Application Value of Radiomics Features Based on PSMA PET/CT in Diagnosis of Clinically Significant Prostate Cancer: A Comparative Analysis of the PRIMARY and PI-RADS Scores

Application Value of Radiomics Features Based on PSMA PET/CT in Diagnosis of Clinically Significant Prostate Cancer: A Comparative Analysis of the PRIMARY and PI-RADS Scores | 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 Application Value of Radiomics Features Based on PSMA PET/CT in Diagnosis of Clinically Significant Prostate Cancer: A Comparative Analysis of the PRIMARY and PI-RADS Scores Yujun Geng, Ming Zhang, Shumei Li, Jingwen Li, Xinzhi Wang, Junzhang Tian, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4229122/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Objectives: The aim of our study was to explore the role of radiomic features derived from positron emission tomography (PSMA-PET)/computed tomography (CT) images in diagnosis of clinically significant prostate cancer (csPCa). Additionally, we aimed to investigate correlations between these features and other PET/CT parameters. Finally, we compared these radiomic features with the PRIMARY and PI-RADS scores to assess their complementarity and enhance the diagnostic capabilities for prostate cancer. Methods : A total of 110 patients with a certain pathological diagnosis were included, and a total of 1155 sets of radiomic features were extracted from these images for each patient. We employed the LASSO regression algorithm (Lasso) to select these features and collect MRI interpretation results (PI-RADS v2.1) via consensus reading for each patient. Two clinical physicians with more than three years of experience in nuclear medicine scored the medical images (PRIMARY) for all patients, and a consensus was reached. Finally, we compared diagnostic capabilities between radiomic features and indices/scores based on medical imaging (magnetic resonance (MRI) and positron emission tomography (PET/CT)). Results: After the Lasso algorithm was applied, three sets of radiomic features, log-sigma-1-mm-3D_glcm_Correlation, log-sigma-3-mm-3D_firstorder_Minimum, and wavelet-LLH_glcm_Imc2, marked as RF1, RF2, and RF3, respectively, were included in the analysis. The area under the curve (AUC) for diagnosing csPCa was 0.8744 (95% CI=0.806-0.943), 0.8413 (95% CI=0.762-0.920), and 0.8602 (95% CI=0.625-0.841), with accuracies of 0.8364, 0.8273, and 0.8273, respectively. The kappa values with the maximum standard uptake value (SUVmax) were 0.757, 0.779, and 0.737; Gleason scores were 0.696, 0.688, and 0.668 of three radiomics features, respectively. The combined analysis of radiomic features(RF1) and MRI results yielded an accuracy of 0.8727, a recall of 0.8364, and a specificity of 0.9091. Conclusion: Radiomic features based on PSMA PET/CT images correlate strongly with the SUVmax and pathological diagnostic results (Gleason score) of prostate cancer patients. These methods can compensate for the insufficient specificity of MRI for prostate cancer diagnosis but not for PET/CT. Prostate cancer Radiomics Magnetic resonance imaging (MRI) Positron emission tomography (PET) Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction Prostate cancer is the second most common disease that threatens the health of men worldwide and leads to approximately 1.9 million new disease cases and more than 800 thousand deaths each year[ 1 ]. Therefore, early diagnosis and treatment of prostate cancer are crucial. Recently, with the rapid development of medical imaging technology and biochemical chemistry, positron emission tomography (PET) technology based on labelled prostate specific membrane antigen (PSMA) has played an increasingly important role in diagnosis of prostate cancer, including lesion risk classification[ 2 ] [ 3 ], pathological score prediction[ 4 ] and postoperative recurrence and metastasis prediction[ 5 ] [ 6 ]. Numerous studies have indicated that PET has excellent diagnostic and predictive abilities[ 7 ], especially when the lesion is located in the central and transitional zones of the prostate[ 8 ], and its diagnostic ability is much better than that of multiparameter magnetic resonance imaging (mp-MRI) [ 9 ]. PSMA is a type II membrane glycoprotein that is highly expressed in prostate cancer cells and has become one of the most widely studied molecular probes in the field of nuclear medicine[ 10 ]. However, as in the current situation, such an examination still has several shortcomings. First, if the size of the lesion focus is small, it is not noticed[ 11 ] [ 12 ]. Second, standard uptake values (SUVs) obtained from medical images require manual delineation, leading to potential subjective bias due to observer variability[ 13 ] [ 14 ]. However, the current calculation process for radiomic features based on nuclear medicine images is relatively objective[ 15 ], with high repeatability and relatively low demands on the expertise of clinical physicians. Therefore, the objective of our current study was to identify crucial radiomic features for diagnosing clinically significant prostate cancer (csPCa). We aimed to compare these features with those of MRI (PI-RADS version 2.1) [ 16 ] and PET (PRIMARY) [ 17 ] data to obtain additional valuable information from PSMA-PET/CT imaging data. In addition, we conducted a comparative analysis of the values of radiomic features based on PET/CT images and other relevant examination parameters, such as total lesions of PSMA (TLP) and metabolic tumour volume (MTV) [ 7 ]. This study aimed to further analyse the origins of radiomic feature values to facilitate future research and discoveries[ 15 ]. 2. Materials and methods This retrospective study was approved by the Institutional Review Board of the Affiliated Guangdong Second Provincial General Hospital of Jinan University. Written informed consent was obtained from all the participants. Study population A total of 255 patients' 18 F-PSMA-PET/CT medical images form department of nuclear medicine of Meizhou People’s Hospital were included in this study (Fig. 1 ). The cases were all confirmed by pathological biopsy to be csPCa or non-caPCa. Our inclusion criteria were as follows. (1) The quality of the image was better with a clear pathological diagnosis, and the image was confirmed many times during the subsequent follow-up. A related Gleason score was assigned for prostate cancer according to the pathological results. (2) The time interval between PET scanning and pathological puncture biopsy or MRI examination was less than 2 months. The exclusion criteria for patients were as follows: (1) before PET scanning, the patient underwent partial prostate resection surgery or had some other drug treatment; (2) severe disease in other organs accompanied by metastasis in other regions; (3) poor image quality; and (4) lacked complete clinical information. Finally, a total of 110 (mean age: 70.3 ± 8.05 years, range from 48 to 89 years) patients with qualified data were included in our study; 55 patients had clinically significant lesions (Gleason score > = 7), and 55 patients had non clinically significant lesions (Gleason score < 7). We also evaluated several clinical parameters, including age and prostate-specific antigen (PSA) level. PET image acquisition and reconstruction All PET/CT images were acquired with a 120 KV positron emission tomography (PET/CT) scanner (Siemens Healthcare, Biograph20) at 100 mA from the head to the thigh. After intravenous injection of 18 F-PSMA (2.5 MBq/kg body mass) for 90 minutes, PET/CT imaging was performed, ranging from the skull base to the thigh. Spiral CT (tube voltage: 120 kV, tube current: 30–210 mA, pitch: 0.56:1, layer thickness and reconstruction layer thickness: 3.75 mm) was performed first, followed by PET (7–9 beds, each bed for 2.5 minutes, matrix 256×256, with a layer thickness and interval of 3.27 mm). Attenuation correction and iterative reconstruction of PET images using CT data, with 2 iterations and a subset of 28; display maximum intensity projection (MIP) images, PET images, CT images, and PET/CT fusion images based on our PACS workstation system. Based on the obtained PET/CT images, we manually delineated the regions of interest at the focal concentration and measured SUVmax, SUVmean, and MTV (TLP = SUVmean*MTV). According to previous studies, the focal concentration is expected to be 3.5 times greater than that in the surrounding areas. Two physicians with more than three years of experience in nuclear medicine were tasked with assessing the images and providing relevant PRIMARY scores[ 16 ]. In cases of discrepancies, a consensus was reached through discussion. The 5-point PRIMARY score was calculated as follows: 1, no significant concentration within the prostate; 2, diffuse transition or central zone concentration; 3, focal transition zone activity above twice the background transition zone; 4, focal peripheral zone activity of any intensity; and 5, an SUV greater than 12 (Fig. 2 ). Region of interest (ROI) segmentation and feature extraction We imported all images into open-source 3D-slicer ( https://www.slicer.org ) software[ 18 ] and reconstructed them in three dimensions for better observation. The two attending physicians in our department were requested to mask the whole prostate as an ROI via the software[ 19 ]. In the process of drawing, they carried out discussions each other. When their opinions were inconsistent, the final decision was made by the supervising physician. Based on open-source PyRadiomics[ 20 ], a total of 1155 radiomic features, including 28 morphological features, 20 original first-order features, 24 grey-level cooccurrence matrix (GLCM) features, 14 grey-level dependence matrix (GLDM) features, 16 grey-level run-length matrix (GLRLM) features, 16 grey-level size-zone matrix (GLSZM) features, 5 neighbouring grey-level dependence matrix (NGLDM) features and 1032 features, were calculated by the wavelet and Laplacian of Gaussian (LoG) method. Intraclass correlation coefficients (ICCs) were calculated between the two physicians, and features with lower ICCs (less than 0.75) were eliminated due to their instability. The least absolute shrinkage and selection operator (LASSO) algorithm was used to select features (the weights of irrelevant features were reduced to 0) (Fig. 3 ). Statistical analysis All statistical calculations were performed with SPSS software and Python 3.8 (version 1.11.1). The receiver operating characteristic (ROC) curve and box plot were generated with GraphPad Prism software. The Mann‒Whitney U test was used to compare differences in continuous variables, and the chi-square test was used for categorical variables. The Cohen correlation coefficient was used to calculate correlations: 0-0.2 indicated slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81-1.0, perfect agreement. For interpretation of PI-RADS results, we employed two methods. Method 1 considered scores 1–3 benign and 4–5 malignant; Method 2 considered scores 1–2 benign and 3–5 malignant. Regarding the PRIMARY score, cases were classified as benign if 1–2 and malignant if 3–5. The area under the curve (AUC), accuracy (ACC), sensitivity (SEN), specificity (SPE), 95% confidence interval (95% CI), and F1 score were calculated to compare each predictor. Finally, in the combined analysis of radiomic features and PI-RADS scores, a PI-RADS score of 5 was considered malignant; for other PI-RADS scores ranging from 1–4, we relied on radiomics-based judgement. The same approach was used for the PRIMARY score; a score of 4 or 5 points was considered malignant. For scores less than 4 points, we adopted the radiomics results. A P value less than 0.05 was considered to indicate statistical significance. Before utilizing RF2, the original values were multiplied by -1, as the AUC value of the original values was 0.1653 (the reverse prediction direction). 3. Results Patient clinical characteristics A total of 110 patients were included in this study; 60 were confirmed to have prostate adenocarcinoma by pathological biopsy, including 5 with a Gleason score of 6 and 55 with a Gleason score > = 7. Among all the patients, 28 had PI-RADS scores of 1–2 (26.3%), 20 had PI-RADS scores of 3 (18.2%), 18 had PI-RADS scores of 4 (28.2%), and 44 (27.3%) had PI-RADS scores of 5. The PRIMARY score results indicated 50 (45%) negative cases and 60 (55%) positive cases. Among all malignant cases, 6 involved concurrent lymph node and bone metastases; 6 had only involved bone metastases ( Table.1 ). Radiomics feature results A total of 292 sets of radiomic features were deleted due to their lower intraclass correlation coefficients (ICCs < 0.70). After LASSO regression and 5-fold cross-validation, a total of 3 groups of radiomic features were selected to establish a diagnostic model (Fig. 3 ). These data are log-sigma-1-mm-3D_glcm_Correlation, log-sigma-3-mm-3D_firstorder_Minimum, and wavelet-LLH_glcm_Imc2; we labelled them RF1, RF2, and RF3, respectively. Before utilizing RF2, the original values were multiplied by -1. The box plot results indicate a significant difference in the distribution of the three sets of radiomic features between the csPCa and noncsPCa groups (Fig. 5 ). The AUC, ACC of the three radiomics features for the csPCa diagnosis was 0.8744 (95% CI:0.806–0.943), 0.8364, 0.8413 (95% CI:0.762–0.920), 0.8273, 0.8602 (95% CI:0.787–0.933), 0.8273, respectively. Their diagnostic capabilities surpassed PSA (AUC:0.7468,95% CI:0.625–0.841), MTV (AUC:0.6134,95% CI:0.506–0.721), and TLP (AUC:0.7848,95% CI:0.695–0.874), but were slightly lower than SUVmax (AUC:0.8691,95% CI:0.796–0.942) (Fig. 4 ) ( Table.2 ). The results of the Cohen correlation kappa indicated a substantial correlation between the three groups of radiomic features and SUVmax (kappa = 0.757, 0.779, 0.737, respectively), TLP (kappa = 0.645, 0.692, 0.602, respectively), and GS (kappa = 0.696, 0.688, 0.668, respectively). However, there was a fair correlation with the MTV (kappa = 0.326, 0.322, 0.247) ( Table.3 ). PI-RADS and PRIMARY scores The ACC of Method 1 for PI-RADS was 0.8545, the recall was 0.9091, and the specificity was 0.8000. The ACC of Method 2 for PI-RADS was 0.7091, the recall was 0.9455, and the specificity was 0.4727. The ACC of PRIMARY was 0.8455, the recall was 0.8909, and the specificity was 0.8000 ( Table.2 ). The results of the Cohen correlation kappa indicated a substantial correlation between PI-RADS (method 1) and the PRIMARY score and between SUVmax (kappa = 0.612, 0.774) and the Gleason score (kappa = 0.713, 0.694). There was a slight correlation with MTV (kappa = 0.188, 0.198) and fair agreement with TLP (kappa = 0.439, 0.491) ( Table.3 ). Combination method The ACC of RF1 + SUVmax was 0.8545, the recall was 0.8980, and the specificity was 0.8197 (Fig. 6 ). The ACC of combination1 (RF1 + PI-RADS) was 0.8727, the recall was 0.8364, and the specificity was 0.9091. The ACC of combination2 (RF1 + PRIMARY) was 0.8455, the recall was 0.7969, and the specificity was 0.9130 ( Table.4 ). 4. Discussion In this study, we primarily analysed the significance of the radiomics features log-sigma-1-mm-3D_glcm_Correlation, log-sigma-3-mm-3D_firstorder_Minimum, and wavelet-LLH_glcm_Imc2 (marked as RF1, RF2, and RF3, respectively) based on PSMA PET/CT for diagnosing clinically significant prostate cancer. The results indicate that the feature log-sigma-1-mm-3D_glcm_Correlation has a higher AUC (0.8744; 95% CI = 0.806–0.943) for diagnosing clinically significant prostate cancer; in addition, the feature log-sigma-3-mm-3D_firstorder_Minimum had a stronger correlation with SUVmax on PSMA-PET/CT (kappa = 0.779, 95% CI = 0.651–0.885). Although the accuracy of the three radiomic features for diagnosing csPCa was not as high as that of SUVmax (AUC: 0.8691; ACC: 0.8455), RF1 showed comparable performance to that of SUVmax, and it exhibited substantial agreement with the Gleason score of the lesion (kappa = 0.696, 95% CI: 0.566–0.816). The extent to which these radiomic features complement SUVmax, whether there are hidden features, and exploration of these aspects were limited in this study due to the limited number of cases. We hope that future researchers can delve further into these discussions, as revealing additional clinical information based on medical imaging will be highly beneficial. In our current study, the cut-off value for SUVmax for diagnosing clinically significant prostate cancer was determined to be 12.6, which closely aligns with the findings of Lv et al[ 7 ]., whose cut-off value was 11.5, with a diagnostic AUC of 0.864. In fact, many studies on PSMA PET/CT for diagnosing prostate cancer heavily rely on SUVmax values. A score of 5 in the PRIMARY is based on whether SUVmax exceeds 12.0[ 17 ]. According to the study by Liu et al. [ 14 ], the SUVmax of a normal prostate gland is approximately 3.40 (range: 2.00-4.40), and a value greater than 3.50 can be considered the focal concentration. Compared to visual identification, radiomics features based on image pixel values demonstrate greater objectivity and repeatability[ 15 ]. However, this does not imply that it is a perfect indicator. First, before extracting features, physicians need to manually delineate the region of interest of the whole prostate gland, introducing observer errors[ 21 ]. Second, its interpretability is not high; further research is needed to understand its biological significance. Therefore, the current importance of deep learning-based medical image segmentation techniques is evident[ 22 ]. On the one hand, this approach can significantly reduce the workload of clinical physicians, and on the other hand, it can assist researchers in quantifying certain biological parameters, such as the tumour volume[ 23 ] and prostate gland area[ 24 ]. At present, an increasing number of research results have shown that PSMA PET/CT has the best prostate diagnostic ability [ 10 ], with performance that is far better than that of mp-MRI [ 9 ], especially when low-signal lesions are located in the central and transitional zones of the prostate [ 8 ]. However, it is worth noting that in our current study, the diagnostic accuracy of MRI was slightly better than that of the PRIMARY score, with a diagnostic accuracy of 85.45%. Nevertheless, this result should be interpreted with caution, as categorizing cases with a PI-RADS score of 3 as either benign or malignant is problematic, given that it inherently represents a 50% probability of being malignant or benign. In our study, a total of 20 patients had a PI-RADS score of 3, and only 2 (10%) of these cases were malignant. Our study results are consistent with the findings of Guo et al.’s team [ 9 ], who reported that approximately 91.5% of cases with a PI-RADS score of 3 were negative. According to the study by the Emmett team [ 17 ], this figure is approximately 76.92% (20/26). Similarly, a PI-RADS score of 5 plays a crucial role in diagnosing prostate cancer [ 25 ] [ 26 ] [ 27 ]. In our study, there were approximately 44 patients with a PI-RADS score of 5, with 93.2% (3/44) of the lesions being malignant. These findings align closely with those of Guo et al. [ 9 ], who reported that approximately 25.05% of patients had a PI-RADS score of 5 but contributed to 41.3% (92/223) of the malignant cases. In our study, after incorporating radiomic features based on PET/CT, the accuracy reached 87.27%, and the specificity reached 92.73%. Approximately 95% of patients with a PI-RADS score of 3 could avoid biopsy, but in patients with a PI-RADS score of 4, this proportion was 61.1%. Therefore, we strongly recommend PET/CT for patients with a PI-RADS score of 3. Radiomics [ 15 ] [ 28 ] and various deep-learning techniques [ 29 ] [ 30 ] [ 31 ] have been widely applied in various clinical studies in recent years, such as detection of prostate cancer lesions [ 32 ], benign-malignant classification of lesions [ 33 ], and prediction of prognosis [ 34 ] [ 35 ] [ 36 ]. For instance, the study by Yi's team [ 12 ] confirmed that radiomics features can effectively identify malignant lesions that appear negative in PET/CT, and scholars such as Yao et al. [ 37 ] have demonstrated that radiomics features can efficiently diagnose lymph node metastasis and extracapsular extension in prostate cancer. In our study, we did not utilize any machine learning algorithm models, primarily due to the limited sample size, with concerns about the risk of overfitting. However, the three sets of radiomics features in this study showed substantial agreement with the pathological results for the lesions (kappa = 0.696, 0.688, 0.668). The two sets of data also demonstrated high consistency with SUVmax (kappa = 0.757, 0.779, 0.737) but were not entirely identical. The main advantage lies in their objectivity, independence of subjective judgement, and straightforward calculation process with extremely high repeatability. There are still some shortcomings in our research. First, the number of patients in this study was relatively small. Our study results call for the participation of more patients and institutions in the future to increase the feasibility and broaden the applicability of the findings. Second, our analysis method was overly simplistic, and we did not explore the hidden variables behind various features (whether they were radiomic features based on PET/CT, MRI, or PSA) [ 38 ] [ 39 ]. Investigating these aspects would be meaningful in future research. 5. Conclusion Radiomic features based on PSMA PET/CT exhibit strong diagnostic capabilities for clinically significant prostate cancer. These parameters showed a substantial correlation with the pathological grade (Gleason score) of prostate cancer and SUVmax based on PET images. This finding suggests the potential of these devices to address the specificity limitations in MRI-based diagnosis to a certain extent but not PET/CT. Abbreviations ACC Accuracy AUC Area under ROC curve CI Confidence interval CT Computed tomography ICC Intraclass correlation coefficient MRI Magnetic resonance imaging MTV Metabolic tumour volume PCa Prostate cancer PET Positron emission tomography PSMA Prostate specific membrane antigen PSA Prostate specific antigen PI-RADS Prostate Imaging Reporting and Data System ROC Receiver operating characteristic ROI Region of interest SEN Sensitivity SPE Specificity SUV Standardized uptake value TLP Total lesions of PSMA Declarations Ethics approval and consent to participate Funding This study was funded by the National Science Foundation of China (Grant number: 82001792) and the Science Foundation of Guangdong Second Provincial General Hospital (Grant Number: 3D-A2021009). Availability of data and materials The authors confirm that the data supporting the findings of this study are available from the corresponding author, upon reasonable request. Author Contribution Yujun Geng Conceived and designed the experiments, Performed the experiments, Writing - original draft. Ming Zhang and Jingwen Li Analyzed the data, Figures; Yujun Geng and Junzhang Tian Contributed reagents/materials/analysis tools. Xinzhi Wang: Performed the experiments. Xiaofen Ma Conceived and designed the experiments, Performed the experiments. All authors reviewed the manuscript. References Siegel, R.L., et al., Cancer statistics, 2023. CA Cancer J Clin, 2023. 73 (1): p. 17-48. Waller, J., R. Flavell, and C.L. Heath, High Accuracy of PSMA PET in Initial Staging of High-Risk Prostate Cancer. Radiol Imaging Cancer, 2020. 2 (4): p. e204025. Basso Dias, A., et al., Impact of (18)F-DCFPyL PET on Staging and Treatment of Unfavorable Intermediate or High-Risk Prostate Cancer. Radiology, 2022. 304 (3): p. 600-608. Emmett, L., et al., Reproducibility and Accuracy of the PRIMARY Score on PSMA PET and of PI-RADS on Multiparametric MRI for Prostate Cancer Diagnosis Within a Real-World Database. J Nucl Med, 2024. 65 (1): p. 94-99. Ulaner, G.A., et al., (18)F-DCFPyL PET/CT for Initially Diagnosed and Biochemically Recurrent Prostate Cancer: Prospective Trial with Pathologic Confirmation. Radiology, 2022. 305 (2): p. 419-428. Murad, V., et al., Association of PSMA PET-derived Parameters and Outcomes of Patients Treated for Oligorecurrent Prostate Cancer. Radiology, 2023. 309 (3): p. e231407. Lv, J., et al., A single-center, multi-factor, retrospective study to improve the diagnostic accuracy of primary prostate cancer using [(68)Ga]Ga-PSMA-11 total-body PET/CT imaging. Eur J Nucl Med Mol Imaging, 2023. Deniffel, D., et al., Avoiding Unnecessary Biopsy: MRI-based Risk Models versus a PI-RADS and PSA Density Strategy for Clinically Significant Prostate Cancer. Radiology, 2021. 300 (2): p. 369-379. Guo, S., et al., The PRIMARY Score: Diagnostic Performance and Added Value Compared With MRI in Detecting Clinically Significant Prostate Cancer. Clin Nucl Med, 2024. 49 (1): p. 37-44. Lawhn-Heath, C., et al., Prostate-specific Membrane Antigen PET in Prostate Cancer. Radiology, 2021. 299 (2): p. 248-260. Zamboglou, C., et al., Uncovering the invisible-prevalence, characteristics, and radiomics feature-based detection of visually undetectable intraprostatic tumor lesions in (68)GaPSMA-11 PET images of patients with primary prostate cancer. Eur J Nucl Med Mol Imaging, 2021. 48 (6): p. 1987-1997. Yi, Z., et al., Machine learning-based prediction of invisible intraprostatic prostate cancer lesions on 68 Ga-PSMA-11 PET/CT in patients with primary prostate cancer. European Journal of Nuclear Medicine and Molecular Imaging, 2021. 49 (5): p. 1523-1534. Benomar, A., et al., Measuring Interrater Reliability. Radiology, 2023. 309 (3): p. e230492. Liu, C., et al., (68)Ga-PSMA PET/CT Combined with PET/Ultrasound-Guided Prostate Biopsy Can Diagnose Clinically Significant Prostate Cancer in Men with Previous Negative Biopsy Results. J Nucl Med, 2020. 61 (9): p. 1314-1319. Gillies, R.J., P.E. Kinahan, and H. Hricak, Radiomics: Images Are More than Pictures, They Are Data. Radiology, 2016. 278 (2): p. 563-77. Turkbey, B., et al., Prostate Imaging Reporting and Data System Version 2.1: 2019 Update of Prostate Imaging Reporting and Data System Version 2. Eur Urol, 2019. 76 (3): p. 340-351. Emmett, L., et al., The PRIMARY Score: Using Intraprostatic (68)Ga-PSMA PET/CT Patterns to Optimize Prostate Cancer Diagnosis. J Nucl Med, 2022. 63 (11): p. 1644-1650. Fedorov, A., et al., 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging, 2012. 30 (9): p. 1323-41. Han, C., et al., Radiomics Models Based on Apparent Diffusion Coefficient Maps for the Prediction of High-Grade Prostate Cancer at Radical Prostatectomy: Comparison With Preoperative Biopsy. J Magn Reson Imaging, 2021. 54 (6): p. 1892-1901. van Griethuysen, J.J.M., et al., Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res, 2017. 77 (21): p. e104-e107. Mackin, D., et al., Measuring Computed Tomography Scanner Variability of Radiomics Features. Invest Radiol, 2015. 50 (11): p. 757-65. Huang, Y., et al., Segment anything model for medical images? Med Image Anal, 2024. 92 : p. 103061. Wang, F., et al., Comparison of Multiple Segmentation Methods for Volumetric Delineation of Primary Prostate Cancer with Prostate-Specific Membrane Antigen-Targeted (18)F-DCFPyL PET/CT. J Nucl Med, 2024. 65 (1): p. 87-93. Jafari, E., et al., A convolutional neural network-based system for fully automatic segmentation of whole-body [(68)Ga]Ga-PSMA PET images in prostate cancer. Eur J Nucl Med Mol Imaging, 2023. Chatterjee, A., et al., Validation of Prostate Tissue Composition by Using Hybrid Multidimensional MRI: Correlation with Histologic Findings. Radiology, 2022. 302 (2): p. 368-377. Tavakoli, A.A., et al., Contribution of Dynamic Contrast-enhanced and Diffusion MRI to PI-RADS for Detecting Clinically Significant Prostate Cancer. Radiology, 2023. 306 (1): p. 186-199. Johnston, E.W., et al., VERDICT MRI for Prostate Cancer: Intracellular Volume Fraction versus Apparent Diffusion Coefficient. Radiology, 2019. 291 (2): p. 391-397. Hesketh, R.L., A.X. Zhu, and R. Oklu, Radiomics and circulating tumor cells: personalized care in hepatocellular carcinoma? Diagn Interv Radiol, 2015. 21 (1): p. 78-84. Li, J., et al., Predicting muscle invasion in bladder cancer based on MRI: A comparison of radiomics, and single-task and multi-task deep learning. Computer Methods and Programs in Biomedicine, 2023. 233 . Hu, Z., et al., Obtaining PET/CT images from non-attenuation corrected PET images in a single PET system using Wasserstein generative adversarial networks. Physics in Medicine & Biology, 2020. 65 (21). Song, F., et al., A Novel Feature Engineering Method Based on Latent Representation Learning for Radiomics: Application in NSCLC Subtype Classification. IEEE Journal of Biomedical and Health Informatics, 2023: p. 1-11. Zhao, Y., et al., Deep Neural Network for Automatic Characterization of Lesions on 68Ga-PSMA PET/CT Images. Annu Int Conf IEEE Eng Med Biol Soc, 2019. 2019 : p. 951-954. Zang, S., et al., Development and validation of (68)Ga-PSMA-11 PET/CT-based radiomics model to detect primary prostate cancer. EJNMMI Res, 2022. 12 (1): p. 63. Seifert, R., et al., Semiautomatically Quantified Tumor Volume Using (68)Ga-PSMA-11 PET as a Biomarker for Survival in Patients with Advanced Prostate Cancer. J Nucl Med, 2020. 61 (12): p. 1786-1792. Ferdinandus, J., et al., Prognostic biomarkers in men with metastatic castration-resistant prostate cancer receiving [177Lu]-PSMA-617. Eur J Nucl Med Mol Imaging, 2020. 47 (10): p. 2322-2327. Gafita, A., et al., Measuring response in metastatic castration-resistant prostate cancer using PSMA PET/CT: comparison of RECIST 1.1, aPCWG3, aPERCIST, PPP, and RECIP 1.0 criteria. Eur J Nucl Med Mol Imaging, 2022. 49 (12): p. 4271-4281. Yao, F., et al., Machine learning-based radiomics for multiple primary prostate cancer biological characteristics prediction with 18F-PSMA-1007 PET: comparison among different volume segmentation thresholds. La radiologia medica, 2022. 127 (10): p. 1170-1178. Lee, W., et al., Transformer-based Deep Neural Network for Breast Cancer Classification on Digital Breast Tomosynthesis Images. Radiol Artif Intell, 2023. 5 (3): p. e220159. Shamshad, F., et al., Transformers in medical imaging: A survey. Med Image Anal, 2023. 88 : p. 102802. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-4229122","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":291847843,"identity":"5f5d3a39-c53a-4320-8af4-261f3468ec45","order_by":0,"name":"Yujun Geng","email":"","orcid":"","institution":"Jinan University Affiliated Guangdong Second Provincial General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yujun","middleName":"","lastName":"Geng","suffix":""},{"id":291847844,"identity":"0c557353-140e-4990-aad9-0ccc181501ad","order_by":1,"name":"Ming Zhang","email":"","orcid":"","institution":"Meizhou People's Hosptal (Meizhou Academy of Medical Sciences)","correspondingAuthor":false,"prefix":"","firstName":"Ming","middleName":"","lastName":"Zhang","suffix":""},{"id":291847845,"identity":"2375a207-aaa1-4443-965f-39677e69c06d","order_by":2,"name":"Shumei Li","email":"","orcid":"","institution":"Jinan University Affiliated Guangdong Second Provincial General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Shumei","middleName":"","lastName":"Li","suffix":""},{"id":291847846,"identity":"2fd94a98-1964-4cff-b260-e1d3210cee32","order_by":3,"name":"Jingwen Li","email":"","orcid":"","institution":"Jinan University Affiliated Guangdong Second Provincial General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jingwen","middleName":"","lastName":"Li","suffix":""},{"id":291847847,"identity":"e39006ee-1071-4924-af3c-026239a5226c","order_by":4,"name":"Xinzhi Wang","email":"","orcid":"","institution":"Jinan University Affiliated Guangdong Second Provincial General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xinzhi","middleName":"","lastName":"Wang","suffix":""},{"id":291847851,"identity":"32de7007-2ed3-4f07-8248-5d12440d88c0","order_by":5,"name":"Junzhang Tian","email":"","orcid":"","institution":"Jinan University Affiliated Guangdong Second Provincial General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Junzhang","middleName":"","lastName":"Tian","suffix":""},{"id":291847853,"identity":"3dfc536a-9eb2-4330-9cef-bb0ae5caf6e3","order_by":6,"name":"Xiaofen Ma","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4UlEQVRIiWNgGAWjYDACZiCuADOYDz5IqKghUssZEIOdLdngwZljRNoE1sLPYyb5sIWZsGqD48wPHxyouBPN38xgVpHYwMbA396dgFeLZDObscGBM89yZxxmSLuRuEOGQeLM2Q14tfAzM5hJf2w7nNtwmOHYjcQzbAwGErn4tbAxs3+TOPjvcO78w4xtBYltzIS18DPzmEkcbDicu+EwMxsDUVokm3mKDQ4cO5y78TAbs0TCmWM8BP1icP74xgcHag7nzjt//uPHHxU1cvztvfi1YAAe0pSPglEwCkbBKMAKACXHTHp3fZ+pAAAAAElFTkSuQmCC","orcid":"","institution":"Jinan University Affiliated Guangdong Second Provincial General Hospital","correspondingAuthor":true,"prefix":"","firstName":"Xiaofen","middleName":"","lastName":"Ma","suffix":""}],"badges":[],"createdAt":"2024-04-07 01:59:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4229122/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4229122/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":55316897,"identity":"72091524-ff06-4423-8c97-01a271250179","added_by":"auto","created_at":"2024-04-25 15:49:33","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":442006,"visible":true,"origin":"","legend":"\u003cp\u003eFlow\u003cstrong\u003e \u003c/strong\u003ediagram of the inclusion and exclusion of cases\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-4229122/v1/4614b07927d4594f69eb821b.png"},{"id":55318717,"identity":"41608a80-00ae-4dc5-80d1-8355d1316cde","added_by":"auto","created_at":"2024-04-25 15:57:33","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":297392,"visible":true,"origin":"","legend":"\u003cp\u003ePRIMARY score of 5 points\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-4229122/v1/148a40392d4326d16ee2df52.png"},{"id":55316898,"identity":"6d1d27bf-9c61-4e2d-96ff-929e1821a2ac","added_by":"auto","created_at":"2024-04-25 15:49:33","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":125704,"visible":true,"origin":"","legend":"\u003cp\u003eFeatures selection by Lasso. The best coefficient of Lasso is 0.0256 and a total of 3 features were enrolled.\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-4229122/v1/ded04e0aba22f4565f097330.png"},{"id":55316901,"identity":"a6cf3edd-f9a3-4449-a202-1ef0f3ddad08","added_by":"auto","created_at":"2024-04-25 15:49:33","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":233203,"visible":true,"origin":"","legend":"\u003cp\u003eThe ROC curve of relevant predictors for csPCa diagnosis.\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-4229122/v1/0e35882fbb27a56a32de7b09.png"},{"id":55318718,"identity":"c3d3ca54-13e7-44ba-a187-571bebfa360c","added_by":"auto","created_at":"2024-04-25 15:57:33","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":83014,"visible":true,"origin":"","legend":"\u003cp\u003eThe difference in SUVmax and related radiomics features between the csPCa and non-csPCa. A The scatter plot of RF1; B The scatter plot of RF2; C The scatter plot of RF3; D The scatter plot of SUVmax.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-4229122/v1/44cf9b213a3f307d74f8fab1.png"},{"id":55316900,"identity":"08301f2e-d637-465f-a803-916a9273efe0","added_by":"auto","created_at":"2024-04-25 15:49:33","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":313854,"visible":true,"origin":"","legend":"\u003cp\u003eA 64 years old elderly male has a symmetrical diffuse distribution of PSMA-PET/CT, with an SUVmax of 15.5 and a PI-RADS score of 2. Due to SUVmax greater than 12.0, it is considered a malignant lesion (PRIMARY: 5). The pathological result is prostatitis with slight bleeding, but all three sets of radiomics features indicate benign (less than the Cutoff value).\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-4229122/v1/257e79f7d363188e89fa7e33.png"},{"id":57045773,"identity":"d6b0984a-1e75-4fbb-9c86-bc7a79bcdbeb","added_by":"auto","created_at":"2024-05-24 00:47:16","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2317223,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4229122/v1/cb5f23dd-e90c-4346-bfb9-fa71568759cf.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Application Value of Radiomics Features Based on PSMA PET/CT in Diagnosis of Clinically Significant Prostate Cancer: A Comparative Analysis of the PRIMARY and PI-RADS Scores","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eProstate cancer is the second most common disease that threatens the health of men worldwide and leads to approximately 1.9\u0026nbsp;million new disease cases and more than 800 thousand deaths each year[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Therefore, early diagnosis and treatment of prostate cancer are crucial. Recently, with the rapid development of medical imaging technology and biochemical chemistry, positron emission tomography (PET) technology based on labelled prostate specific membrane antigen (PSMA) has played an increasingly important role in diagnosis of prostate cancer, including lesion risk classification[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], pathological score prediction[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] and postoperative recurrence and metastasis prediction[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Numerous studies have indicated that PET has excellent diagnostic and predictive abilities[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], especially when the lesion is located in the central and transitional zones of the prostate[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], and its diagnostic ability is much better than that of multiparameter magnetic resonance imaging (mp-MRI) [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003ePSMA is a type II membrane glycoprotein that is highly expressed in prostate cancer cells and has become one of the most widely studied molecular probes in the field of nuclear medicine[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. However, as in the current situation, such an examination still has several shortcomings. First, if the size of the lesion focus is small, it is not noticed[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Second, standard uptake values (SUVs) obtained from medical images require manual delineation, leading to potential subjective bias due to observer variability[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. However, the current calculation process for radiomic features based on nuclear medicine images is relatively objective[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], with high repeatability and relatively low demands on the expertise of clinical physicians.\u003c/p\u003e \u003cp\u003eTherefore, the objective of our current study was to identify crucial radiomic features for diagnosing clinically significant prostate cancer (csPCa). We aimed to compare these features with those of MRI (PI-RADS version 2.1) [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] and PET (PRIMARY) [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] data to obtain additional valuable information from PSMA-PET/CT imaging data. In addition, we conducted a comparative analysis of the values of radiomic features based on PET/CT images and other relevant examination parameters, such as total lesions of PSMA (TLP) and metabolic tumour volume (MTV) [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. This study aimed to further analyse the origins of radiomic feature values to facilitate future research and discoveries[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e"},{"header":"2. Materials and methods","content":"\u003cp\u003e This retrospective study was approved by the Institutional Review Board of the Affiliated Guangdong Second Provincial General Hospital of Jinan University. Written informed consent was obtained from all the participants.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStudy population\u003c/b\u003e \u003c/p\u003e \u003cp\u003eA total of 255 patients' \u003csup\u003e18\u003c/sup\u003eF-PSMA-PET/CT medical images form department of nuclear medicine of Meizhou People\u0026rsquo;s Hospital were included in this study (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The cases were all confirmed by pathological biopsy to be csPCa or non-caPCa. Our inclusion criteria were as follows. (1) The quality of the image was better with a clear pathological diagnosis, and the image was confirmed many times during the subsequent follow-up. A related Gleason score was assigned for prostate cancer according to the pathological results. (2) The time interval between PET scanning and pathological puncture biopsy or MRI examination was less than 2 months. The exclusion criteria for patients were as follows: (1) before PET scanning, the patient underwent partial prostate resection surgery or had some other drug treatment; (2) severe disease in other organs accompanied by metastasis in other regions; (3) poor image quality; and (4) lacked complete clinical information. Finally, a total of 110 (mean age: 70.3\u0026thinsp;\u0026plusmn;\u0026thinsp;8.05 years, range from 48 to 89 years) patients with qualified data were included in our study; 55 patients had clinically significant lesions (Gleason score\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;7), and 55 patients had non clinically significant lesions (Gleason score\u0026thinsp;\u0026lt;\u0026thinsp;7). We also evaluated several clinical parameters, including age and prostate-specific antigen (PSA) level.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ePET image acquisition and reconstruction\u003c/b\u003e \u003c/p\u003e \u003cp\u003eAll PET/CT images were acquired with a 120 KV positron emission tomography (PET/CT) scanner (Siemens Healthcare, Biograph20) at 100 mA from the head to the thigh. After intravenous injection of \u003csup\u003e18\u003c/sup\u003eF-PSMA (2.5 MBq/kg body mass) for 90 minutes, PET/CT imaging was performed, ranging from the skull base to the thigh. Spiral CT (tube voltage: 120 kV, tube current: 30\u0026ndash;210 mA, pitch: 0.56:1, layer thickness and reconstruction layer thickness: 3.75 mm) was performed first, followed by PET (7\u0026ndash;9 beds, each bed for 2.5 minutes, matrix 256\u0026times;256, with a layer thickness and interval of 3.27 mm). Attenuation correction and iterative reconstruction of PET images using CT data, with 2 iterations and a subset of 28; display maximum intensity projection (MIP) images, PET images, CT images, and PET/CT fusion images based on our PACS workstation system. Based on the obtained PET/CT images, we manually delineated the regions of interest at the focal concentration and measured SUVmax, SUVmean, and MTV (TLP\u0026thinsp;=\u0026thinsp;SUVmean*MTV). According to previous studies, the focal concentration is expected to be 3.5 times greater than that in the surrounding areas.\u003c/p\u003e \u003cp\u003eTwo physicians with more than three years of experience in nuclear medicine were tasked with assessing the images and providing relevant PRIMARY scores[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. In cases of discrepancies, a consensus was reached through discussion. The 5-point PRIMARY score was calculated as follows: 1, no significant concentration within the prostate; 2, diffuse transition or central zone concentration; 3, focal transition zone activity above twice the background transition zone; 4, focal peripheral zone activity of any intensity; and 5, an SUV greater than 12 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eRegion of interest (ROI) segmentation and feature extraction\u003c/b\u003e \u003c/p\u003e \u003cp\u003eWe imported all images into open-source 3D-slicer (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.slicer.org\u003c/span\u003e\u003cspan address=\"https://www.slicer.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) software[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] and reconstructed them in three dimensions for better observation. The two attending physicians in our department were requested to mask the whole prostate as an ROI via the software[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. In the process of drawing, they carried out discussions each other. When their opinions were inconsistent, the final decision was made by the supervising physician.\u003c/p\u003e \u003cp\u003eBased on open-source PyRadiomics[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], a total of 1155 radiomic features, including 28 morphological features, 20 original first-order features, 24 grey-level cooccurrence matrix (GLCM) features, 14 grey-level dependence matrix (GLDM) features, 16 grey-level run-length matrix (GLRLM) features, 16 grey-level size-zone matrix (GLSZM) features, 5 neighbouring grey-level dependence matrix (NGLDM) features and 1032 features, were calculated by the wavelet and Laplacian of Gaussian (LoG) method. Intraclass correlation coefficients (ICCs) were calculated between the two physicians, and features with lower ICCs (less than 0.75) were eliminated due to their instability. The least absolute shrinkage and selection operator (LASSO) algorithm was used to select features (the weights of irrelevant features were reduced to 0) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eStatistical analysis\u003c/b\u003e \u003c/p\u003e \u003cp\u003eAll statistical calculations were performed with SPSS software and Python 3.8 (version 1.11.1). The receiver operating characteristic (ROC) curve and box plot were generated with GraphPad Prism software. The Mann‒Whitney U test was used to compare differences in continuous variables, and the chi-square test was used for categorical variables. The Cohen correlation coefficient was used to calculate correlations: 0-0.2 indicated slight agreement; 0.21\u0026ndash;0.40, fair agreement; 0.41\u0026ndash;0.60, moderate agreement; 0.61\u0026ndash;0.80, substantial agreement; and 0.81-1.0, perfect agreement. For interpretation of PI-RADS results, we employed two methods. Method 1 considered scores 1\u0026ndash;3 benign and 4\u0026ndash;5 malignant; Method 2 considered scores 1\u0026ndash;2 benign and 3\u0026ndash;5 malignant. Regarding the PRIMARY score, cases were classified as benign if 1\u0026ndash;2 and malignant if 3\u0026ndash;5. The area under the curve (AUC), accuracy (ACC), sensitivity (SEN), specificity (SPE), 95% confidence interval (95% CI), and F1 score were calculated to compare each predictor. Finally, in the combined analysis of radiomic features and PI-RADS scores, a PI-RADS score of 5 was considered malignant; for other PI-RADS scores ranging from 1\u0026ndash;4, we relied on radiomics-based judgement. The same approach was used for the PRIMARY score; a score of 4 or 5 points was considered malignant. For scores less than 4 points, we adopted the radiomics results. A P value less than 0.05 was considered to indicate statistical significance. Before utilizing RF2, the original values were multiplied by -1, as the AUC value of the original values was 0.1653 (the reverse prediction direction).\u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003e \u003cb\u003ePatient clinical characteristics\u003c/b\u003e \u003c/p\u003e \u003cp\u003eA total of 110 patients were included in this study; 60 were confirmed to have prostate adenocarcinoma by pathological biopsy, including 5 with a Gleason score of 6 and 55 with a Gleason score\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;7. Among all the patients, 28 had PI-RADS scores of 1\u0026ndash;2 (26.3%), 20 had PI-RADS scores of 3 (18.2%), 18 had PI-RADS scores of 4 (28.2%), and 44 (27.3%) had PI-RADS scores of 5. The PRIMARY score results indicated 50 (45%) negative cases and 60 (55%) positive cases. Among all malignant cases, 6 involved concurrent lymph node and bone metastases; 6 had only involved bone metastases (\u003cb\u003eTable.1\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003eRadiomics feature results\u003c/b\u003e \u003c/p\u003e \u003cp\u003eA total of 292 sets of radiomic features were deleted due to their lower intraclass correlation coefficients (ICCs\u0026thinsp;\u0026lt;\u0026thinsp;0.70). After LASSO regression and 5-fold cross-validation, a total of 3 groups of radiomic features were selected to establish a diagnostic model (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). These data are log-sigma-1-mm-3D_glcm_Correlation, log-sigma-3-mm-3D_firstorder_Minimum, and wavelet-LLH_glcm_Imc2; we labelled them RF1, RF2, and RF3, respectively. Before utilizing RF2, the original values were multiplied by -1. The box plot results indicate a significant difference in the distribution of the three sets of radiomic features between the csPCa and noncsPCa groups (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe AUC, ACC of the three radiomics features for the csPCa diagnosis was 0.8744 (95% CI:0.806\u0026ndash;0.943), 0.8364, 0.8413 (95% CI:0.762\u0026ndash;0.920), 0.8273, 0.8602 (95% CI:0.787\u0026ndash;0.933), 0.8273, respectively. Their diagnostic capabilities surpassed PSA (AUC:0.7468,95% CI:0.625\u0026ndash;0.841), MTV (AUC:0.6134,95% CI:0.506\u0026ndash;0.721), and TLP (AUC:0.7848,95% CI:0.695\u0026ndash;0.874), but were slightly lower than SUVmax (AUC:0.8691,95% CI:0.796\u0026ndash;0.942) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003e) (\u003cb\u003eTable.2\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe results of the Cohen correlation kappa indicated a substantial correlation between the three groups of radiomic features and SUVmax (kappa\u0026thinsp;=\u0026thinsp;0.757, 0.779, 0.737, respectively), TLP (kappa\u0026thinsp;=\u0026thinsp;0.645, 0.692, 0.602, respectively), and GS (kappa\u0026thinsp;=\u0026thinsp;0.696, 0.688, 0.668, respectively). However, there was a fair correlation with the MTV (kappa\u0026thinsp;=\u0026thinsp;0.326, 0.322, 0.247) (\u003cb\u003eTable.3\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003ePI-RADS and PRIMARY scores\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe ACC of Method 1 for PI-RADS was 0.8545, the recall was 0.9091, and the specificity was 0.8000. The ACC of Method 2 for PI-RADS was 0.7091, the recall was 0.9455, and the specificity was 0.4727. The ACC of PRIMARY was 0.8455, the recall was 0.8909, and the specificity was 0.8000 (\u003cb\u003eTable.2\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eThe results of the Cohen correlation kappa indicated a substantial correlation between PI-RADS (method 1) and the PRIMARY score and between SUVmax (kappa\u0026thinsp;=\u0026thinsp;0.612, 0.774) and the Gleason score (kappa\u0026thinsp;=\u0026thinsp;0.713, 0.694). There was a slight correlation with MTV (kappa\u0026thinsp;=\u0026thinsp;0.188, 0.198) and fair agreement with TLP (kappa\u0026thinsp;=\u0026thinsp;0.439, 0.491) (\u003cb\u003eTable.3\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003eCombination method\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe ACC of RF1\u0026thinsp;+\u0026thinsp;SUVmax was 0.8545, the recall was 0.8980, and the specificity was 0.8197 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). The ACC of combination1 (RF1\u0026thinsp;+\u0026thinsp;PI-RADS) was 0.8727, the recall was 0.8364, and the specificity was 0.9091. The ACC of combination2 (RF1\u0026thinsp;+\u0026thinsp;PRIMARY) was 0.8455, the recall was 0.7969, and the specificity was 0.9130 (\u003cb\u003eTable.4\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eIn this study, we primarily analysed the significance of the radiomics features log-sigma-1-mm-3D_glcm_Correlation, log-sigma-3-mm-3D_firstorder_Minimum, and wavelet-LLH_glcm_Imc2 (marked as RF1, RF2, and RF3, respectively) based on PSMA PET/CT for diagnosing clinically significant prostate cancer. The results indicate that the feature log-sigma-1-mm-3D_glcm_Correlation has a higher AUC (0.8744; 95% CI\u0026thinsp;=\u0026thinsp;0.806\u0026ndash;0.943) for diagnosing clinically significant prostate cancer; in addition, the feature log-sigma-3-mm-3D_firstorder_Minimum had a stronger correlation with SUVmax on PSMA-PET/CT (kappa\u0026thinsp;=\u0026thinsp;0.779, 95% CI\u0026thinsp;=\u0026thinsp;0.651\u0026ndash;0.885). Although the accuracy of the three radiomic features for diagnosing csPCa was not as high as that of SUVmax (AUC: 0.8691; ACC: 0.8455), RF1 showed comparable performance to that of SUVmax, and it exhibited substantial agreement with the Gleason score of the lesion (kappa\u0026thinsp;=\u0026thinsp;0.696, 95% CI: 0.566\u0026ndash;0.816). The extent to which these radiomic features complement SUVmax, whether there are hidden features, and exploration of these aspects were limited in this study due to the limited number of cases. We hope that future researchers can delve further into these discussions, as revealing additional clinical information based on medical imaging will be highly beneficial.\u003c/p\u003e \u003cp\u003eIn our current study, the cut-off value for SUVmax for diagnosing clinically significant prostate cancer was determined to be 12.6, which closely aligns with the findings of Lv et al[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]., whose cut-off value was 11.5, with a diagnostic AUC of 0.864. In fact, many studies on PSMA PET/CT for diagnosing prostate cancer heavily rely on SUVmax values. A score of 5 in the PRIMARY is based on whether SUVmax exceeds 12.0[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. According to the study by Liu et al. [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], the SUVmax of a normal prostate gland is approximately 3.40 (range: 2.00-4.40), and a value greater than 3.50 can be considered the focal concentration. Compared to visual identification, radiomics features based on image pixel values demonstrate greater objectivity and repeatability[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. However, this does not imply that it is a perfect indicator. First, before extracting features, physicians need to manually delineate the region of interest of the whole prostate gland, introducing observer errors[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Second, its interpretability is not high; further research is needed to understand its biological significance. Therefore, the current importance of deep learning-based medical image segmentation techniques is evident[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. On the one hand, this approach can significantly reduce the workload of clinical physicians, and on the other hand, it can assist researchers in quantifying certain biological parameters, such as the tumour volume[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] and prostate gland area[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAt present, an increasing number of research results have shown that PSMA PET/CT has the best prostate diagnostic ability [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], with performance that is far better than that of mp-MRI [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], especially when low-signal lesions are located in the central and transitional zones of the prostate [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. However, it is worth noting that in our current study, the diagnostic accuracy of MRI was slightly better than that of the PRIMARY score, with a diagnostic accuracy of 85.45%. Nevertheless, this result should be interpreted with caution, as categorizing cases with a PI-RADS score of 3 as either benign or malignant is problematic, given that it inherently represents a 50% probability of being malignant or benign. In our study, a total of 20 patients had a PI-RADS score of 3, and only 2 (10%) of these cases were malignant. Our study results are consistent with the findings of Guo et al.\u0026rsquo;s team [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], who reported that approximately 91.5% of cases with a PI-RADS score of 3 were negative. According to the study by the Emmett team [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], this figure is approximately 76.92% (20/26).\u003c/p\u003e \u003cp\u003eSimilarly, a PI-RADS score of 5 plays a crucial role in diagnosing prostate cancer [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. In our study, there were approximately 44 patients with a PI-RADS score of 5, with 93.2% (3/44) of the lesions being malignant. These findings align closely with those of Guo et al. [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], who reported that approximately 25.05% of patients had a PI-RADS score of 5 but contributed to 41.3% (92/223) of the malignant cases. In our study, after incorporating radiomic features based on PET/CT, the accuracy reached 87.27%, and the specificity reached 92.73%. Approximately 95% of patients with a PI-RADS score of 3 could avoid biopsy, but in patients with a PI-RADS score of 4, this proportion was 61.1%. Therefore, we strongly recommend PET/CT for patients with a PI-RADS score of 3.\u003c/p\u003e \u003cp\u003eRadiomics [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] and various deep-learning techniques [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] have been widely applied in various clinical studies in recent years, such as detection of prostate cancer lesions [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], benign-malignant classification of lesions [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], and prediction of prognosis [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e] [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e] [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. For instance, the study by Yi's team [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] confirmed that radiomics features can effectively identify malignant lesions that appear negative in PET/CT, and scholars such as Yao et al. [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e] have demonstrated that radiomics features can efficiently diagnose lymph node metastasis and extracapsular extension in prostate cancer. In our study, we did not utilize any machine learning algorithm models, primarily due to the limited sample size, with concerns about the risk of overfitting. However, the three sets of radiomics features in this study showed substantial agreement with the pathological results for the lesions (kappa\u0026thinsp;=\u0026thinsp;0.696, 0.688, 0.668). The two sets of data also demonstrated high consistency with SUVmax (kappa\u0026thinsp;=\u0026thinsp;0.757, 0.779, 0.737) but were not entirely identical. The main advantage lies in their objectivity, independence of subjective judgement, and straightforward calculation process with extremely high repeatability.\u003c/p\u003e \u003cp\u003eThere are still some shortcomings in our research. First, the number of patients in this study was relatively small. Our study results call for the participation of more patients and institutions in the future to increase the feasibility and broaden the applicability of the findings. Second, our analysis method was overly simplistic, and we did not explore the hidden variables behind various features (whether they were radiomic features based on PET/CT, MRI, or PSA) [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e] [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Investigating these aspects would be meaningful in future research.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eRadiomic features based on PSMA PET/CT exhibit strong diagnostic capabilities for clinically significant prostate cancer. These parameters showed a substantial correlation with the pathological grade (Gleason score) of prostate cancer and SUVmax based on PET images. This finding suggests the potential of these devices to address the specificity limitations in MRI-based diagnosis to a certain extent but not PET/CT.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eACC Accuracy\u003c/p\u003e\n\u003cp\u003eAUC Area under ROC curve\u003c/p\u003e\n\u003cp\u003eCI Confidence interval\u003c/p\u003e\n\u003cp\u003eCT Computed tomography\u003c/p\u003e\n\u003cp\u003eICC Intraclass correlation coefficient\u003c/p\u003e\n\u003cp\u003eMRI Magnetic resonance imaging\u003c/p\u003e\n\u003cp\u003eMTV Metabolic tumour volume\u003c/p\u003e\n\u003cp\u003ePCa Prostate cancer\u003c/p\u003e\n\u003cp\u003ePET Positron emission tomography\u003c/p\u003e\n\u003cp\u003ePSMA Prostate specific membrane antigen\u003c/p\u003e\n\u003cp\u003ePSA Prostate specific antigen\u003c/p\u003e\n\u003cp\u003ePI-RADS Prostate Imaging Reporting and Data System\u003c/p\u003e\n\u003cp\u003eROC Receiver operating characteristic\u003c/p\u003e\n\u003cp\u003eROI Region of interest\u003c/p\u003e\n\u003cp\u003eSEN Sensitivity\u003c/p\u003e\n\u003cp\u003eSPE Specificity\u003c/p\u003e\n\u003cp\u003eSUV Standardized uptake value\u003c/p\u003e\n\u003cp\u003eTLP Total lesions of PSMA\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval and consent to participate\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was funded by the National Science Foundation of China (Grant number: 82001792) and the Science Foundation of Guangdong Second Provincial General Hospital (Grant Number: 3D-A2021009). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors confirm that the data supporting the findings of this study are available from the corresponding author, upon reasonable request.\u0026nbsp;\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eYujun Geng Conceived and designed the experiments, Performed the experiments, Writing - original draft. Ming Zhang and Jingwen Li Analyzed the data, Figures; Yujun Geng and Junzhang Tian Contributed reagents/materials/analysis tools. Xinzhi Wang: Performed the experiments. Xiaofen Ma Conceived and designed the experiments, Performed the experiments. All authors reviewed the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSiegel, R.L., et al., Cancer statistics, 2023. CA Cancer J Clin, 2023. \u003cstrong\u003e73\u003c/strong\u003e(1): p. 17-48.\u003c/li\u003e\n\u003cli\u003eWaller, J., R. Flavell, and C.L. Heath, High Accuracy of PSMA PET in Initial Staging of High-Risk Prostate Cancer. Radiol Imaging Cancer, 2020. \u003cstrong\u003e2\u003c/strong\u003e(4): p. e204025.\u003c/li\u003e\n\u003cli\u003eBasso Dias, A., et al., Impact of (18)F-DCFPyL PET on Staging and Treatment of Unfavorable Intermediate or High-Risk Prostate Cancer. Radiology, 2022. \u003cstrong\u003e304\u003c/strong\u003e(3): p. 600-608.\u003c/li\u003e\n\u003cli\u003eEmmett, L., et al., Reproducibility and Accuracy of the PRIMARY Score on PSMA PET and of PI-RADS on Multiparametric MRI for Prostate Cancer Diagnosis Within a Real-World Database. J Nucl Med, 2024. \u003cstrong\u003e65\u003c/strong\u003e(1): p. 94-99.\u003c/li\u003e\n\u003cli\u003eUlaner, G.A., et al., (18)F-DCFPyL PET/CT for Initially Diagnosed and Biochemically Recurrent Prostate Cancer: Prospective Trial with Pathologic Confirmation. Radiology, 2022. \u003cstrong\u003e305\u003c/strong\u003e(2): p. 419-428.\u003c/li\u003e\n\u003cli\u003eMurad, V., et al., Association of PSMA PET-derived Parameters and Outcomes of Patients Treated for Oligorecurrent Prostate Cancer. Radiology, 2023. \u003cstrong\u003e309\u003c/strong\u003e(3): p. e231407.\u003c/li\u003e\n\u003cli\u003eLv, J., et al., A single-center, multi-factor, retrospective study to improve the diagnostic accuracy of primary prostate cancer using [(68)Ga]Ga-PSMA-11 total-body PET/CT imaging. Eur J Nucl Med Mol Imaging, 2023.\u003c/li\u003e\n\u003cli\u003eDeniffel, D., et al., Avoiding Unnecessary Biopsy: MRI-based Risk Models versus a PI-RADS and PSA Density Strategy for Clinically Significant Prostate Cancer. Radiology, 2021. \u003cstrong\u003e300\u003c/strong\u003e(2): p. 369-379.\u003c/li\u003e\n\u003cli\u003eGuo, S., et al., The PRIMARY Score: Diagnostic Performance and Added Value Compared With MRI in Detecting Clinically Significant Prostate Cancer. Clin Nucl Med, 2024. \u003cstrong\u003e49\u003c/strong\u003e(1): p. 37-44.\u003c/li\u003e\n\u003cli\u003eLawhn-Heath, C., et al., Prostate-specific Membrane Antigen PET in Prostate Cancer. Radiology, 2021. \u003cstrong\u003e299\u003c/strong\u003e(2): p. 248-260.\u003c/li\u003e\n\u003cli\u003eZamboglou, C., et al., Uncovering the invisible-prevalence, characteristics, and radiomics feature-based detection of visually undetectable intraprostatic tumor lesions in (68)GaPSMA-11 PET images of patients with primary prostate cancer. Eur J Nucl Med Mol Imaging, 2021. \u003cstrong\u003e48\u003c/strong\u003e(6): p. 1987-1997.\u003c/li\u003e\n\u003cli\u003eYi, Z., et al., Machine learning-based prediction of invisible intraprostatic prostate cancer lesions on 68 Ga-PSMA-11 PET/CT in patients with primary prostate cancer. European Journal of Nuclear Medicine and Molecular Imaging, 2021. \u003cstrong\u003e49\u003c/strong\u003e(5): p. 1523-1534.\u003c/li\u003e\n\u003cli\u003eBenomar, A., et al., Measuring Interrater Reliability. Radiology, 2023. \u003cstrong\u003e309\u003c/strong\u003e(3): p. e230492.\u003c/li\u003e\n\u003cli\u003eLiu, C., et al., (68)Ga-PSMA PET/CT Combined with PET/Ultrasound-Guided Prostate Biopsy Can Diagnose Clinically Significant Prostate Cancer in Men with Previous Negative Biopsy Results. J Nucl Med, 2020. \u003cstrong\u003e61\u003c/strong\u003e(9): p. 1314-1319.\u003c/li\u003e\n\u003cli\u003eGillies, R.J., P.E. Kinahan, and H. Hricak, Radiomics: Images Are More than Pictures, They Are Data. Radiology, 2016. \u003cstrong\u003e278\u003c/strong\u003e(2): p. 563-77.\u003c/li\u003e\n\u003cli\u003eTurkbey, B., et al., Prostate Imaging Reporting and Data System Version 2.1: 2019 Update of Prostate Imaging Reporting and Data System Version 2. Eur Urol, 2019. \u003cstrong\u003e76\u003c/strong\u003e(3): p. 340-351.\u003c/li\u003e\n\u003cli\u003eEmmett, L., et al., The PRIMARY Score: Using Intraprostatic (68)Ga-PSMA PET/CT Patterns to Optimize Prostate Cancer Diagnosis. J Nucl Med, 2022. \u003cstrong\u003e63\u003c/strong\u003e(11): p. 1644-1650.\u003c/li\u003e\n\u003cli\u003eFedorov, A., et al., 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging, 2012. \u003cstrong\u003e30\u003c/strong\u003e(9): p. 1323-41.\u003c/li\u003e\n\u003cli\u003eHan, C., et al., Radiomics Models Based on Apparent Diffusion Coefficient Maps for the Prediction of High-Grade Prostate Cancer at Radical Prostatectomy: Comparison With Preoperative Biopsy. J Magn Reson Imaging, 2021. \u003cstrong\u003e54\u003c/strong\u003e(6): p. 1892-1901.\u003c/li\u003e\n\u003cli\u003evan Griethuysen, J.J.M., et al., Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res, 2017. \u003cstrong\u003e77\u003c/strong\u003e(21): p. e104-e107.\u003c/li\u003e\n\u003cli\u003eMackin, D., et al., Measuring Computed Tomography Scanner Variability of Radiomics Features. Invest Radiol, 2015. \u003cstrong\u003e50\u003c/strong\u003e(11): p. 757-65.\u003c/li\u003e\n\u003cli\u003eHuang, Y., et al., Segment anything model for medical images? Med Image Anal, 2024. \u003cstrong\u003e92\u003c/strong\u003e: p. 103061.\u003c/li\u003e\n\u003cli\u003eWang, F., et al., Comparison of Multiple Segmentation Methods for Volumetric Delineation of Primary Prostate Cancer with Prostate-Specific Membrane Antigen-Targeted (18)F-DCFPyL PET/CT. J Nucl Med, 2024. \u003cstrong\u003e65\u003c/strong\u003e(1): p. 87-93.\u003c/li\u003e\n\u003cli\u003eJafari, E., et al., A convolutional neural network-based system for fully automatic segmentation of whole-body [(68)Ga]Ga-PSMA PET images in prostate cancer. Eur J Nucl Med Mol Imaging, 2023.\u003c/li\u003e\n\u003cli\u003eChatterjee, A., et al., Validation of Prostate Tissue Composition by Using Hybrid Multidimensional MRI: Correlation with Histologic Findings. Radiology, 2022. \u003cstrong\u003e302\u003c/strong\u003e(2): p. 368-377.\u003c/li\u003e\n\u003cli\u003eTavakoli, A.A., et al., Contribution of Dynamic Contrast-enhanced and Diffusion MRI to PI-RADS for Detecting Clinically Significant Prostate Cancer. Radiology, 2023. \u003cstrong\u003e306\u003c/strong\u003e(1): p. 186-199.\u003c/li\u003e\n\u003cli\u003eJohnston, E.W., et al., VERDICT MRI for Prostate Cancer: Intracellular Volume Fraction versus Apparent Diffusion Coefficient. Radiology, 2019. \u003cstrong\u003e291\u003c/strong\u003e(2): p. 391-397.\u003c/li\u003e\n\u003cli\u003eHesketh, R.L., A.X. Zhu, and R. Oklu, Radiomics and circulating tumor cells: personalized care in hepatocellular carcinoma? Diagn Interv Radiol, 2015. \u003cstrong\u003e21\u003c/strong\u003e(1): p. 78-84.\u003c/li\u003e\n\u003cli\u003eLi, J., et al., Predicting muscle invasion in bladder cancer based on MRI: A comparison of radiomics, and single-task and multi-task deep learning. Computer Methods and Programs in Biomedicine, 2023. \u003cstrong\u003e233\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eHu, Z., et al., Obtaining PET/CT images from non-attenuation corrected PET images in a single PET system using Wasserstein generative adversarial networks. Physics in Medicine \u0026amp; Biology, 2020. \u003cstrong\u003e65\u003c/strong\u003e(21).\u003c/li\u003e\n\u003cli\u003eSong, F., et al., A Novel Feature Engineering Method Based on Latent Representation Learning for Radiomics: Application in NSCLC Subtype Classification. IEEE Journal of Biomedical and Health Informatics, 2023: p. 1-11.\u003c/li\u003e\n\u003cli\u003eZhao, Y., et al., Deep Neural Network for Automatic Characterization of Lesions on 68Ga-PSMA PET/CT Images. Annu Int Conf IEEE Eng Med Biol Soc, 2019. \u003cstrong\u003e2019\u003c/strong\u003e: p. 951-954.\u003c/li\u003e\n\u003cli\u003eZang, S., et al., Development and validation of (68)Ga-PSMA-11 PET/CT-based radiomics model to detect primary prostate cancer. EJNMMI Res, 2022. \u003cstrong\u003e12\u003c/strong\u003e(1): p. 63.\u003c/li\u003e\n\u003cli\u003eSeifert, R., et al., Semiautomatically Quantified Tumor Volume Using (68)Ga-PSMA-11 PET as a Biomarker for Survival in Patients with Advanced Prostate Cancer. J Nucl Med, 2020. \u003cstrong\u003e61\u003c/strong\u003e(12): p. 1786-1792.\u003c/li\u003e\n\u003cli\u003eFerdinandus, J., et al., Prognostic biomarkers in men with metastatic castration-resistant prostate cancer receiving [177Lu]-PSMA-617. Eur J Nucl Med Mol Imaging, 2020. \u003cstrong\u003e47\u003c/strong\u003e(10): p. 2322-2327.\u003c/li\u003e\n\u003cli\u003eGafita, A., et al., Measuring response in metastatic castration-resistant prostate cancer using PSMA PET/CT: comparison of RECIST 1.1, aPCWG3, aPERCIST, PPP, and RECIP 1.0 criteria. Eur J Nucl Med Mol Imaging, 2022. \u003cstrong\u003e49\u003c/strong\u003e(12): p. 4271-4281.\u003c/li\u003e\n\u003cli\u003eYao, F., et al., Machine learning-based radiomics for multiple primary prostate cancer biological characteristics prediction with 18F-PSMA-1007 PET: comparison among different volume segmentation thresholds. La radiologia medica, 2022. \u003cstrong\u003e127\u003c/strong\u003e(10): p. 1170-1178.\u003c/li\u003e\n\u003cli\u003eLee, W., et al., Transformer-based Deep Neural Network for Breast Cancer Classification on Digital Breast Tomosynthesis Images. Radiol Artif Intell, 2023. \u003cstrong\u003e5\u003c/strong\u003e(3): p. e220159.\u003c/li\u003e\n\u003cli\u003eShamshad, F., et al., Transformers in medical imaging: A survey. Med Image Anal, 2023. \u003cstrong\u003e88\u003c/strong\u003e: p. 102802.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Prostate cancer, Radiomics, Magnetic resonance imaging (MRI), Positron emission tomography (PET)","lastPublishedDoi":"10.21203/rs.3.rs-4229122/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4229122/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eObjectives: \u003c/strong\u003eThe aim of our study was to explore the role of radiomic features derived from positron emission tomography (PSMA-PET)/computed tomography (CT) images in diagnosis of clinically significant prostate cancer (csPCa). Additionally, we aimed to investigate correlations between these features and other PET/CT parameters. Finally, we compared these radiomic features with the PRIMARY and PI-RADS scores to assess their complementarity and enhance the diagnostic capabilities for prostate cancer.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e: A total of 110 patients with a certain pathological diagnosis were included, and a total of 1155 sets of radiomic features were extracted from these images for each patient. We employed the LASSO regression algorithm (Lasso) to select these features and collect MRI interpretation results (PI-RADS v2.1) via consensus reading for each patient. Two clinical physicians with more than three years of experience in nuclear medicine scored the medical images (PRIMARY) for all patients, and a consensus was reached. Finally, we compared diagnostic capabilities between radiomic features and indices/scores based on medical imaging (magnetic resonance (MRI) and positron emission tomography (PET/CT)).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003eAfter the Lasso algorithm was applied, three sets of radiomic features, log-sigma-1-mm-3D_glcm_Correlation, log-sigma-3-mm-3D_firstorder_Minimum, and wavelet-LLH_glcm_Imc2, marked as RF1, RF2, and RF3, respectively, were included in the analysis. The area under the curve (AUC) for diagnosing csPCa was 0.8744 (95% CI=0.806-0.943), 0.8413 (95% CI=0.762-0.920), and 0.8602 (95% CI=0.625-0.841), with accuracies of 0.8364, 0.8273, and 0.8273, respectively. The kappa values with the maximum standard uptake value (SUVmax) were 0.757, 0.779, and 0.737; Gleason scores were 0.696, 0.688, and 0.668 of three radiomics features, respectively. The combined analysis of radiomic features(RF1) and MRI results yielded an accuracy of 0.8727, a recall of 0.8364, and a specificity of 0.9091.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion: \u003c/strong\u003eRadiomic features based on PSMA PET/CT images correlate strongly with the SUVmax and pathological diagnostic results (Gleason score) of prostate cancer patients. These methods can compensate for the insufficient specificity of MRI for prostate cancer diagnosis but not for PET/CT.\u003c/p\u003e","manuscriptTitle":"Application Value of Radiomics Features Based on PSMA PET/CT in Diagnosis of Clinically Significant Prostate Cancer: A Comparative Analysis of the PRIMARY and PI-RADS Scores","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-04-25 15:49:28","doi":"10.21203/rs.3.rs-4229122/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"32704e17-4744-421e-b626-bab861469e33","owner":[],"postedDate":"April 25th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-05-24T00:39:08+00:00","versionOfRecord":[],"versionCreatedAt":"2024-04-25 15:49:28","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4229122","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4229122","identity":"rs-4229122","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}