedings of the. IEEE/CVF conference on computer vision and pattern recognition. 2019: 3867-3876. Xu L, Li RS, Wu RZ, et al. Small lesion depiction and quantification accuracy of. oncological 18 F-FDG PET/CT with small voxel and Bayesian penalized likelihood reconstruction. EJNMMI Phys . 2022;9(1):23. Rep S, Tomse P, Jensterle L, et al. Image reconstruction using small-voxel size improves small. lesion detection for positron emission tomography. Radiol Oncol . 2022;56(2):142-149. Yang FJ, Ai SY, Wu R, et al. Impact of total variation regularized expectation maximization. reconstruction on the image quality of 68 Ga-PSMA PET: a phantom and patient study. Br J Radiol . 2021;94(1120):20201356. Liu L, Liu H, Xu S, et al. The impact of total variation regularized expectation maximization. reconstruction on68 Ga-DOTA-TATE PET/CT images in patients with neuroendocrine tumor. Front Med (Lausanne). 2022;9: 845806. Schaefferkoetter J, Yan J, Ortega C, et al. Convolutional neural networks for improving image quality with noisy PET data. EJNMMI Res . 2020;10(1):105. Sanaat A, Shiri I, Arabi H, et al. Deep learning-assisted ultra-fast/low-dose whole-body. PET/CT imaging. Eur J Nucl Med Mol Imaging. 2021;48(8):2405–15. Yang H, Chen S, Qi M, et al. Investigation of PET image quality with acquisition time/bed and. enhancement of lesion quantification accuracy through deep progressive learning. EJNMMI Phys . 2024;11(1):7. Barrington SF, Sulkin T, Forbes A, et al. All that glitters is not gold—new reconstruction. methods using Deauville criteria for patient reporting. Eur J Nucl Med Mol Imaging. 2018;45(2):316–7. Boellaard R, Sera T, Kaalep T, et al. Updating PET/CT performance standards and PET/CT. interpretation criteria should go hand in hand. EJNMMI Res. 2019;9(1):95. Boellaard R, Delgado-Bolton R, Oyen WJ, et al. FDG PET/CT: EANM procedure guidelines for tumour imaging: version 2.0. Eur J Nucl Med Mol Imaging . 2015;42(2):328-354. Teoh EJ, McGowan DR, Bradley KM, et al. Novel penalised likelihood reconstruction of PET. in the assessment of histologically verified small pulmonary nodules. Eur Radiol . 2016;26(2):576-584. Wu Z, Guo B, Huang B, et al. Phantom and clinical assessment of small pulmonary nodules. using Q.Clear reconstruction on a silicon-photomultiplier-based time-of-flight PET/CT system. Sci Rep . 2021;11(1):10328. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 17 Jan, 2026 Read the published version in BMC Medical Imaging → Version 1 posted Editorial decision: Revision requested 27 Oct, 2025 Reviews received at journal 23 Oct, 2025 Reviewers agreed at journal 30 Sep, 2025 Reviews received at journal 06 Jun, 2025 Reviews received at journal 02 Jun, 2025 Reviewers agreed at journal 28 May, 2025 Reviewers agreed at journal 09 May, 2025 Reviewers invited by journal 07 May, 2025 Editor assigned by journal 06 May, 2025 Submission checks completed at journal 04 May, 2025 First submitted to journal 04 May, 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-6366594","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":454175783,"identity":"3cc0780c-ae53-4b79-b5b2-39d1962807a1","order_by":0,"name":"Lei Xu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0ElEQVRIiWNgGAWjYDACdgYDIGnDwNgA4rERo4UZrCWNdC2HoTxitBgcZt4mzdt2Po952hkDhg9lhxn4Zzfg1yLZzFYG1HK7mHF2jgHjjHOHGSTuHMCvhZ+ZxwykJbERqIWZt+0wg4FEAn4tbBAt5yBa/hKjBWrLAYgWRmK0AP1SbDnnXDJQS1rBwZ5z6TwSNwhoMTjevPHGmzK7xI2zkzc++FFmLcc/g4AWIGCR4gVGh2EDA8MBII+HoHogYP744w8DgzwxSkfBKBgFo2BkAgCCRD9rUuj1PwAAAABJRU5ErkJggg==","orcid":"","institution":"Nanjing Medical University","correspondingAuthor":true,"prefix":"","firstName":"Lei","middleName":"","lastName":"Xu","suffix":""},{"id":454175784,"identity":"eb31a540-9b2f-4bfb-be93-a226960b20cb","order_by":1,"name":"Rui Yang","email":"","orcid":"","institution":"Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Rui","middleName":"","lastName":"Yang","suffix":""},{"id":454175785,"identity":"3507a0ea-4805-4d1d-b0a6-bd7b75f25b92","order_by":2,"name":"Ru-shuai Li","email":"","orcid":"","institution":"Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Ru-shuai","middleName":"","lastName":"Li","suffix":""},{"id":454175786,"identity":"8a1bc449-dd79-43ad-b0d6-46f83ac1c03c","order_by":3,"name":"Ren-cong Liu","email":"","orcid":"","institution":"Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Ren-cong","middleName":"","lastName":"Liu","suffix":""},{"id":454175789,"identity":"7810e698-9c52-4992-af92-9aec65d9d0b7","order_by":4,"name":"Qing-le Meng","email":"","orcid":"","institution":"Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Qing-le","middleName":"","lastName":"Meng","suffix":""},{"id":454175791,"identity":"358b4ed0-2e55-4b8d-a096-43f1178b8928","order_by":5,"name":"Feng Wang","email":"","orcid":"","institution":"Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Feng","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2025-04-03 06:53:16","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6366594/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6366594/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12880-026-02166-w","type":"published","date":"2026-01-17T16:29:21+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":82603942,"identity":"880df9d8-819e-43d1-acff-88e3b03d6eed","added_by":"auto","created_at":"2025-05-13 09:54:33","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2117700,"visible":true,"origin":"","legend":"\u003cp\u003ePlots of CR (a), BV (b), CNR (c), residual lung error (d), normalized activity (e) and RCR (f) of hot sphere diameter for different reconstruction methods.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6366594/v1/d9cc9368fa73c8ae45252782.png"},{"id":82603945,"identity":"65e969fe-9fcf-4c18-8126-9339f5dc94e3","added_by":"auto","created_at":"2025-05-13 09:54:33","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":115404,"visible":true,"origin":"","legend":"\u003cp\u003eThe\u003cstrong\u003e \u003c/strong\u003emean and standard deviation\u003cstrong\u003e \u003c/strong\u003eof liver SUV\u003csub\u003emean\u003c/sub\u003e (a), SUV\u003csub\u003esd\u003c/sub\u003e (b), COV (c), and normalized COV (d) for different reconstruction methods.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6366594/v1/688f21cc758342c796b042a6.png"},{"id":82603947,"identity":"10a684ee-1527-4986-86b7-797a7be96f29","added_by":"auto","created_at":"2025-05-13 09:54:33","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":220271,"visible":true,"origin":"","legend":"\u003cp\u003eThe\u003cstrong\u003e \u003c/strong\u003emean and standard deviation\u003cstrong\u003e \u003c/strong\u003eof lesion SUV\u003csub\u003emax\u003c/sub\u003e (a), TBR (b), CNR (c), and volume (d) with different reconstruction methods. Normalization of lesion SUV\u003csub\u003emax \u003c/sub\u003e(e), TBR (f), and CNR (g) from OSEM group, and PET-derived volume normalization from CT measurement (h).\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6366594/v1/9e18be00ab326052542791e1.png"},{"id":82603948,"identity":"5525bd71-cd57-405b-a3d1-041efa7db469","added_by":"auto","created_at":"2025-05-13 09:54:33","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":210737,"visible":true,"origin":"","legend":"\u003cp\u003eThe\u003cstrong\u003e \u003c/strong\u003emean and standard deviation\u003cstrong\u003e \u003c/strong\u003eof lesion SUV\u003csub\u003emax\u003c/sub\u003e (a), TBR (b), CNR (c), and volume (d) divided by lesion diameter with different reconstruction methods. The normalization of lesion SUV\u003csub\u003emax \u003c/sub\u003e(e), TBR (f), and CNR (g) from OSEM group, and PET-derived volume normalization from CT measurement (h) divided by lesion diameter.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6366594/v1/d637f999f567828b6ef2d580.png"},{"id":82605600,"identity":"7129e344-27f6-4737-903f-c876e232fe58","added_by":"auto","created_at":"2025-05-13 10:02:33","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1746107,"visible":true,"origin":"","legend":"\u003cp\u003eLinear regression of liver COV(a) and lesion SUV\u003csub\u003emax\u003c/sub\u003e (b) between OSEM and DPR groups as well as ROSEM group.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6366594/v1/991ab33386e1ab64c012850b.png"},{"id":82603949,"identity":"ca5b2f88-ddb6-434a-bf7d-908674ad0a9b","added_by":"auto","created_at":"2025-05-13 09:54:33","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1019367,"visible":true,"origin":"","legend":"\u003cp\u003eVisual assessment of the image noise (a), lesion depiction (b), and overall image quality (c) with different reconstruction methods.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6366594/v1/108347d6f60f782a8727d9bd.png"},{"id":82603952,"identity":"6f16f914-4c6b-4b36-9548-9ec55d2b7242","added_by":"auto","created_at":"2025-05-13 09:54:33","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":248227,"visible":true,"origin":"","legend":"\u003cp\u003eA patient injected 380 MBq \u003csup\u003e18\u003c/sup\u003eF-FDG diagnosed with thyroid cancer and lung metastasis (162 cm, 70 kg, 2 minutes duration time per bed, and resting for 48 minutes). The diameter of the lung nodule is 0.8 cm on the CT axial view, with a volume of 0.12 cm\u003csup\u003e3\u003c/sup\u003e. For the purposes of illustration, a red circle has been placed on each reconstruction images. The images displayed from top to bottom represent phantom images, maximum intensity projection (MIP) images, liver transverse images, and lung nodule images. The images displayed from left to right were reconstructed using the following methods: OSEM (a), ROSEM (b), DPR1 (c), DPR2 (d), DPR3 (e), DPR4 (f), and DPR5 (g). The display window is designated as SUV\u003csub\u003ebw \u003c/sub\u003e[0,6]. The small sphere and lung nodule were clearly discernible in the ROSEM and DPR2 to DPR5 reconstruction images, but were barely perceptible in the OSEM and the DPR1 MIP images.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-6366594/v1/bc19f95d5c3555d3a9e55ce4.png"},{"id":100614675,"identity":"ccc5848b-b450-46e4-a0d4-61a51a2df30b","added_by":"auto","created_at":"2026-01-19 17:23:14","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6621855,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6366594/v1/c8e584da-3c53-4ba2-ad7b-1ad77c28778d.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Investigate the quantification accuracy of small lesions in oncological 18F-FDG PET/CT using a deep progressive learning reconstruction method","fulltext":[{"header":"Background","content":"\u003cp\u003e \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT is a highly effective functional imaging technology that has been widely adopted in clinical practice [\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. In order to achieve an accurate diagnosis, staging/restaging, and treatment monitoring, it is essential to have precise quantification of the radiotracer and high-quality images. However, image noise and contrast ratio can limit the image quality and quantitative accuracy. Clinical practice has demonstrated that the PET reconstruction algorithm has a substantial effect impact on both image quality and the accuracy of measuring the standardized uptake value (SUV) [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The most commonly used reconstruction algorithm at present is the ordered subset expectation maximization (OSEM) approach. However, the OSEM approach requires a considerable number of iterations to attain optimal quantitative accuracy in the presence of elevated image noise levels. Accordingly, the iterative process must be terminated prematurely. To address this issue, an additional post-processing technique, such as Gaussian filtering, has been employed to reduce the image noise. Nevertheless, this results in a certain degree of compromising with regard to image resolution and lesion detectability. To address this limitation, a regularized image reconstruction algorithm (ROSEM) that permits complete iterative convergence was developed, demonstrating enhanced performance compared to OSEM [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. ROSEM incorporates the pixel-to-pixel total variation, global noise equivalent counts, and local sensitivity profile into a smooth penalty term in the iterative reconstruction process. This process serves to suppress background noise in images and enhance contrast between lesions, thereby obviating the necessity for post-processing filters.\u003c/p\u003e \u003cp\u003eIn recent years, convolutional neural networks (CNNs) have yielded state-of-the-art results in PET image denoising largely due to the rapid advancement of deep learning. Xing et al. [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e] trained a deep learning-based denoising model using 90-second acquisition duration OSEM images with the objective of achieving target 180-second acquisition duration OSEM images. The model demonstrated the capacity to reduce the acquisition duration by 25% \u0026minus;\u0026thinsp;50% without compromising lesion contrast. In a related study, Mehranian et al. [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] developed a deep learning enhancement model using full-duration regularized OSEM images as the training target images. The model was subsequently demonstrated to be capable of reducing acquisition time and injected dose by 50%. Nevertheless, it is difficult to directly learn from input images to target images when there is a considerable disparity between them. Furthermore, the majority of studies utilize deep learning as a post-processing technique, rather than integrating the neural network model into the iterative process. Therefore, incorporating deep learning into the conventional iterative optimization model may prove an effective means of enhancing PET image quality. In a study inspired by a progressive learning strategy, Lv et al. [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] proposed a deep progressive learning (DPL) method for PET image reconstruction. This method serves to bridge the gap between low-quality images and high-quality images through the implementation of denoising and then enhancement steps. The results demonstrate the feasibility of this method for reducing noise and improving lesion contrast in PET images.\u003c/p\u003e \u003cp\u003eIn light of the superior performance of DPL, a revolutionary new deep learning iterative PET reconstruction technique, the Deep Progressive Learning Reconstruction (DPR) algorithm, is currently commercially available on the Chinese market (HYPER DPR, United Imaging Healthcare, China). The DPR algorithm effectively integrates the strengths of multiple CNNs with the exceptional performance of a comprehensive total-body PET/CT training dataset (uEXPLORER, United Imaging Healthcare, China). This integration results in the generation of high-contrast, low-noise PET images, while simultaneously minimizing the associated side effects such as low count statistics, decay effect and partial-volume effect. Preliminary studies have demonstrated that the DPR method can reduce the administered activity of \u003csup\u003e18\u003c/sup\u003eF-FDG by up to two-thirds in a clinical setting while maintaining image quality and allowing for more accurate quantification of lesions in overweight or obese patients [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Nevertheless, the performance of DPR on small lesions remain to be fully elucidated. The aim of this study was to examine the accuracy of the depiction and quantification of \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT images reconstructed using the DPR algorithm for small lung lesions with a diameter of less than 2.0 cm. A compared quantitative and qualitative comparison of image quality metrics was conducted between DPR reconstruction with different filter strengths and OSEM and ROSEM, using data from a NEMA phantom and patient data. The image quality and the accuracy of quantification of small lesions were evaluated in order to determine the optimal filter strength factor for DPR reconstruction.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eDPR algorithm\u003c/h2\u003e \u003cp\u003eDPR is an iterative image reconstruction technique that integrates a denoising network (CNN-DE) and an enhancement network (CNN-EH) into the iterative part of the OSEM algorithm, thereby producing low-noise and high-contrast \u003csup\u003e18\u003c/sup\u003eF-FDG PET images. The CNN-DE was previously trained to differentiate between noise and signal without influencing the metabolic distribution. The CNN-EH was subjected to a pre-training process with the objective of enhancing image contrast. Furthermore, the PET scan reconstruction also integrates the time-of-flight (TOF) and point-spread function (PSF) methods, which can be combined with conventional image filters. The deep learning CNNs were trained using data from uEXPLORER, a total body PET scanner, in which the PET images provided high contrast and minimal noise [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. The dimensions of the training image were 249 \u0026times; 249 \u0026times; 671 with a voxel size of 2.4 \u0026times; 2.4 \u0026times; 2.68 mm\u003csup\u003e3\u003c/sup\u003e and 499 \u0026times; 499 \u0026times; 1342 with a voxel size of 1.2 \u0026times; 1.2 \u0026times; 1.34 mm\u003csup\u003e3\u003c/sup\u003e for the 2.4 mm and 1.2 mm networks, respectively. In order to train the CNN-DE, target images are utilized, which are 15-minute scanning low noise uEXPLORER images. The input images are introduced at a rate of 10% of the total counts. To train the CNN-EH, images with two OSEM iterations and a higher number of OSEM iterations were employed as the target images and input images, respectively. The CNN-DE and CNN-EH are both designed based on the feedback network [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. The training dataset was constructed using a total of 161,040 image slice pairs from 80 patients. This comprised 53,680 and 107,360 slice pairs for the 2.4 mm and 1.2 mm networks, respectively. The test dataset was constructed using a total of 40,260 image slice pairs from 20 patients, including 13,420 and 26,840 slice pairs for the 2.4 mm and 1.2 mm models, respectively. Further details on the design of the DPR algorithm design, as well as the network training and testing procedures, can be found in the reference literature [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePhantom study\u003c/h3\u003e\n\u003cp\u003eIn this study, a National Electrical Manufacturers Association (NEMA) International Electrotechnical Commission (IEC) body phantom was utilized, comprising six fillable spheres with inner diameters of 10 mm, 13 mm, 17 mm, 22 mm, 28 mm, and 37 mm, respectively. Furthemore, a cylinder lung insert with a diameter of 5 cm and a length of 16 cm was positioned at the center of the phantom. The phantom background was filled with an \u003csup\u003e18\u003c/sup\u003eF-FDG solution at an activity concentration of 8.26 kBq/ml, while the four smallest spheres were filled with an \u003csup\u003e18\u003c/sup\u003eF-FDG solution at 33.04 kBq/ml. This yielded a sphere-to-background ratio of 4:1.\u003c/p\u003e\n\u003ch3\u003ePatients study\u003c/h3\u003e\n\u003cp\u003eThe study population consisted of 30 consecutive patients who were referred to the Nanjing First Hospital between July 20 and December 20, 2022, and underwent \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT examinations for cancer staging or restaging. The study\u0026rsquo;s inclusion criteria were as follows: the presence of \u003csup\u003e18\u003c/sup\u003eF-FDG-avid lesions was identified in the lungs, the target lesion was able to be segmented on CT images with a diameter of less than 2.0 cm, and the availability of list-mode raw data for additional PET reconstructions. The exclusion criteria were as follows: the \u003csup\u003e18\u003c/sup\u003eF-FDG uptake time was found to be longer than 100 minutes or less than 40 minutes in three patients. Furthermore, it was observed that respiratory motion had a considerable impact on the accuracy of lesion segmentation on PET images in a single patient. In conclusion, the study cohort comprised 26 patients, with a female-to-male ratio of 11:15, and an age range of 33 to 78 years. This retrospective study was approved by the institutional review board of Nanjing First Hospital and did not require to written informed consent.\u003c/p\u003e\n\u003ch3\u003ePET/CT procedure for the phantom and patient study\u003c/h3\u003e\n\u003cp\u003eThe PET/CT scans were conducted using a digital PET/CT scanner (uMI780, United Imaging Healthcare, Shanghai, China), with a 30 cm axial length and a system sensitivity of 15 kcps/MBq. Prior to the administration of \u003csup\u003e18\u003c/sup\u003eF-FDG, patients were required to fast for a minimum of six hours and their blood glucose levels were confirmed to be less than 10 mmol/mL. A weight-based dose of 5.0 MBq/kg \u003csup\u003e18\u003c/sup\u003eF-FDG was administered to the patients as an intravenous bolus. Prior to the commencement of the scanning procedure, patients were instructed to consume a volume of water between 0.5 and 1.0 liters. The patients and phantom were subjected to CT scanning with a fixed tube voltage of 120 kV and an automated milliampere-second (mAs) technique for dose modulation. This provided the necessary anatomical information and attenuation correction for the PET images. Subsequently, a whole-body PET scan was acquired in 3D list mode for a period of two minutes per bed position. This provided coverage of the area from the skull base to the mid-thigh of the patient, or approximately 30 cm to encompass the entire phantom.\u003c/p\u003e\n\u003ch3\u003ePET image reconstruction\u003c/h3\u003e\n\u003cp\u003eThe PET images were reconstructed into seven groups, comprising the routine OSEM, regularized OSEM, and DPR algorithms, with five filter strengths ranging from 1 to 5, indicating smooth to sharp. In the following sections, the aforementioned groups will henceforth be referred to as OSEM, ROSEM, DPR1, DPR2, DPR3, DPR4, and DPR5, respectively. The OSEM group was implemented using two iterations and 20 subsets. Additionally, a Gaussian filter with a full width at half maximum of 3 mm, a 256 \u0026times; 256 matrix, and a 600 mm field of view (FOV) was employed. Additionally, a slice thickness of 3 mm, TOF, and a PSF model were employed. The necessary corrections, including those for scatter, random, dead time, decay, attenuation, and normalization, were incorporated into the reconstruction. The ROSEM group employed HYPER Iterative, a commercially available implementation of Bayesian penalized likelihood (BPL) algorithms that incorporates a total variation regulator and the sensitivity profile of PET scanners into the penalization term. The penalty factor is a hyperparameter that regulates the image contrast and smoothness. The operator can adjust this value between 0 and 1. In this study, a penalty factor of 0.8 was selected based on the findings of a preliminary study [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], which demonstrated that this factor yielded optimal accuracy in depicting and quantifying small lesions in oncological \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT. Furthermore, the DPR algorithm was employed for PET reconstruction without the application of any additional post-processing methods. The FOV, matrix, and slice thickness utilized in the ROSEM and DPR groups were identical to those employed in the OSEM group. The CT images were reconstructed with a FOV of 600 mm, a matrix of 512 \u0026times; 512, and a slice thickness of 3 mm, with 1.5 mm increments.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eQuantitative evaluation of the phantom and patient images\u003c/h2\u003e \u003cp\u003eThe quantitative analyses were conducted by a senior nuclear radiologist on a dedicated workstation (uWI-MI, United Imaging Healthcare, China). For the IEC body phantom study, the quality of the PET images was evaluated in accordance with the NEMA NU-2012 protocol. The region of interest (ROI) was delineated at the center of each hot sphere with a diameter matching that of the sphere. The background ROI was delineated in the peripheral area of the phantom background at the central slice of the spheres, as well as \u0026plusmn;\u0026thinsp;1cm and \u0026plusmn;\u0026thinsp;2cm from the central axis. A total of 60 background ROIs of varying size, with 12 ROIs on each of the five slices, are to be drawn. It is imperative that the locations of all ROIs remain fixed between successive measurements, and that the mean counts in each background ROI are duly recorded.\u003c/p\u003e \u003cp\u003eThe percent contrast recovery (CR) and background variation (BV) were calculated using the appropriate equations, namely (1) and (2). The contrast-to-noise ratio (CNR), a measure of the signal level in the presence of noise, was calculated using equations (3). The residual lung error (LE) was calculated using Eq.\u0026nbsp;(4). The radioactivity concentration ratio (RCR) was determined by dividing the measured activity by the injected activity of the hot sphere. Furthermore, the normalized activity of four hot spheres was calculated as the mean activity concentration of all reconstruction groups in relation to that of the OSEM group. This demonstrates the comparative change resulting from the various reconstructions, with the OSEM serving as the reference.\u003c/p\u003e \u003cp\u003eIn the context of the patient study, a volume of interest (VOI) with a diameter of 3 cm was delineated manually at the same position on a homogeneous area of the right liver lobe for each image. The mean and standard deviation (SUV\u003csub\u003emean\u003c/sub\u003e and SUV\u003csub\u003esd\u003c/sub\u003e) were recorded, respectively. The coefficient of variation (COV) in the liver, which serves as a measure of background noise, was calculated by dividing the SUV\u003csub\u003esd\u003c/sub\u003e by the SUV\u003csub\u003emean\u003c/sub\u003e. A semi-automatic segmentation tool (MI-Oncology, United Imaging Healthcare, Shanghai, China) was utilized to delineate each small \u003csup\u003e18\u003c/sup\u003eF-FDG-avid lung lesion on PET and CT images. The volume and maximum of standardized uptake value (SUV\u003csub\u003emax\u003c/sub\u003e) of the lesion were quantified on the PET image using a 41% SUV\u003csub\u003emax\u003c/sub\u003e threshold. Only lesions with a diameter of less than 2.0 cm, as determined on the corresponding CT image, were included in the analysis. Furthermore, the lesion was segmented using CT images, and the resulting volume was employed as a standard reference for comparison with PET- and CT-derived volumes. To evaluate the image contrast, the tumor-to-background ratio (TBR) was calculated by dividing the lesion\u0026rsquo;s SUV\u003csub\u003emax\u003c/sub\u003e by the liver\u0026rsquo;s SUV\u003csub\u003emean\u003c/sub\u003e. The contrast-to-noise ratio (CNR) of the lesion was calculated by dividing the lesion SUV\u003csub\u003emax\u003c/sub\u003e by the liver\u0026rsquo;s SUV\u003csub\u003esd\u003c/sub\u003e.\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:{CR}_{H,j}=\\frac{{C}_{H,j}/{C}_{B,j}-1}{A-1}\\times\\:100\\%\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(1\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:{BV}_{j}=\\frac{\\sqrt{\\sum\\:_{k=1}^{K}{\\left({C}_{B,j,k}-{C}_{B,j}\\right)}^{2}/\\left(K-1\\right)}}{{C}_{B,j}}\\times\\:100\\%\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(2\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:{CNR}_{H,j}=\\frac{{CR}_{H,j}}{{BV}_{j}}\\times\\:100\\%\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(3\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equd\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e\n$$\\:{LE}_{i}=\\frac{{C}_{lung,i}}{{C}_{B,i}}\\times\\:100\\%\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(4\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe notation \u003cem\u003eC\u003c/em\u003e\u003csub\u003e\u003cem\u003eH,j\u003c/em\u003e\u003c/sub\u003e represents the average counts within a ROI for hot sphere \u003cem\u003ej\u003c/em\u003e. The symbol \u003cem\u003eBV\u003c/em\u003e\u003csub\u003e\u003cem\u003ej\u003c/em\u003e\u003c/sub\u003e signifies the background variability for sphere \u003cem\u003ej\u003c/em\u003e, and \u003cem\u003eA\u003c/em\u003e is the standard hot sphere to background concentration ratio. The symbol \u003cem\u003eC\u003c/em\u003e\u003csub\u003e\u003cem\u003elung,i\u003c/em\u003e\u003c/sub\u003e denotes the average counts for an ROI drawn in the lung insert on slice \u003cem\u003ei\u003c/em\u003e, while \u003cem\u003eC\u003c/em\u003e\u003csub\u003e\u003cem\u003elung,i\u003c/em\u003e\u003c/sub\u003e represents the average background counts for slice \u003cem\u003ei\u003c/em\u003e. Finally, the symbol \u003cem\u003eK\u003c/em\u003e\u0026thinsp;=\u0026thinsp;60 signifies the number of selected background ROIs.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eVisual image quality assessment for the patient images\u003c/h3\u003e\n\u003cp\u003eThe quality of the PET image was evaluated by two nuclear radiologists, each with over 10 years of experience, on a dedicated workstation (uWS-MI R004, United Imaging Healthcare, China). To mitigate potential bias, the order in which the seven series of PET images were evaluated was randomized for each patient. Furthermore, the readers were blinded to the patient\u0026rsquo;s clinical information and reconstruction settings. A five-point Likert scale was employed to evaluate three perspectives: image noise, the lesion depiction, and the overall image quality. The rating scale employed was as follows: 1\u0026thinsp;=\u0026thinsp;The image quality is poor, with an excessive amount of noise or unfavorable lesion contrast, and insufficient lesion depiction. 2\u0026thinsp;=\u0026thinsp;The image quality is deemed unacceptable, with suboptimal noise and blurring of lesions, which impairs diagnostic confidence. 3\u0026thinsp;=\u0026thinsp;Acceptable image quality, comparable to that of routine oncological \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT images, with an appropriate level of noise and average lesion delineation that enables a clinical diagnosis. 4\u0026thinsp;=\u0026thinsp;The image quality is optimal, with the desired level of noise and the capacity to detect small lesions with precision, thereby fostering robust diagnostic confidence. A score of 5 indicates an excellent image quality, with optimal noise, clear contrast of lesions, and accurate depiction and quantification of small lesions, thereby instilling complete confidence in the diagnostic process. The scores for image quality, which were independently assessed by the two raters, were recorded for the purposes of evaluating inter-reader agreement test. In the event of a discrepancy between the two radiologists, a consensus meeting was convened with a third physician proficient in nuclear medicine to reach a final determination.\u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eThe mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation is employed to present the data. A two-tailed paired t-test was employed to assess quantitative data that exhibited a normal distribution, as determined by the Shapiro-Wilk normality test. The \u003cem\u003ep\u003c/em\u003e-value was adjusted using the Benjamini \u0026amp; Hochberg correction to account for the false discovery rate resulting from multiple comparisons. A simple linear regression was employed to quantify the degree of agreement between liver COV and lesion SUV\u003csub\u003emax\u003c/sub\u003e between OSEM and the other reconstruction groups. To assess the inter-reader agreement, a Cohen\u0026rsquo;s kappa test was employed. The third reader provided qualitative assessments of image quality scores, which were subsequently evaluated using a matched-pairs Wilcoxon signed-rank test. A p-value of less than 0.05 was considered to indicate statistical significance. All data were processed using GraphPad Prism 9.0 and Microsoft Excel 2022 software.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003ePhantom study\u003c/h2\u003e \u003cp\u003eThe CR, BV, CNR, LE, RCR, and normalized activity of the NEMA phantom with various reconstruction methods are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. For all reconstruction groups, the trend of CRs exhibited an initial increase, followed by a decrease with the increase in the diameter of the hot sphere. The highest values for the CRs were observed for DPR4 and DPR5, followed by DPR2, DPR3, and ROSEM, and reaching their lowest values for DPR1 and OSEM across all spheres. With the exception of the 22 mm hot sphere, the CRs exhibited by DPR1 were equivalent to those exhibited by OSEM for each hot sphere. Moreover, the CRs exhibited a near equivalence between DPR4 and DPR5, with the discrepancies confined to a narrow range (-0.4\u0026ndash;2.1%). As the diameter of the hot sphere increased, the discrepancy in CRs responsible for DPR reconstruction groups diminished. The CRs of the DPR2 and DPR3 were observed to be higher than those of ROSEM at the diameters of the 17-mm and 22-mm hot spheres. It is noteworthy that ROSEM achieved a CR of 71.2% for the 10 mm hot sphere, which was higher than the corresponding values for DPR2 and DPR3 of 64.4% and 68.5%, respectively. Upon increasing the diameter of the sphere to 13 mm, the CRs exhibited a near-equivalence between DPR3 (75.1%) and ROSEM (75.0%) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). The BVs of each reconstruction group exhibited a declining trend as the diameter of the hot sphere increased, with the gap progressively narrowing to a narrow range. Additionally, the lowest BVs were observed in DPR2, followed by DPR3, DPR1, ROSEM, DPR4, and DPR5, while the highest BVs were observed in OSEM at the same diameter of the hot sphere. Furthermore, the BVs of DPR1 were found to be comparable to those of ROSEM. Notably, the BVs for DPR2 to DPR3 were observed to be 2.2\u0026ndash;2.8% for the smallest 10 mm hot sphere. The corresponding values for OSEM and ROSEM were 5.1% and 3.2%, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb).\u003c/p\u003e \u003cp\u003eThe CNRs of each reconstruction group demonstrated an increase as the diameter of the spheres increased. The CNRs of each sphere exhibited the highest values for DPR2, followed closely by DPR3, and the lowest values for OSEM. In particular, the CNR of DPR2 was 29.27 for the smallest sphere with a diameter of 10 mm. Moreover, the CNRs of DPR1 were observed to be comparable to those of ROSEM, with the exception of the 10 mm hot sphere (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec). In all reconstruction groups, the highest residual lung error was observed for DPR1 (7.6%), followed by OSEM (7.0%) and ROSEM (5.6%). The discrepancies between DPR2 and DPR5 were minimal, spanning a narrow range of 4.3\u0026ndash;3.6% (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ed). All reconstruction groups exhibited a mean normalized activity value exceeding 1.0, demonstrating a slight increase with the enhancement filter strength. The mean normalized activity of DPR1 was found to be highly comparable to that of OSEM, while that of ROSEM was found to be similar to that of DPR2 and DPR3 (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ee). A positive correlation was observed between the RCRs and the diameter of the hot sphere for all reconstruction groups. Furthermore, an increase in the strength of the filter factor was accompanied by a corresponding raise in the RCRs of each sphere. The DPR4 and DPR5 groups exhibited the highest RCRs, while the OSEM and DPR1 groups demonstrated the lowest RCRs. The RCRs of DPR1 were found to be equivalent to those of OSEM, with the exception of the 22 mm hot sphere. Furthermore, the RCRs of the ROSEM group were observed to be higher than those of the DPR2 and DPR3 groups in the 10 mm and 13 mm hot spheres, while being lower than those of the DPR2 and DPR3 groups in the 17 mm and 22 mm hot spheres (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ef).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003ePatient characteristics\u003c/h2\u003e \u003cp\u003eThe study cohort comprised 26 lung lesions. The diameter of these lesions was 0.94\u0026thinsp;\u0026plusmn;\u0026thinsp;0.18 cm (range 0.7\u0026ndash;1.4 cm), which corresponds to a volume of 0.26\u0026thinsp;\u0026plusmn;\u0026thinsp;0.14 cm\u003csup\u003e3\u003c/sup\u003e (range 0.10\u0026ndash;0.74 cm\u003csup\u003e3\u003c/sup\u003e) as determined by CT imaging. The mean body weight of the patients was 65.7\u0026thinsp;\u0026plusmn;\u0026thinsp;11.6 kg (range 41\u0026ndash;93 kg), while the mean height was 1.65\u0026thinsp;\u0026plusmn;\u0026thinsp;0.07 m (range 1.52\u0026ndash;1.78 m). The administered activity of \u003csup\u003e18\u003c/sup\u003eF-FDG was 327.12\u0026thinsp;\u0026plusmn;\u0026thinsp;48.4 MBq (range 218.3\u0026ndash;411.14 MBq), with an average uptake time of 62.81\u0026thinsp;\u0026plusmn;\u0026thinsp;14.82 minutes (range 46\u0026ndash;97 minutes). The primary cancer type and other clinical characteristics are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e for reference.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePatient clinical characteristics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameter\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eValue\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber of patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e26\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e58.3\u0026thinsp;\u0026plusmn;\u0026thinsp;14.8 [33, 78]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGender\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeight (m)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.65\u0026thinsp;\u0026plusmn;\u0026thinsp;0.07 [1.52, 1.78]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeight (kg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e65.7\u0026thinsp;\u0026plusmn;\u0026thinsp;11.6 [41, 93]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody mass index (kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24.14\u0026thinsp;\u0026plusmn;\u0026thinsp;3.56 [17.06, 32.18]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInjected activity (MBq)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e327.12\u0026thinsp;\u0026plusmn;\u0026thinsp;48.4 [218.3, 411.14]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInjected activity per weight (MBq/kg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.04\u0026thinsp;\u0026plusmn;\u0026thinsp;0.66 [3.95, 6.50]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUptake time (minutes)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e62.81\u0026thinsp;\u0026plusmn;\u0026thinsp;14.82 [46, 97]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePrimary cancer type\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLung cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParaganglioma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLymphoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRectal cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePancreatic cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThyroid cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNeuroendocrine tumor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVestibular adenocarcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBreast cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGastric cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLiver cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eQuantitative image evaluation for the patient study\u003c/h2\u003e \u003cp\u003eThe mean SUV\u003csub\u003emean\u003c/sub\u003e for the liver was approximately 2.55 across all reconstruction groups (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea). The liver SUV\u003csub\u003esd\u003c/sub\u003e demonstrated an increase with the enhancement filter strength in the DPR groups, with the exception of DPR1. The lowest liver SUV\u003csub\u003esd\u003c/sub\u003e were observed in the DPR2 reconstruction group. The mean liver SUV\u003csub\u003esd\u003c/sub\u003e of DPR5 was slightly higher than that of the OSEM group, while the mean liver SUV\u003csub\u003esd\u003c/sub\u003e of DPR3 were comparable to those of the ROSEM group (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb). The liver COVs for all reconstruction groups were less than 15%, with an average of less than 11% (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec). The DPR1 to DPR4 groups exhibited significantly lower liver COVs than the OSEM group (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), whereas no significant difference was observed between the DPR5 and OSEM groups (p\u0026thinsp;=\u0026thinsp;0.48) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed). The liver COVs for the ROSEM group were observed to be higher than those for the DPR1 group (p\u0026thinsp;=\u0026thinsp;0.09), yet lower than those for the DPR3 group (p\u0026thinsp;=\u0026thinsp;0.65). The liver COVs for DPR2 were found to be significantly lower than those for ROSEM (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), while the COVs for DPR4 and DPR5 were found to be significantly higher than those for ROSEM (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Moreover, the mean lesion SUV\u003csub\u003emax\u003c/sub\u003e exhibited a positive correlation with increasing filter strength factor (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea, \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ee). The lesion SUV\u003csub\u003emax\u003c/sub\u003e of the ROSEM and all DPR groups was significantly higher than that of the OSEM group, with the exception of the DPR1 group (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05 and p\u0026thinsp;=\u0026thinsp;0.40, respectively). The lesion SUV\u003csub\u003emax\u003c/sub\u003e of the ROSEM group was found to be significantly higher than that of the DPR1, DPR2, and DPR3 groups, respectively (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Meanwhile, the DPR4 and DPR5 groups demonstrated results comparable to the ROSEM group (p\u0026thinsp;=\u0026thinsp;0.61and p\u0026thinsp;=\u0026thinsp;0.18, respectively). The mean normalized lesion SUV\u003csub\u003emax\u003c/sub\u003e increased by 53\u0026ndash;76% for the DPR2 to DPR5 groups, and by 71% for the ROSEM groups in comparison to the OSEM group (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ee).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eQuantitative results of the patient study (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMethods\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLiver SUV\u003csub\u003emean\u003c/sub\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLiver COV (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLesion SUV\u003csub\u003emax\u003c/sub\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLesion TBR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLesion CNR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLesion volume (cm\u003csup\u003e3\u003c/sup\u003e)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOSEM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.55\u0026thinsp;\u0026plusmn;\u0026thinsp;0.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e9.80\u0026thinsp;\u0026plusmn;\u0026thinsp;1.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e4.02\u0026thinsp;\u0026plusmn;\u0026thinsp;0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e1.63\u0026thinsp;\u0026plusmn;\u0026thinsp;0.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e16.68\u0026thinsp;\u0026plusmn;\u0026thinsp;4.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c7\"\u003e \u003cp\u003e0.64\u0026thinsp;\u0026plusmn;\u0026thinsp;0.29\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eROSEM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.55\u0026thinsp;\u0026plusmn;\u0026thinsp;0.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e5.71\u0026thinsp;\u0026plusmn;\u0026thinsp;1.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e6.82\u0026thinsp;\u0026plusmn;\u0026thinsp;1.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e2.73\u0026thinsp;\u0026plusmn;\u0026thinsp;0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e49.38\u0026thinsp;\u0026plusmn;\u0026thinsp;15.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c7\"\u003e \u003cp\u003e0.31\u0026thinsp;\u0026plusmn;\u0026thinsp;0.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDPR1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.55\u0026thinsp;\u0026plusmn;\u0026thinsp;0.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e4.91\u0026thinsp;\u0026plusmn;\u0026thinsp;1.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e3.88\u0026thinsp;\u0026plusmn;\u0026thinsp;0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e1.57\u0026thinsp;\u0026plusmn;\u0026thinsp;0.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e32.98\u0026thinsp;\u0026plusmn;\u0026thinsp;10.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c7\"\u003e \u003cp\u003e0.65\u0026thinsp;\u0026plusmn;\u0026thinsp;0.33\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDPR2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.55\u0026thinsp;\u0026plusmn;\u0026thinsp;0.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e4.21\u0026thinsp;\u0026plusmn;\u0026thinsp;0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e6.11\u0026thinsp;\u0026plusmn;\u0026thinsp;1.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e2.46\u0026thinsp;\u0026plusmn;\u0026thinsp;0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e59.92\u0026thinsp;\u0026plusmn;\u0026thinsp;15.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c7\"\u003e \u003cp\u003e0.37\u0026thinsp;\u0026plusmn;\u0026thinsp;0.18\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDPR3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.55\u0026thinsp;\u0026plusmn;\u0026thinsp;0.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e5.84\u0026thinsp;\u0026plusmn;\u0026thinsp;1.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e6.38\u0026thinsp;\u0026plusmn;\u0026thinsp;1.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e2.57\u0026thinsp;\u0026plusmn;\u0026thinsp;0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e45.09\u0026thinsp;\u0026plusmn;\u0026thinsp;11.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c7\"\u003e \u003cp\u003e0.35\u0026thinsp;\u0026plusmn;\u0026thinsp;0.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDPR4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.55\u0026thinsp;\u0026plusmn;\u0026thinsp;0.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e8.25\u0026thinsp;\u0026plusmn;\u0026thinsp;1.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e6.73\u0026thinsp;\u0026plusmn;\u0026thinsp;1.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e2.70\u0026thinsp;\u0026plusmn;\u0026thinsp;0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e33.28\u0026thinsp;\u0026plusmn;\u0026thinsp;7.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c7\"\u003e \u003cp\u003e0.33\u0026thinsp;\u0026plusmn;\u0026thinsp;0.15\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDPR5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.55\u0026thinsp;\u0026plusmn;\u0026thinsp;0.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e10.3\u0026thinsp;\u0026plusmn;\u0026thinsp;1.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e6.96\u0026thinsp;\u0026plusmn;\u0026thinsp;1.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e2.79\u0026thinsp;\u0026plusmn;\u0026thinsp;0.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e27.46\u0026thinsp;\u0026plusmn;\u0026thinsp;6.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c7\"\u003e \u003cp\u003e0.34\u0026thinsp;\u0026plusmn;\u0026thinsp;0.19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.26 \u0026plusmn;\u0026thinsp;0.14\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"7\"\u003e*Data are presented as the mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe DPR5 group exhibited the highest mean lesion TBR of all the reconstruction groups. As the filter strength increased, the lesion TBR also increased (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb). The TBRs of the DPR2 to DPR5 groups were found to be significantly higher than those of the OSEM group (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), whereas the DPR1 and OSEM exhibited comparable TBRs (p\u0026thinsp;=\u0026thinsp;0.37). The lesion TBR of the DPR1, DPR2, and DPR3 groups was significantly lower than that of the ROSEM group (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), while the lesion TBR of the DPR4 and DPR5 groups was comparable to that of the ROSEM group (p\u0026thinsp;=\u0026thinsp;0.70 and p\u0026thinsp;=\u0026thinsp;0.26, respectively). The mean normalized lesion TBR increased by 1.53 to 1.75 times for the DPR2 to DPR5 groups, and by 1.71 times for the ROSEM group in comparison to the OSEM group (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ef). The mean lesion CNR of all the DPR groups and the ROSEM group were found to be significantly higher than that of the OSEM group (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). The DPR2 group exhibited the highest CNRs among all reconstruction groups (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ec). The CNR of the ROSEM group was found to be significantly lower than that of the DPR2 group (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), but significantly higher than that of the DPR1, DPR4, and DPR5 groups (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). The mean CNR of the DPR3 groups was observed to be smaller than that of the ROSEM group, although the difference was not statistically significant (p\u0026thinsp;=\u0026thinsp;0.75). The mean normalized lesion CNR exhibited a 1.67-to-3.63-fold increase for the DPR groups and a 2.98-fold increase for the ROSEM group in comparison to the OSEM group (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eg). The volume of the lesions was found to be significantly higher in all the DPR groups compared to that measured on CT images (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Nevertheless, the lesion volume derived from the ROSEM group was statistically equivalent to that measured on CT images (p\u0026thinsp;=\u0026thinsp;0.48) (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ed). The lesion volume of the OSEM group was found to be significantly larger than that of the ROSEM group and all DPR groups, with the exception of DPR1 (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05 and p\u0026thinsp;\u0026gt;\u0026thinsp;0.99, respectively). The mean lesion volume was observed to be 2.66 times that of the OSEM and DPR1 groups, and 1.49 to 1.33 times that of the DPR2 to DPR5 groups, and 1.22 times that of the ROSEM group in comparison to the CT volumetric measurement (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eh). Therefore, the volume derived from the ROSEM group was found to be more accurate than those derived from the OSEM group and all the DPR groups when CT volumetric measurement was employed as the standard reference.\u003c/p\u003e \u003cp\u003eIn a subsequent subgroup analysis, the small lesions were divided into two categories based on their diameters: sub-centimeter (D\u0026thinsp;\u0026lt;\u0026thinsp;1.0 cm, n\u0026thinsp;=\u0026thinsp;16, range 0.7\u0026ndash;0.9 cm) and medium-size (1.0\u0026thinsp;\u0026le;\u0026thinsp;D\u0026thinsp;\u0026lt;\u0026thinsp;2.0 cm, n\u0026thinsp;=\u0026thinsp;10, range 1.0 -1.8 cm). In each category, the lesion SUV\u003csub\u003emax\u003c/sub\u003e of the DPR groups exhibited an increasing trend with the enhancement of filter strength. The mean normalized SUV\u003csub\u003emax\u003c/sub\u003e for sub-centimeter lesions was observed to be higher than that for medium-size lesions across the ROSEM and all the DPR groups, with the exception of the DPR1 group at the same filter strength \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ee). The mean normalized SUV\u003csub\u003emax\u003c/sub\u003e was greater than 1.6 for sub-centimeter lesions (range 1.61\u0026ndash;1.85) and 1.4 for medium-sized lesions (range 1.41\u0026ndash;1.61) across all the DPR groups, with the exception of the DPR1 group. Nevertheless, the average normalized SUV\u003csub\u003emax\u003c/sub\u003e exhibited only minor differences between sub-centimeter and medium-sized lesions in the DPR1 group. The average normalized lesion SUV\u003csub\u003emax\u003c/sub\u003e of the ROSEM group demonstrated a significant improvement, reaching 1.85 times that of sub-centimeter lesions and 1.51 times that of medium-sized lesions, respectively. A comparable pattern was observed in the metric of lesion TBR (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ef). The impact of lesion diameter on the lesion CNR for different reconstruction techniques is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eg. The mean lesion CNR of the DPR groups initially increases with filter strength, reaching a maximum value, and then subsequently declines. This trend reaches a maximum at the DPR2 groups for both the sub-centimeter and medium-size categories. The mean normalized lesion CNR of the sub-centimeter group was observed to be greater than that of the medium-size group for all reconstruction procedures, with the exception of the DPR1 groups. The DPR2 group exhibited the most substantial increase in both groups, with a 3.81-fold rise in the sub-centimeter cohort and a 3.34-fold surge in the medium-size group. When the reconstruction settings were identical, the mean of the normalized lesion volume for the sub-centimeter group was observed to be greater than that of the medium-size group (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eh). The mean of the normalized lesion volume for the OSEM group exhibited the most pronounced fluctuations between the sub-centimeter and medium-size categories, with a range of 2.96 to 2.18. In contrast, the mean of the normalized lesion volume for the ROSEM group was approximately 1.2, with a value of 1.25 for the sub-centimeter group and 1.17 for the medium-size group. Moreover, minor discrepancies were identified between the sub-centimeter and medium-size categories within the DPR5 group.\u003c/p\u003e \u003cp\u003eA linear regression of liver COV and lesion SUV\u003csub\u003emax\u003c/sub\u003e revealed a significant degree of agreement between the OSEM method and other reconstruction methods (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e. The liver COV concordance correlation coefficients for the DPR groups (R\u0026thinsp;=\u0026thinsp;0.91\u0026ndash;0.94) were all greater than 0.91, with the exception of the DPR1 group (R\u0026thinsp;=\u0026thinsp;0.60). The liver COV concordance correlation coefficients were found to be comparable between the DPR1 and ROSEM groups (R\u0026thinsp;=\u0026thinsp;0.60 and R\u0026thinsp;=\u0026thinsp;0.61, respectively). As the filter strength increased, the concordance correlation coefficients of lesion SUV\u003csub\u003emax\u003c/sub\u003e between the OSEM and DPR groups demonstrated a decline (R\u0026thinsp;=\u0026thinsp;0.91\u0026thinsp;\u0026minus;\u0026thinsp;0.75). The lowest concordance correlation coefficient was observed between the OSEM and ROSEM groups (R\u0026thinsp;=\u0026thinsp;0.71).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eVisual image scores for the patient study\u003c/h2\u003e \u003cp\u003eThe inter-reader agreement for visual noise, lesion depiction, and overall image quality was substantial agreement between the readers, with kappa values of 0.695, 0.697, and 0.771, respectively. The results for visual noise, lesion depiction, and overall image quality scores are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. As the filter strength increased, the visual noise score initially increased and subsequently declined (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea). The visual noise score achieved by the DPR2 group was the highest (4.54\u0026thinsp;\u0026plusmn;\u0026thinsp;0.50), while that achieved by the OSEM group and the DPR5 group was the lowest (2.88\u0026thinsp;\u0026plusmn;\u0026thinsp;0.32 and 2.88\u0026thinsp;\u0026plusmn;\u0026thinsp;0.58, respectively). The visual noise scores for the DPR1, DPR2, and DPR3 groups were found to be superior to those of OSEM (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). In contrast, the DPR4 and DPR5 groups exhibited higher visual scores, although the difference was not statistically significant when compared to the OSEM group (p\u0026thinsp;=\u0026thinsp;0.149 and p\u0026thinsp;=\u0026thinsp;0.923, respectively). In comparison to the ROSEM group, the DPR2, DPR4, and DPR5 groups exhibited elevated visual noise scores relative to the ROSEM group (p\u0026thinsp;=\u0026thinsp;0.021, p\u0026thinsp;=\u0026thinsp;0.0008, and p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, respectively), whereas the DRP1 and DPR3 groups exhibited the same visual noise scores as the ROSEM groups (p\u0026thinsp;=\u0026thinsp;0.748 and p\u0026thinsp;=\u0026thinsp;0.608, respectively).\u003c/p\u003e \u003cp\u003eThe lesion depiction score was observed to improve with an elevate filter strength (4.35\u0026ndash;4.75) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb). The DPR4 (4.85\u0026thinsp;\u0026plusmn;\u0026thinsp;0.36) and DPR5 (4.85\u0026thinsp;\u0026plusmn;\u0026thinsp;0.36) groups received the highest scores, respectively, while the DPR1 (2.61\u0026thinsp;\u0026plusmn;\u0026thinsp;0.74) group received the lowest score. The ROSEM and DPR groups exhibited higher lesion depiction scores in comparison to the OSEM groups, with the exception of the DPR1 group (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01 and p\u0026thinsp;=\u0026thinsp;0.52, respectively). Moreover, the lesion depiction scores of the DPR2 to DPR5 groups did not differ significantly from the ROSEM group (p\u0026thinsp;=\u0026thinsp;0.08\u0026ndash;0.87), while the lesion depiction score of the DPR1 group was inferior to that of the ROSEM group (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The overall image quality exhibited an initial improvement, but subsequently demonstrated a decline with an increase in the filter strength factor (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ec). The highest overall image quality score was achieved by the DPR3 group (4.04\u0026thinsp;\u0026plusmn;\u0026thinsp;0.19), while the lowest score was received by the DPR1 group (2.73\u0026thinsp;\u0026plusmn;\u0026thinsp;0.44). The overall image quality scores for the DPR2, DPR3, and DPR4 groups were found to be superior to those of the OSEM group (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Although the DPR1 group exhibited a lower overall image quality score than the OSEM group, this difference was not found to be statistically significant (p\u0026thinsp;=\u0026thinsp;0.21). It is noteworthy that the DPR5 group exhibited a marginally higher overall quality score than the OSEM group (p\u0026thinsp;=\u0026thinsp;0.82), indicating that radiologists may be amenable to the DPR5 group. Moreover, the overall quality scores of the DPR2 and DPR3 groups were not statistically different from the ROSEM group (p\u0026thinsp;=\u0026thinsp;0.56 and p\u0026thinsp;=\u0026thinsp;0.85, respectively). In contrast, the overall quality scores of the DPR1, DPR4, and DPR5 groups were significantly higher than that of the ROSEM group (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe objective of this study was to evaluate the performance of the DPR algorithm with regard to depiction and quantification accuracy for small lesions in oncological \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT imaging. Moreover, the optimal filter strength factor of the DPR algorithm was determined using both phantom and patient data. The results of the phantom study indicated that the DPR2 and DPR3 reconstructions yielded the highest hot sphere CNR, in comparison to the OSEM and ROSEM reconstructions. The patient study demonstrated that the DPR2 reconstruction exhibited lower image noise than the OSEM, the ROSEM, and DPR3 reconstructions. Moreover, the DPR3 reconstruction exhibited comparable SUV\u003csub\u003emax\u003c/sub\u003e, CNR, and TBR for small lesions to the ROSEM reconstruction. However, the lesion volume derived from the ROSEM reconstruction was found to be more accurate than those of the DPR2 and DPR3 reconstructions, with CT measurements serving as the standard reference. In conclusion, it can be posited that the DPR reconstruction with a filter strength factor between 2 and 3 may provide high-quality images that may be beneficial for the detection of small lesions and the increase in quantification accuracy.\u003c/p\u003e \u003cp\u003eIn the domain of oncology, the capacity to identify and characterize minute, low-intensity/uptake lesions is of paramount importance for the early diagnosis and staging of patients. Nevertheless, it remains a significant challenge to achieve accurate quantification of radiotracer uptake in the small lesions, primarily due to the overwhelming influence of noise in low signal-to-noise ratio (SNR) images. In order to address this challenge, Rep S et al. [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] proposed a small-voxel OSEM reconstruction (2 mm in-line pixel size) and evaluated the PET image quality using phantoms with a low target-to-background ratio. The findings indicated that the small-voxel reconstruction method can reliably facilitate the precise delineation of small lesions, enhance lesion contrast, and improve image quality. However, the study did not validate its findings in patient data. Furthermore, the image noise increased as a result of the lower counts per voxel, which may necessitate a longer acquisition time to compensate for the lower counts\u0026rsquo; statistics. Another potential avenue for improvement is the application of more sophisticated reconstruction algorithms, including those based on BPL and deep learning techniques. The ROSEM reconstruction, a novel BPL method, has been demonstrated to enhance spatial resolution and lesion contrast for small lesions, thereby facilitating more accurate quantification than the OSEM reconstruction. Nevertheless, the efficacy of small lesion detection is contingent upon the penalization factor. Prior research has indicated that the ROSEM reconstruction with a penalization factor of 0.8\u0026ndash;0.9 improves the depiction and quantification accuracy of small lung lesions in oncological \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT in comparison to the OSEM reconstruction [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. These findings are consistent with those observed for other tracers, including \u003csup\u003e68\u003c/sup\u003eGa-PSMA-11 [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] and \u003csup\u003e68\u003c/sup\u003eGa-DOTATATE [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], although the penalty factor would be regularized as 0.1\u0026ndash;0.2. The findings of this study align with those of previous investigations, indicating that the ROSEM algorithm can elevate the SUV\u003csub\u003emax\u003c/sub\u003e of lesions by 85% in sub-centimeter lesions and 51% in medium-sized lesions, respectively.\u003c/p\u003e \u003cp\u003eRecent studies have indicated that deep learning-based image denoising methods have the potential to enhance image quality although this may also lead to a reduction in CNR for small lesions [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. The direct learning of high-SNR images from low-SNR images is inherently unstable, resulting in suboptimal outcomes. The DPR algorithm can address the challenges posed by this issue through a progressive learning approach. In this approach, a denoising network (CNN-DE) is immediately followed by the initial expectation maximization iterations, resulting in an image with minimal noise. In the second expectation maximization iteration, the contrast of small lesions is gradually recovered and then enhanced through an enhancement network (CNN-EH). The DPR algorithm has the potential to enhance the detectability and the accuracy of quantification for small pulmonary nodules. Nevertheless, the efficacy of this approach may be contingent on the filter strength factor, lesion size, and lesion contrast or TBR. The quantitative analysis of the IEC body phantom study revealed that the CNR increased by a factor of 2.8 for DPR2 and 2.3 for DPR3 in comparison to the OSEM method when imaging on the 10-mm hot sphere. However, when the diameter of the hot sphere was increased to 13 mm, the gains decreased to 2.44 and 2.01, respectively, for the same filter strength factor. These results were further confirmed in the patient study and are consistent with the findings of previous studies [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. The clinical data indicated that the DPR2 and DPR3 groups increased by over 41% and 48%, respectively, for medium-centimeter lesions SUV\u003csub\u003emax\u003c/sub\u003e in comparison to the OSEM group. In the case of sub-centimeter lesions, the SUV\u003csub\u003emax\u003c/sub\u003e increase was observed to be up to 61% and 68%, respectively, which represented a greater gain than that observed in a previous study [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. The aforementioned study evaluated the performance of the DPR and OSEM algorithms in quantifying SUV in 63 sub-centimeter lesions with a mean diameter of (0.76\u0026thinsp;\u0026plusmn;\u0026thinsp;0.15) cm. The results demonstrated that the average SUV\u003csub\u003emax\u003c/sub\u003e (11.46) of DPR was approximately 30% higher than that of OSEM (8.90). However, the filter strengthening factor was not provided. Furthermore, the study indicated that the DPR2 algorithm can achieve the greatest gain in CNR in both sub-centimeter and medium-centimeter lesions compared to the OSEM algorithm. In contrast, the DPR3 algorithm demonstrated superior TBR in comparison to the DPR2 algorithm, indicating that the DPR3 algorithm is more effective in enhancing the lesion contrast. Furthermore, the results indicated that the DPR3 reconstruction could provide a marginally more precise estimation of lesion volume than the DPR2 reconstruction when a CT-derived volume was employed as the standard reference. However, all the DPR groups demonstrated an overestimation of the volume of small lesions. To the best of our knowledge, no previous studies have been conducted on the impact of the SUV\u003csub\u003emax\u003c/sub\u003e threshold on lesion quantification accuracy using the DPR algorithm. The present study suggests that a threshold of greater than 41% for SUV\u003csub\u003emax\u003c/sub\u003e may enhance the accuracy of lesion segmentation.\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e illustrate PET images of a patient diagnosed with thyroid cancer and lung metastasis, obtained through the application of diverse reconstruction techniques. The images demonstrate the presence of numerous lesions in the maximum intensity projection (MIP) images of the patient. While the small lesions appear indistinct and blurred in the images reconstructed by the OSEM and the DPR1 algorithms, they become more pronounced in the images reconstructed by the DPR2 to DPR5 algorithms. This indicates that an inappropriate selection of the filter strength factor may result in an over-or under-estimation of noise, which could lead to images that are either oversmoothed or under-smoothed (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). An increase in the filter strength factor may result in the generation of enhanced images; however, it may also lead to the introduction of noise (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). The selection of an optimal filter strength factor is often a challenging process, influenced by a number of factors, including the specific radiotracer used, the preference of the radiologists, the size of the patient, and the assessment of image quality. Accordingly, the filter strength factor is frequently specified as a range, in accordance with the recommendations of the equipment manufacturer. In clinical practice, it is recommended that the filter strength factor be fixed in order to maintain consistency in SUV measurements. In the visual analysis, the DPR2 reconstruction demonstrated the highest noise score, while the DPR3 reconstruction exhibited the highest lesion depiction score and the highest overall image quality score. As radiologists are primarily concerned with diagnosis, their preference may be influenced by images with a lower background noise level or higher lesion contrast. These images may facilitate the diagnosis of small and low-contrast lesions. The study concluded that, based on visual inspection and quantitative measurements, the optimal choice is DPR reconstruction with a filter strength factor of 3, as it provides superior contrast and lower noise for small lesions.\u003c/p\u003e \u003cp\u003eA reliable and precise measurement of radiotracer uptake is of paramount importance for differential diagnosis, treatment planning, and therapy response evaluation [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Nevertheless, the application of advanced reconstruction techniques, such as the ROSEM and DPR algorithms, has the potential to elevate the SUV of minor lesions and enhance contrast recovery. Such findings may consequently influence the criteria for image interpretation and the evaluation of therapy in subsequent studies. Consequently, it is necessary to update the quantitative interpretation criteria in line with the latest developments in PET technology in the clinical practice. To address these issues, the European Association of Nuclear Medicine (EANM) has recommended that at least two images should be reconstructed for a routine PET examination, with one image displaying the optimal image quality and the other adhering to the EANM Research Ltd. (EARL) specification for quantitative measurement [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Teoh EJ et al. [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] put forth the proposition that elevated SUV\u003csub\u003emax\u003c/sub\u003e thresholds may be justified when employing semi-quantitative analyses for the purpose of diagnosing malignancy. Wu Z. et al. [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] measured the recovery coefficient (RC) of each phantom sphere with a diameter of 10\u0026ndash;37 mm, subsequently establishing an association between RC values and the sphere diameter. This can be employed in the partial-volume-effect (PVE) correction to enhance the accuracy of SUV measurements of pulmonary nodules. The results of the phantom and the patient studies demonstrated that the DPR reconstruction method enhanced the accuracy of quantification in relation to the true uptake and the test-retest reliability through the PVE correction. This could facilitate the evaluation of treatment response in follow-up studies, particularly in the case of small lesions. Furthermore, it was observed that DPR1 with the highest smooth strength could achieve comparable and agreeable CR and lesion SUV\u003csub\u003emax\u003c/sub\u003e with the OSEM reconstruction. This could serve as an alternative solution to address the concern regarding the increasing SUV. It is thus recommended that a harmonized DPR protocol should be implemented in clinical practice in order to ensure consistent results across multi-center trials.\u003c/p\u003e \u003cp\u003eIt is essential to recognize that this study is constrained by a number of limitations. Firstly, it should be noted that this study was conducted at a single center with a limited number of enrolled patients. A large-scale multicenter study is expected to be anticipated, particularly with regard to the quantification accuracy of SUVs, which is of paramount importance in multicenter or cross-machine PET studies. Secondly, further investigation is required to elucidate the relationship between SUV measurement under the DPR reconstruction and pathological results requires. Such information could potentially facilitate the early differential diagnosis of lung cancer. Thirdly, this study focused exclusively on the comparison of depiction and quantification accuracy in the small lesions among the PET images reconstructed by the OSEM, ROSEM, and DPR algorithms. Further investigation is necessary to determine the potential benefits of the DPR algorithm in the detection of large lesions, the improvement of image quality for obese patients, and the reduction of acquisition time. This is a topic that our research team plans to examine in greater depth. Ultimately, the DPR method has only been trained and applied with \u003csup\u003e18\u003c/sup\u003eF-FDG PET image reconstruction. Consequently, in order for the DPR algorithm to be applied to other tracers, such as prostate-specific membrane antigen (PSMA) and fibroblast activation protein inhibitor (FAPI), it is necessary to either retrain the network using a new dataset or retune it via transfer learning.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThe study demonstrated that the DPR method exhibited superior performance in comparison to the OSEM method in terms of small lesion contrast, image noise suppression, contrast recovery, and volumetric quantification accuracy. The application of a filter strength factor of 3 to the DPR method yielded the optimal contrast-to-noise ratio, thereby demonstrating the highest detectability of small lesions. The findings indicated that DPR3 yielded results comparable to those of the ROSEM method in terms of enhancing the quality of PET images.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCNN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003econvolutional neural network\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDPR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003edeep progressive reconstruction\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePET/CT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003epositron emission tomography/computed tomography\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOSEM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eordered subset expectation maximization\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eBPL\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBayesian penalized likelihood\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROSEM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eregularized OSEM\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNEMA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNational Electrical Manufacturers Association\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003econtrast recovery\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eBV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ebackground variability\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCNR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003econtrast-to-noise ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRCR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eradioactivity concentration ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eregion of interest\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eVOI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003evolume of interest\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003estandard deviation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSUV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003estandard uptake value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCOV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ecoefficient of variation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eAvailability for data and materials\u003c/h2\u003e \u003cp\u003eThe datasets used and analyzed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eEthical approval and consent to participate\u003c/strong\u003e \u003cp\u003e All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study was approved by the institutional ethical review board of Nanjing First Hospital, Nanjing Medical University (KY20171208-02), and the written informed consent was waived due to the retrospective nature of this study.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003e Patients signed informed consent regarding the publication of their data and photographs.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eCompeting interests\u003c/strong\u003e \u003cp\u003eThe authors declare that there are no conflicts of interest regarding the publication of this paper.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis research was supported by funding from a Jiangsu Provincial Frontier Grant (BE2017612), a Nanjing Municipal Health Science and Technology Development Fund (ZKX22036), a Nanjing Municipal Health Science and Technology Development Fund (YKK20104), and Jiangsu Provincial Medical Key Discipline Cultivation Unit (JSDW202247).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eLX wrote the manuscript and performed the data analysis. LX, RY, QL-M, and RS-L acquired the data. RC-L and FW carried out the image interpretation. LX, FW, and QL-M conceived and designed the study. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eWe thank all members of the research group.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets used and analyzed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eTahari AK, Wahl RL. Quantitative FDG PET/CT in the community: experience from. interpretation of outside oncologic PET/CT exams in referred cancer patients. \u003cem\u003eJ Med Imaging Radiat Oncol\u003c/em\u003e. 2014;58(2):183-188.\u003c/li\u003e\n\u003cli\u003eAnand SS, Singh H, Dash AK. Clinical applications of PET and PET-CT. Med J Armed Forces. India. 2009;65(4):353\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eFletcher JW, Djulbegovic B, Soares HP, et al. Recommendations on the use of 18F-FDG PET in oncology. J Nucl Med. 2008;49:480\u0026ndash;508.\u003c/li\u003e\n\u003cli\u003eAdams MC, Turkington TG, Wilson JM, et al. A systematic review of the factors affecting. accuracy of SUV measurements. \u003cem\u003eAJR Am J Roentgenol\u003c/em\u003e. 2010;195(2):310-320. \u003c/li\u003e\n\u003cli\u003eTong S, Alessio AM, Kinahan PE. Image reconstruction for PET/CT scanners: past achievements. and future challenges. \u003cem\u003eImaging Med\u003c/em\u003e. 2010;2(5):529-545. \u003c/li\u003e\n\u003cli\u003eXu L, Cui C, Li R, et al. Phantom and clinical evaluation of the effect of a new Bayesian penalized likelihood reconstruction algorithm (HYPER Iterative) on \u003csup\u003e68\u003c/sup\u003eGa-DOTA-NOC PET/CT image quality. \u003cem\u003eEJNMMI Res\u003c/em\u003e. 2022;12(1):73.\u003c/li\u003e\n\u003cli\u003eXing Y, Qiao W, Wang T, et al. Deep learning-assisted PET imaging achieves fast scan/low-dose examination. \u003cem\u003eEJNMMI Phys\u003c/em\u003e. 2022;9(1):7. \u003c/li\u003e\n\u003cli\u003eMehranian A, Wollenweber SD, Walker MD, et al. Image enhancement of whole-body oncology [\u003csup\u003e18\u003c/sup\u003eF]-FDG PET scans using deep neural networks to reduce noise. \u003cem\u003eEur J Nucl Med Mol Imaging\u003c/em\u003e. 2022;49(2):539-549. \u003c/li\u003e\n\u003cli\u003eLv Y, Xi C. PET image reconstruction with deep progressive learning. \u003cem\u003ePhys Med Biol\u003c/em\u003e. 2021;66(10):10.\u003c/li\u003e\n\u003cli\u003eWang T, Qiao W, Wang Y, et al. Deep progressive learning achieves whole-body low-dose \u003csup\u003e18\u003c/sup\u003eF-FDG PET imaging. \u003cem\u003eEJNMMI Phys\u003c/em\u003e. 2022;9(1):82. \u003c/li\u003e\n\u003cli\u003eLi J, Xi C, Dai H, et al. Enhanced PET imaging using progressive conditional deep image prior. \u003cem\u003ePhys Med Biol\u003c/em\u003e. 2023;68(17):10.1088/1361-6560/acf091. Published 2023 Sep 1. \u003c/li\u003e\n\u003cli\u003eBadawi RD, Shi H, Hu P, et al. First Human Imaging Studies with the EXPLORER Total-Body. PET Scanner. \u003cem\u003eJ Nucl Med\u003c/em\u003e. 2019;60(3):299-303. \u003c/li\u003e\n\u003cli\u003eLi Z, Yang J, Liu Z, et al. Feedback network for image super-resolution[C]//Proceedings of the. IEEE/CVF conference on computer vision and pattern recognition. 2019: 3867-3876.\u003c/li\u003e\n\u003cli\u003eXu L, Li RS, Wu RZ, et al. Small lesion depiction and quantification accuracy of. oncological \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT with small voxel and Bayesian penalized likelihood reconstruction. \u003cem\u003eEJNMMI Phys\u003c/em\u003e. 2022;9(1):23. \u003c/li\u003e\n\u003cli\u003eRep S, Tomse P, Jensterle L, et al. Image reconstruction using small-voxel size improves small. lesion detection for positron emission tomography. \u003cem\u003eRadiol Oncol\u003c/em\u003e. 2022;56(2):142-149. \u003c/li\u003e\n\u003cli\u003eYang FJ, Ai SY, Wu R, et al. Impact of total variation regularized expectation maximization. reconstruction on the image quality of \u003csup\u003e68\u003c/sup\u003eGa-PSMA PET: a phantom and patient study. \u003cem\u003eBr J Radiol\u003c/em\u003e. 2021;94(1120):20201356.\u003c/li\u003e\n\u003cli\u003eLiu L, Liu H, Xu S, et al. The impact of total variation regularized expectation maximization. reconstruction on68 Ga-DOTA-TATE PET/CT images in patients with neuroendocrine tumor. Front Med (Lausanne). 2022;9: 845806.\u003c/li\u003e\n\u003cli\u003eSchaefferkoetter J, Yan J, Ortega C, et al. Convolutional neural networks for improving image quality with noisy PET data. \u003cem\u003eEJNMMI Res\u003c/em\u003e. 2020;10(1):105.\u003c/li\u003e\n\u003cli\u003eSanaat A, Shiri I, Arabi H, et al. Deep learning-assisted ultra-fast/low-dose whole-body. PET/CT imaging. Eur J Nucl Med Mol Imaging. 2021;48(8):2405\u0026ndash;15.\u003c/li\u003e\n\u003cli\u003eYang H, Chen S, Qi M, et al. Investigation of PET image quality with acquisition time/bed and. enhancement of lesion quantification accuracy through deep progressive learning. \u003cem\u003eEJNMMI Phys\u003c/em\u003e. 2024;11(1):7.\u003c/li\u003e\n\u003cli\u003eBarrington SF, Sulkin T, Forbes A, et al. All that glitters is not gold\u0026mdash;new reconstruction. methods using Deauville criteria for patient reporting. Eur J Nucl Med Mol Imaging. 2018;45(2):316\u0026ndash;7.\u003c/li\u003e\n\u003cli\u003eBoellaard R, Sera T, Kaalep T, et al. Updating PET/CT performance standards and PET/CT. interpretation criteria should go hand in hand. EJNMMI Res. 2019;9(1):95.\u003c/li\u003e\n\u003cli\u003eBoellaard R, Delgado-Bolton R, Oyen WJ, et al. FDG PET/CT: EANM procedure guidelines for tumour imaging: version 2.0. \u003cem\u003eEur J Nucl Med Mol Imaging\u003c/em\u003e. 2015;42(2):328-354.\u003c/li\u003e\n\u003cli\u003eTeoh EJ, McGowan DR, Bradley KM, et al. Novel penalised likelihood reconstruction of PET. in the assessment of histologically verified small pulmonary nodules. \u003cem\u003eEur Radiol\u003c/em\u003e. 2016;26(2):576-584. \u003c/li\u003e\n\u003cli\u003eWu Z, Guo B, Huang B, et al. Phantom and clinical assessment of small pulmonary nodules. using Q.Clear reconstruction on a silicon-photomultiplier-based time-of-flight PET/CT system. \u003cem\u003eSci Rep\u003c/em\u003e. 2021;11(1):10328. \u003c/li\u003e\n\u003c/ol\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":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"PET/CT, 18F-FDG, Deep progressive learning, Small lesions, Image quality","lastPublishedDoi":"10.21203/rs.3.rs-6366594/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6366594/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eTo investigate the impact of deep progressive learning reconstruction (DPR) on small lesion detection and image quality compared to ordered subset expectation maximum (OSEM) and regularized OSEM (ROSEM) in \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT imaging.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThe NEMA phantom was filled with \u003csup\u003e18\u003c/sup\u003eF-FDG solution, with a hot sphere-to-background ratio of 4:1. Twenty-six patients with \u003csup\u003e18\u003c/sup\u003eF-FDG-avid lung lesions (diameter\u0026thinsp;\u0026lt;\u0026thinsp;2.0 cm) were enrolled in the study. The PET images were reconstructed by seven groups: routine OSEM, ROSEM with a penalization factor of 0.8 (ROSEM), and DPR reconstructions with five different filter strength factors ranging from smooth to sharp:1\u0026ndash;5 (DPR1, DPR2, DPR3, DPR4, and DPR5). The contrast recovery (CR), background variability (BV), contrast-to-noise ratio (CNR), and radioactivity concentration ratio (RCR) were measured in the phantom study. The maximum standardized uptake values (SUV\u003csub\u003emax\u003c/sub\u003e), target-to-background ratio (TBR), CNR, the volume of the lesions, and the coefficient of variation (COV) of the liver were calculated and compared between these methods in the patient study. Two radiologists evaluated the image quality using a five-point Likert scale.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eIn the phantom study, the DPR2 to DPR5 and ROSEM groups achieved higher CR, CNR, RCR, and lower BV than the OSEM group. For the smallest 10-mm hot sphere, the DPR3 achieved a CNR of 24.46, while the ROSEM and OSEM groups achieved the corresponding values of 22.25 and 10.39, respectively. In the patient study, the DPR1 to DPR4 groups exhibited significantly lower liver COVs than the OSEM group (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Nevertheless, no significant difference was observed between the DPR3 and ROSEM groups (p\u0026thinsp;=\u0026thinsp;0.65). The lesion SUVs, TBRs, and CNRs of the DPR2 to DPR4 groups were found to be significantly higher than those of the OSEM group (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). However, the lesion SUVs and TBRs of the DPR4 and DPR5 groups were found to be equivalent to those of the ROSEM group (SUVs: p\u0026thinsp;=\u0026thinsp;0.19\u0026ndash;0.61, and TBRs: p\u0026thinsp;=\u0026thinsp;0.26\u0026ndash;0.70). Moreover, the CNRs of the DPR3 and ROSEM groups were found to be comparable (p\u0026thinsp;=\u0026thinsp;0.75). The volume of the lesion obtained from the ROSEM group was statistically equivalent to that measured on CT images (p\u0026thinsp;=\u0026thinsp;0.48), but those were overestimated by all DPR groups (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). The overall image quality scores for DPR2, DPR3, and DPR4 were found to be superior to those obtained with OSEM (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), while those for DPR2 and DPR3 groups were not statistically different from the ROSEM group (p\u0026thinsp;=\u0026thinsp;0.56, and p\u0026thinsp;=\u0026thinsp;0.85).\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe DPR method demonstrated a significant improvement in lesion contrast, TBR, and volumetric quantification accuracy for small lesions compared to the OSEM method in the \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT imaging. The DPR method, with a filter strength factor of 3, demonstrated comparable enhancement of PET image quality to the ROSEM method.\u003c/p\u003e","manuscriptTitle":"Investigate the quantification accuracy of small lesions in oncological 18F-FDG PET/CT using a deep progressive learning reconstruction method","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-13 09:54:28","doi":"10.21203/rs.3.rs-6366594/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-10-27T13:12:52+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-23T08:37:10+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"16125669890555996921253779392383495285","date":"2025-09-30T08:12:57+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-06T10:01:18+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-02T13:33:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"240318208955387784226006861266867243767","date":"2025-05-29T00:37:52+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"88992278663209737787933666012259464066","date":"2025-05-09T07:50:28+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-05-07T06:22:37+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-06T05:51:55+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-04T14:29:34+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2025-05-04T14:28:29+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"e08cad5e-2273-4b5c-875d-d8153a3fccbe","owner":[],"postedDate":"May 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-01-19T16:46:55+00:00","versionOfRecord":{"articleIdentity":"rs-6366594","link":"https://doi.org/10.1186/s12880-026-02166-w","journal":{"identity":"bmc-medical-imaging","isVorOnly":false,"title":"BMC Medical Imaging"},"publishedOn":"2026-01-17 16:29:21","publishedOnDateReadable":"January 17th, 2026"},"versionCreatedAt":"2025-05-13 09:54:28","video":"","vorDoi":"10.1186/s12880-026-02166-w","vorDoiUrl":"https://doi.org/10.1186/s12880-026-02166-w","workflowStages":[]},"version":"v1","identity":"rs-6366594","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6366594","identity":"rs-6366594","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}