Predicting bone metastasis and high-grade Gleason scores in prostate cancer: a retrospective study integrating clinical features and magnetic resonance imaging radiomics

Introduction

Prostate cancer (PCa) is a prevalent malignant tumor among middle-aged and elderly men, with bone metastasis being the most common metastatic form (1). Patients with bone metastasis experience a significant decline in survival rates, therefore, it is crucial to accurately predict the impact of PCa bone metastasis risk on life expectancy and quality of life. Those diagnosed with high risk PCa face an increased risk of treatment failure and mortality (2). At present, the commonly used prediction methods in clinical practice mainly rely on traditional clinical indicators such as prostate-specific antigen (PSA) levels, Gleason scores, etc. The Gleason scoring System is the most widely used method in clinical practice for assessing the aggressiveness of PCa, as it directly influences the patient’s prognosis (3). In addition, somatic genetic evaluation is also required for patients with metastatic/high-grade PCa (4). Therefore, we propose that the early diagnosis and prediction of PCa bone metastasis and high-grade Gleason Scores through MRI examinations hold significant clinical importance.

Radiomics, introduced by Lambin in 2012 (5), involves extracting multiple quantitative imaging features from medical images to describe the biological characteristics of different diseases. It serves as a bridge between macroscopic imaging information and microscopic pathological and molecular information, providing a quantitative basis for non-invasive diagnostic and therapeutic decision-making (6). Radiomics has been extensively applied in precision diagnosis, classification, prognostic assessment, treatment guidance, and therapeutic monitoring of tumors (7-9). MRI is an important imaging tool for the diagnosis and staging of PCa, which can provide high-resolution anatomical and functional information. Through MRI radiomics, potential biomarkers associated with PCa bone metastasis can be identified, thereby improving the accuracy and reliability of prediction.

In this study, we extracted clinical and pathological data, as well as MRI radiomics data from 168 patients with PCa, aiming to develop and validate a predictive model based on MRI radiomics, which can be is to predict the risk of bone metastasis and high-grade Gleason score. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-412/rc).

Methods

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Zhongshan Hospital of Traditional Chinese Medicine (No. 2022ZSZY-LLK-021), and due to the retrospective nature of this study, the requirement for obtaining patient consent was waived.

Patient population

This study used a retrospective analysis, and the data are sourced from Zhongshan Traditional Chinese Medicine Hospital. We collected clinical and pathological data from 168 patients diagnosed with pathology confirmed PCa from July 2011 to April 2025. The data that we collected included age, tPSA, clinical prostate-specific antigen (cPSA), fPSA, pathological Gleason grade, and pathological histological type. The inclusion and exclusion criteria are shown in Figure 1.

Figure 1 Patient characteristics and inclusion/exclusion criteria. MRI, magnetic resonance imaging.

Image acquisition

All prostate MRI exams were carried out on a 3.0T GE Healthcare system with an eight-channel abdominal coil while the patients lay supine. The protocol—comprising diffusion-weighted imaging (DWI), T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and T2 fat-suppressed imaging (T2-FS)—was implemented in accordance with Prostate Imaging Reporting and Data System (RI-RADS) v2.1 guidelines (10) (Table 1). Each individual enrolled in the study underwent the identical, standardized imaging protocol.

Radiomics feature extraction

After MRI acquisition, all images were manually segmented by a radiologist with over ten years of experience. Inter-rater reliability was quantified with the intraclass correlation coefficient (ICC), and only segmentations with ICC >0.75 were accepted; any disagreements were resolved by a senior consultant radiologist. During the extraction process, patient information will be masked. If the image quality is too low, the sequence is incomplete, or clinical data is missing, the case will be directly abandoned. The aim is to accurately predict bone metastasis and Gleason grading of PCa before treatment through non-invasive magnetic resonance examination of patients. After optimization, the histogram bin width was set to 25, which influenced the gray-level discretization process used to calculate texture features. A larger bin width reduces computational load but may result in the loss of some texture information; the optimal value was determined through testing. The resampled pixel spacing was standardized to 1×1×1 mm³ to ensure consistent image resolution and eliminate the impact of different resolutions on feature extraction. Following standardization, radiomic features were categorized into seven groups: original shape, first-order statistics, gray level co-occurrence matrix (GLCM), gray level dependence matrix (GLDM), gray level run length matrix (GLRLM), gray level size zone matrix (GLSZM), neighboring gray tone difference matrix (NGTDM), and their Laplacian-of-Gaussian-filtered counterparts, all z-score normalized to ensure a uniform distribution.. Using the PyRadiomics (v3.0) toolkit (11), features were extracted from the original images, five different Laplacian of Gaussian (LoG)-filtered images (sigma =1.0, 2.0, 3.0, 4.0, 5.0), and Wavelet-transformed images. The wavelet transformation produces eight sub-bands [low-low-low (LLL), low-low-high (LLH), low-high-low (LHL), low-high-high (LHH), high-low-low (HLL), high-low-high (HLH), high-high-low (HHL), high-high-high (HHH)]. For each image type (original, LoG, Wavelet), the seven feature categories were extracted. The default number of features for each category is as follows: first order statistics: 18; shape-based features: 14 (only applicable to the original image); GLCM features: 24; GLRLM features: 16; GLSZM features: 16; GLDM features: 14; NGTDM features: 5. The total number of features extracted was 1,316 (Figure 2).

Figure 2 MRI radiomics feature extraction and data analysis diagram. AdaBoost, Adaptive Boosting; DT, Decision Tree; KNN, k-Nearest Neighbor; LR, Logistic Regression; MRI, magnetic resonance imaging; NB, Naive Bayes; RF, Random Forest; SVM, Support Vector Machine; XGBoost, Extreme Gradient Boosting.

Feature selection and model validation

One hundred sixty-eight eligible datasets were randomly split into a training set (n=117) and an independent testing set (n=51). Dimensionality reduction began with t-tests to remove non-informative variables, followed by least absolute shrinkage and selection operator (LASSO) regression whose regularisation parameter was tuned by ten-fold cross-validation. This process distilled the initial 1,316 radiomic features to 21 (bone-metastasis task) or 27 (high-grade Gleason task) highly discriminative predictors and curbed redundancy and overfitting. Internal verification used cross-validation; recalibration was applied whenever predicted probabilities deviated from observed outcomes. The validation dataset differed from the development set only by stricter image-quality criteria to safeguard feature-extraction fidelity.

Statistical analysis

The model performance is evaluated on an independent test set using accuracy AUC, comprehensive evaluation of six indicators including sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), all based on scikit learn 1.0 (https://scikit-learn.org) Calculate. The difference in clinical feature distribution between the training set and the test set was determined using independent sample t-test, with a significance level of two-sided P<0.05.

Model construction

The machine learning algorithms used included Random Forest (RF), Logistic Regression (LR), Support Vector Machines (SVMs), Decision Tree (DT), K-Nearest Neighbors (KNN), Naïve Bayes (NB), Adaptive Boosting (AdaBoost) and Extreme Gradient Boosting (XGBoost).

Results

Clinical information features

A total of 168 patients were included in this study, with a mean age of 71.17±9.33 years. Among these patients, 69 had bone metastasis while 99 did not; 119 patients had high-grade Gleason scores, and 49 had low-grade Gleason scores. Those harboring bone metastases and high-grade Gleason patterns showed markedly elevated tPSA and fPSA levels versus patients lacking metastases or harboring low-grade tumors (P<0.05), whereas the training and testing cohorts exhibited no significant divergence (Tables 2-4).

Radiomics results

The research team developed two machine-learning models on two independent datasets: one integrating clinical variables with radiomic features and the other relying solely on radiomic signatures. Model performance was assessed using correlation coefficients (Figure 3), and dimensionality reduction plus feature selection were executed via t-tests (Tables 2,3) and LASSO regression (Figure 4). Ultimately, 21 informative features were retained for bone-metastasis prediction, whereas 27 features were identified for high-grade Gleason score prediction (Table 5).

Figure 3 Correlation coefficients for the (A) bone metastasis prediction and (B) high-grade Gleason score prediction tasks. fPSA, free PSA; PSA, prostate-specific antigen.

Figure 4 LASSO parameter determination for the (A,B) bone metastasis and (C,D) high-grade Gleason score prediction tasks. LASSO, least absolute shrinkage and selection operator; MSE, mean squared error.

In the test set, the XGBoost model that combined clinical and radiomic features achieved the best performance for predicting bone metastasis, with an accuracy of 0.863, AUC of 0.875, sensitivity of 0.762, specificity of 0.933, PPV of 0.889, and NPV of 0.848 (Table 6, Figure 5A), significantly outperforming the radiomics-only counterpart (accuracy 0.725, AUC 0.732). AUC of training set of bone metastasis prediction task can be found in Figure 5B.

Figure 5 ROC curves for bone metastasis and high-grade Gleason score prediction tasks. (A) Test set of bone metastasis prediction task; (B) training set of bone metastasis prediction task; (C) test set of high-grade Gleason score prediction task; (D) training set of high-grade Gleason score prediction task. AdaBoost, Adaptive Boosting; AUC, area under the curve; KNN, k-Nearest Neighbor; ROC, receiver operating characteristic; SVM, Support Vector Machine; XGBoost, Extreme Gradient Boosting.

Similarly, for predicting high-grade Gleason scores, the XGBoost model integrating clinical and radiomic features again delivered the best performance, achieving an accuracy of 0.784, AUC of 0.830, sensitivity of 0.865, PPV of 0.842, and NPV of 0.615 (Table 7, Figure 5C), significantly surpassing the radiomics-only model (accuracy 0.706, AUC 0.778). AUC of training set of high-grade Gleason score prediction task can be found in Figure 5D.

Feature-importance analysis revealed that, in the bone-metastasis task, fPSA was the most influential clinical variable, whereas log-sigma-5-0-mm-3D_glszm_ZoneEntropy was the dominant radiomic feature. For high-grade Gleason-score prediction, fPSA again ranked as the top clinical predictor, and wavelet-HLH_gldm_HighGrayLevelEmphasis emerged as the leading radiomic signature. Conversely, wavelet-HLL_firstorder_Variance exhibited the lowest contribution among radiomic features for bone metastasis, while wavelet-HHL_glszm_HighGrayLevelZoneEmphasis was the least informative radiomic variable for high-grade Gleason-score classification (Figure 6).

Figure 6 Feature weight bar charts for the (A) bone metastasis and (B) high-grade Gleason score prediction tasks. fPSA, free PSA; PSA, prostate-specific antigen.

Discussion

In this study, we included a cohort of 168 patients with pathologically confirmed PCa and extracted a total of 1,316 radiomics features from their prostate MRI scans. The dataset was subsequently divided into training and testing sets to facilitate model development and validation. To pinpoint the most predictive features for bone metastasis and high-grade Gleason patterns, feature selection was performed using t-tests and LASSO regression. These methods enabled us to pinpoint the features that contributed most significantly to the predictive models. The ultimate goal was to establish a non-invasive, radiation-free method for predicting bone metastasis and Gleason grade scores using prostate MRI, thereby enhancing decision-making and patient management in PCa care (11).

Currently, the gold standard for the diagnosis of PCa is pathological examination, which can accurately determine the Gleason score of the patient. Prostate biopsy is an indispensable means for diagnosing PCa and developing treatment plans. In addition, biopsy histology is of great value for local localization (12). However, this method is invasive and carries several associated risks. Improper procedures may lead to the rupture of blood vessels at the puncture site, resulting in bleeding (13,14). Additionally, inadequate preoperative disinfection or improper postoperative care may increase the risk of infection. Furthermore, patients may experience temporary erectile dysfunction as a consequence of the biopsy procedure (15). The necessity for multiple biopsies to achieve a conclusive diagnosis also poses challenges, as there is a risk that sufficient tissue may not be obtained for accurate diagnosis (16). Compounding this issue, some patients may undergo unnecessary biopsies due to false-positive results from screening tests (17). These complications highlight the need for non-invasive diagnostic alternatives, such as the radiomics approach explored in our study, which could mitigate these risks while providing reliable predictions for bone metastasis and Gleason score.

The assessment of bone metastasis in PCa typically involves several imaging modalities, including bone electrochemotherapy (ECT) scans, whole-body bone scintigraphy, positron emission tomography/computed tomography (PET/CT), positron emission tomography/magnetic resonance imaging (PET/MRI) etc., especially 68GaPSMA PET/CT (18,19). Both bone ECT scans and whole-body bone scintigraphy carry the risk of radiation exposure, which may adversely affect the hematopoietic function of the bone marrow, especially in patients with impaired hematopoietic function. Although PET/CT and PET/MRI leverage the high sensitivity of PET and the high specificity of CT/MRI, making them particularly valuable for restaging PCa patients with normal biochemical indicators (20), advanced imaging techniques such as PET and DWI during PET/MRI can assess the metabolic activity of tumors, tumor tissue, tumor cells, and the biological characteristics of tumor tissues (21). Gallium-68-labelled prostate-specific membrane antigen (68GaPSMA) PET/CT alone showed high accuracy in diagnosing and staging high-risk PCa (19). Despite these advantages, the clinical application of these advanced imaging techniques remains limited due to their high costs and the lack of corresponding equipment in primary healthcare facilities (22).

However, both bone metastasis and Gleason grade are critical factors in the diagnosis of PCa, significantly impacting patient quality of life and prognosis. Emerging evidence indicates that 70% of advanced PCa patients progress to bone metastasis and incur skeletal-related events (SREs) (23,24). These can trigger spinal cord compression, pathologic fractures, and sharp quality-of-life decline. The 2019 International Society of Urological Pathology (ISUP) consensus (25) subdivides Gleason 7 into 3+4 (grade 2) and 4+3 (grade 3), assigning grades ≥3 a worse prognosis and greater mortality risks (26). Therefore, we define a Gleason score ≥4+3 as a high-grade and a Gleason score ≤3+4 as a low-grade. The early prediction of bone metastasis and Gleason grade is crucial for effective diagnosis and treatment.

MRI is a non-invasive, radiation-free imaging modality that is widely used in clinical practice, including in primary care hospitals. The results in Tables 6,7 show that the AUC of the radiomics alone model is lower than that of the predictive model that combines clinical and radiomic features. In this study, we combined clinical features with radiomics features extracted from prostate MRI to develop and validate predictive models. Our results showed that the XGBoost model, which combined clinical with radiomics features, achieved high accuracy in predicting bone metastasis (AUC 0.875) and high-grade Gleason scores (AUC 0.830).

Among the findings, the most significant clinical feature was fPSA, while the most significant radiomics features were log-sigma-5-0-mm-3D_glszm_ZoneEntropy for bone metastasis and wavelet-HLH_gldm_HighGrayLevelEmphasis for high-grade Gleason scores. Conversely, the least effective radiomics features were wavelet-LHL_glszm_ZoneEntropy for bone metastasis and wavelet-HHL_glszm_HighGrayLevelZoneEmphasis for high-grade Gleason scores.

These findings suggest a strong correlation between specific radiomics features and the underlying biological characteristics of PCa. For instance, significant radiomics features such as log-sigma-5-0-mm-3D_glszm_ZoneEntropy and wavelet-HLH_gldm_HighGrayLevelEmphasis may reflect tumor heterogeneity and aggressiveness, which are key biological factors influencing the development of bone metastasis and high-grade Gleason scores (27). Recent studies have shown that radiomics features derived from MRI can provide valuable information on tumor biology, such as tumor cellularity, hypoxia, and angiogenesis, all of which are closely related to cancer progression and metastasis. For example, Gugliandolo et al. (28) demonstrated that texture features were highly predictive of Gleason score, PIRADS v2 score, and risk class, with AUC values ranging from 0.74 to 0.94. In another study, Ma et al. evaluated 210 patients with pathology-confirmed extraprostatic extension and found that the radiomics signature outperformed visual assessment by radiologists in predicting extraprostatic extension (29). These studies indicate that radiomics features can serve as surrogate markers for underlying biological processes in PCa.

Moreover, the identification of fPSA as the most significant clinical feature underscores the importance of integrating clinical and imaging data to enhance predictive accuracy. The combination of clinical parameters and radiomics features has been shown to improve diagnostic performance and risk stratification in PCa. Once embedded into Picture Archiving and Communication System (PACS) or Hospital Information System (HIS), the model is expected to automatically extract clinical and radiomic features, enabling rapid, non-invasive prediction of prostate-cancer bone metastasis and high-grade Gleason scores, thereby reducing treatment costs and yielding long-term benefits, although prospective validation remains indispensable.

In this study, we did not observe a clear association between patient age and bone metastasis or high-grade Gleason scores. However, there are limitations in this study. First, due to the retrospective nature of the study, the imaging examination requirements for determining the presence of bone metastasis in enrolled patients include: (I) low-risk PCa (PSA <10 ng/mL, Gleason <7, cT1-T2a) has undergone pelvic and prostate MRI examination, including pelvic, lower lumbar, and proximal femur, which is most commonly involved in PCa bone metastasis; (II) moderate- or high-risk patients have received at least an MRI or CT scan of the entire spine, pelvis, and proximal femur to rule out bone metastasis. Second, radiomics relies on the extraction and analysis of radiological features through appropriate delineation of ROIs, and the variations in the operations performed by different physicians may impact the study’s findings. However, this issue can be mitigated by employing automatically generated contours for extracting radiomics features (30). Finally, the study was conducted at a single center with a relatively small sample size. This is attributable to the intricate challenges of multi-center data harmonization—encompassing scanner heterogeneity, divergent ethical approval pathways, and stringent patient-privacy safeguards—which, to date, have precluded external validation. Although the high-dimensional imaging data contained redundant features, LASSO-based dimensionality reduction yielded strong performance in the internal test set. Nevertheless, inter-site variations in field strength (1.5T vs. 3.0T) and scanning protocols—sequence parameters, reconstruction algorithms, etc.—may still compromise model generalizability, rendering overfitting an objective concern. Going forward, we will actively initiate a multi-institutional collaboration that adheres to harmonized data-acquisition protocols—standardized sequences and parameter ranges—to enlarge the dataset, perform external validation, iteratively enhance model generalizability, and effectively mitigate overfitting. In addition, this study is only preliminary; in addition, Artificial Intelligence will improve all models for the diagnosis and staging of PCa (31).

Conclusions

In summary, PCa still requires pathological results to confirm its diagnosis, however, in this study, we propose a predictive model that combines clinical features with radiomics features from prostate MRI. This model can non-invasively and radiation-free predict bone metastasis and high-grade Gleason Scores in PCa. It provides a valuable tool for clinical diagnosis and treatment that can be achieved in non-large hospitals.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-412/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-412/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-412/prf

Funding: This work was supported by grants from the Zhongshan Social Welfare Science and Technology Research Project (grant Nos. 2022B1107 and 2022B3010) for Fundamental Research.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-412/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Zhongshan Hospital of Traditional Chinese Medicine (No. 2022ZSZY-LLK-021), and due to the retrospective nature of this study, the requirement for obtaining patient consent was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
2. Cooperberg MR, Broering JM, Carroll PR. Time trends and local variation in primary treatment of localized prostate cancer. J Clin Oncol 2010;28:1117-23. [Crossref] [PubMed]
3. Zelic R, Giunchi F, Fridfeldt J, et al. Prognostic Utility of the Gleason Grading System Revisions and Histopathological Factors Beyond Gleason Grade. Clin Epidemiol 2022;14:59-70. [Crossref] [PubMed]
4. Vatrano S, Pepe P, Pepe L, et al. BRCA mutations and prostate cancer: should urologist improve daily clinical practice? Arch Ital Urol Androl 2025;97:13635. [Crossref] [PubMed]
5. Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer 2012;48:441-6. [Crossref] [PubMed]
6. Liu Z, Wang S, Dong D, et al. The Applications of Radiomics in Precision Diagnosis and Treatment of Oncology: Opportunities and Challenges. Theranostics 2019;9:1303-22. [Crossref] [PubMed]
7. Qi YJ, Su GH, You C, et al. Radiomics in breast cancer: Current advances and future directions. Cell Rep Med 2024;5:101719. [Crossref] [PubMed]
8. Corti A, Lo Iacono F, Ronchetti F, et al. Enhancing cardiovascular risk stratification: Radiomics of coronary plaque and perivascular adipose tissue - Current insights and future perspectives. Trends Cardiovasc Med 2025;35:47-59. [Crossref] [PubMed]
9. Badani A, Ozair A, Khasraw M, et al. Immune checkpoint inhibitors for glioblastoma: emerging science, clinical advances, and future directions. J Neurooncol 2025;171:531-47. [Crossref] [PubMed]
10. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res 2017;77:e104-7. [Crossref] [PubMed]
11. Zhao YY, Xiong ML, Liu YF, et al. Magnetic resonance imaging radiomics-based prediction of clinically significant prostate cancer in equivocal PI-RADS 3 lesions in the transitional zone. Front Oncol 2023;13:1247682. [Crossref] [PubMed]
12. Pepe P, Fraggetta F, Galia A, et al. Is quantitative histologic examination useful to predict nonorgan-confined prostate cancer when saturation biopsy is performed? Urology 2008;72:1198-202. [Crossref] [PubMed]
13. Anastasiadis A, Zapała L, Cordeiro E, et al. Complications of prostate biopsy. Expert Rev Anticancer Ther 2013;13:829-37. [Crossref] [PubMed]
14. Li J, Zhu C, Yang S, et al. Non-Invasive Diagnosis of Prostate Cancer and High-Grade Prostate Cancer Using Multiparametric Ultrasonography and Serological Examination. Ultrasound Med Biol 2024;50:600-9. [Crossref] [PubMed]
15. Borghesi M, Ahmed H, Nam R, et al. Complications After Systematic, Random, and Image-guided Prostate Biopsy. Eur Urol 2017;71:353-65. [Crossref] [PubMed]
16. Qiu Y, Liu YF, Shu X, et al. Peritumoral Radiomics Strategy Based on Ensemble Learning for the Prediction of Gleason Grade Group of Prostate Cancer. Acad Radiol 2023;30:S1-S13. [Crossref] [PubMed]
17. Bachour DM, Chahin E, Al-Fahoum S. Frequency of Unnecessarily Biopsies among Patients with Suspicion of Prostate Cancer in Syrian Men. Asian Pac J Cancer Prev 2015;16:5967-70. [Crossref] [PubMed]
18. Bjurlin MA, Rosenkrantz AB, Beltran LS, et al. Imaging and evaluation of patients with high-risk prostate cancer. Nat Rev Urol 2015;12:617-28. [Crossref] [PubMed]
19. Pepe P, Pennisi M. Targeted Biopsy in Men High Risk for Prostate Cancer: (68)Ga-PSMA PET/CT Versus mpMRI. Clin Genitourin Cancer 2023;21:639-42. [Crossref] [PubMed]
20. Umbehr MH, Müntener M, Hany T, et al. The role of 11C-choline and 18F-fluorocholine positron emission tomography (PET) and PET/CT in prostate cancer: a systematic review and meta-analysis. Eur Urol 2013;64:106-17. [Crossref] [PubMed]
21. Schmidt GP, Schoenberg SO, Schmid R, et al. Screening for bone metastases: whole-body MRI using a 32-channel system versus dual-modality PET-CT. Eur Radiol 2007;17:939-49. [Crossref] [PubMed]
22. Bae H, Yoshida S, Matsuoka Y, et al. Apparent diffusion coefficient value as a biomarker reflecting morphological and biological features of prostate cancer. Int Urol Nephrol 2014;46:555-61. [Crossref] [PubMed]
23. Hensel J, Thalmann GN. Biology of Bone Metastases in Prostate Cancer. Urology 2016;92:6-13. [Crossref] [PubMed]
24. Onukwugha E, Yong C, Mullins CD, et al. Skeletal-related events and mortality among older men with advanced prostate cancer. J Geriatr Oncol 2014;5:281-9. [Crossref] [PubMed]
25. van Leenders GJLH, van der Kwast TH, Grignon DJ, et al. The 2019 International Society of Urological Pathology (ISUP) Consensus Conference on Grading of Prostatic Carcinoma. Am J Surg Pathol 2020;44:e87-99. [Crossref] [PubMed]
26. Siech C, Hoeh B, Rohlfsen E, et al. Organ-confined pT2 ISUP4/5 vs. nonorgan confined pT3/4 ISUP2 vs. ISUP3 prostate cancer: Differences in biochemical recurrence-free survival after radical prostatectomy. Urol Oncol 2024;42:448.e1-8. [Crossref] [PubMed]
27. Ferro M, de Cobelli O, Musi G, et al. Radiomics in prostate cancer: an up-to-date review. Ther Adv Urol 2022;14:17562872221109020. [Crossref] [PubMed]
28. Gugliandolo SG, Pepa M, Isaksson LJ, et al. MRI-based radiomics signature for localized prostate cancer: a new clinical tool for cancer aggressiveness prediction? Sub-study of prospective phase II trial on ultra-hypofractionated radiotherapy (AIRC IG-13218). Eur Radiol 2021;31:716-28. [Crossref] [PubMed]
29. Ma S, Xie H, Wang H, et al. MRI-Based Radiomics Signature for the Preoperative Prediction of Extracapsular Extension of Prostate Cancer. J Magn Reson Imaging 2019;50:1914-25. [Crossref] [PubMed]
30. Schmidt RM, Delgadillo R, Ford JC, et al. Assessment of CT to CBCT contour mapping for radiomic feature analysis in prostate cancer. Sci Rep 2021;11:22737. [Crossref] [PubMed]
31. Caputo A, Maffei E, Gupta N, et al. Computer-assisted diagnosis to improve diagnostic pathology: A review. Indian J Pathol Microbiol 2025;68:3-10. [Crossref] [PubMed]