Predicting Response to Neoadjuvant Hormonal Therapy and Prognostic Outcomes in High-Risk Prostate Cancer Using MRI-Based Radiomics

Predicting Response to Neoadjuvant Hormonal Therapy and Prognostic Outcomes in High-Risk Prostate Cancer Using MRI-Based Radiomics | 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 Predicting Response to Neoadjuvant Hormonal Therapy and Prognostic Outcomes in High-Risk Prostate Cancer Using MRI-Based Radiomics Hai Zhou, Zhi-Yu Qian, Zhi-Hao Chen, Jian-Jin Fan, Yong-Xin Mao, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8655062/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 29 Apr, 2026 Read the published version in BMC Cancer → Version 1 posted 10 You are reading this latest preprint version Abstract Purpose This study aimed to develop and validate an MRI-based radiomics model for predicting complete pathological response (pCR) to neoadjuvant hormonal therapy (NHT) and to assess its prognostic value in patients with high-risk prostate cancer (HRPCa). Methods This retrospective study included 140 patients with HRPCa who underwent NHT followed by radical prostatectomy at Huadong Hospital. Radiomic features were extracted from preoperative apparent diffusion coefficient (ADC) maps derived from multiparametric MRI. Feature selection was performed using appropriate statistical methods, and predictive models were constructed using various machine learning classifiers. The performance of the radiomics signature and a combined model integrating clinical MRI and pathological features were evaluated. Results For pCR prediction, one clinical MRI feature (seminal vesicle invasion), one pathological feature (cribriform adenocarcinoma), and eight radiomics features were ultimately selected. The combined model, which incorporated these features with a radiomics signature generated by a Support Vector Machine (SVM) classifier, demonstrated excellent discriminative ability. It achieved areas under the curve (AUC) of 0.909 (95% CI: 0.852–0.965) in the training cohort and 0.947 (95% CI: 0.882–1.012) in the internal validation cohort. Decision curve analysis confirmed its clinical net benefit. Furthermore, Kaplan-Meier analysis revealed that patients predicted by the model to have a higher probability of pCR experienced significantly better disease-free survival compared to those with a lower predicted probability. Conclusion Based on a substantial sample size and rigorous internal validation, the proposed MRI-based radiomics model achieved a high Radiomics Quality Score (RQS). It shows strong potential for the non-invasive prediction of treatment response and prognosis in HRPCa, demonstrating significant clinical value. Neoadjuvant Hormonal Therapy complete pathological response Prostate cancer Radiomics Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1 INTRODUCTION Prostate cancer (PCa) is the most common malignancy and the second leading cause of cancer-related death among men worldwide [ 1 ]. High-risk prostate cancer (HRPCa) represents a risk category associated with a high probability of disease progression or recurrence [ 2 ]. In China, the majority of PCa cases are diagnosed at advanced stages, with a particularly high prevalence of the high-risk subgroup [ 3 ]. Accounting for 15%–20% of clinically localized PCa cases, HRPCa is characterized by an increased likelihood of biochemical recurrence (BCR), metastatic progression, and cancer-specific mortality [ 4 ]. Despite ongoing clinical efforts, a consensus on the optimal treatment strategy for men with HRPCa remains elusive [ 5 ]. Consequently, novel therapeutic strategies, including multimodal approaches, are needed. Neoadjuvant hormonal therapy (NHT) combined with radical prostatectomy (RP) or radiotherapy (RT) may improve outcomes in PCa [ 6 ]. Specifically, neoadjuvant androgen deprivation therapy (ADT) prior to RP, compared to RP alone, can reduce rates of pathological T3 stage (downstaging), positive surgical margins, and lymph node invasion [ 7 ]. Although NHT before RP has demonstrated significant improvements in pathological outcomes, it has not yet been shown to confer a definitive survival benefit for patients with high-risk disease [ 8 ]. Accurately identifying patients who may respond to NHT is clinically important. Previous studies have proposed several strategies for this purpose, such as monitoring prostate-specific antigen (PSA) kinetics [ 9 ] or using prostate-specific membrane antigen positron emission tomography/computed tomography (PSMA PET/CT) [ 10 ]. However, these approaches are largely based on a single modality, lack robust quantifiable metrics, and have demonstrated limited reliability and accuracy. Radiomics, which involves the high-throughput extraction of quantitative features from medical images, holds promise for providing non-invasive insights into tumor heterogeneity and underlying pathophysiology [ 11 ]. Prior research has indicated the potential value of radiomics features derived from multiparametric magnetic resonance imaging (mpMRI) in predicting response to NHT and in assessing treatment response across various cancers, including breast, colorectal, and prostate cancer [ 12 – 13 ]. Nevertheless, in the context of PCa, existing radiomics studies are often limited by small sample sizes and the absence of independent validation cohorts. To address these limitations, we conducted this study utilizing a larger sample size with an internal validation cohort. Our objectives are: (1) to develop and evaluate the efficacy of an mpMRI-based radiomics model, employing different classifiers, for predicting a complete pathological response (pCR) to NHT in patients with HRPCa prior to treatment; and (2) to construct and validate a combined model that integrates clinical MRI features with the optimal radiomics signature to improve predictive performance and explore its potential prognostic value. 2 MATERIALS AND METHODS 2.1 Patient Cohort This retrospective study was approved by the Ethics Committee of Huadong Hospital Affiliated with Fudan University, and written informed consent was obtained from all participants. We enrolled 140 patients diagnosed with high-risk prostate cancer (PCa) who received neoadjuvant hormonal therapy (NHT) followed by radical prostatectomy (RP) at our center between January 2017 and January 2025. All patients had a pathological diagnosis of prostatic adenocarcinoma via prostate biopsy prior to NHT initiation. PCa staging was based on the 2017 TNM classification. According to the 2021 European Association of Urology (EAU) guidelines, high-risk PCa encompasses locally high-risk disease (PSA >20 ng/mL, ISUP grade group 4 or 5, or clinical stage cT2c) and locally advanced disease (clinical stage cT3–4 or cN+, irrespective of PSA level and ISUP grade). Inclusion criteria were: (1) biopsy-confirmed high-risk PCa, and (2) availability of prostate multiparametric magnetic resonance imaging (mpMRI) performed prior to NHT initiation. Exclusion criteria were: (1) presence of metastatic PCa at diagnosis, (2) absence of pre-NHT MRI data, and (3) failure to undergo RP (e.g., patients who underwent transurethral resection of the prostate [TURP] instead). After excluding [X] patients meeting the exclusion criteria, 140 eligible subjects with complete clinicopathological data were finally analyzed. They were divided into a training cohort (n=95) and an internal validation cohort (n=45). 2.2 Clinical and MRI Feature Assessment Clinical characteristics, including patient age and pre-NHT prostate-specific antigen (PSA) level, were retrieved from medical records. MRI morphological features—tumor size (measured on T2-weighted sagittal images), MRI-detected extracapsular extension (ECE), and seminal vesicle invasion (SVI)—were independently evaluated by two associate chief urologists (Reader 1 and Reader 2, each with >10 years of experience in prostate MRI interpretation). In cases of disagreement, a third senior urologist (Reader 3, with >20 years of experience) provided a final assessment. All readers were blinded to clinical outcomes and postoperative pathology. The specific criteria for assessing MRI morphological features, along with representative MR images illustrating ECE and SVI, are provided in Figure S1. 2.3 MRI Acquisition and Tumor Segmentation All patients underwent mpMRI on a SIEMENS Verio 3.0 T scanner within two weeks before NHT initiation. Patients were scanned in the supine position with the prostate at the isocenter. They were instructed to empty their bowels and maintain a moderately full bladder prior to imaging. The scanning protocol included axial, sagittal, and coronal T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and the derived apparent diffusion coefficient (ADC) maps. Detailed sequence parameters are listed in Table S1. Using 3D Slicer software (version 3.4.3), a uroradiologist manually delineated the region of interest (ROI) slice-by-slice along the tumor border on T2WI and high b-value DWI (b = 800 s/mm²) images, covering the entire tumor volume. The ROIs were then propagated to the corresponding ADC maps. During segmentation on DWI, anatomical reference from T2W images was carefully integrated to ensure accuracy. 2.4 Reproducibility Assessment of Radiomics Feature Extraction To evaluate inter- and intra-observer reproducibility, a random subset of 50 patients was selected. For intra-observer assessment, Reader 1 performed manual ROI segmentation on the MRI scans and repeated the process one month later, blinded to the initial contours. For inter-observer assessment, Reader 2 independently delineated the ROIs for the same patients. Both readers were blinded to pathological and clinical outcomes. The intraclass correlation coefficient (ICC) was calculated for each extracted radiomics feature. Features demonstrating an ICC (inter- or intra-observer) ≥ 0.75 were considered to have good reproducibility and were retained for subsequent analysis. 2.5 NHT Regimen and Treatment Response Assessment The NHT regimen comprised subcutaneous injections of either goserelin or leuprorelin (3.6 mg) every 28 days, combined with daily oral bicalutamide (50 mg). All enrolled patients completed three cycles of NHT prior to surgery. Radical prostatectomy (RP) with standard pelvic lymph node dissection (PLND) was performed within 3–4 weeks after completing NHT. Postoperative pathological assessment served as the gold standard for evaluating treatment response. Pathological complete response (pCR) was characterized by features such as reduced glandular volume, decreased glandular density, increased periglandular stromal density, and near-complete degeneration of cancer cells. Minimal residual disease (MRD) was defined as residual tumor with a maximum cross-sectional area < 5 mm, while significant residual disease (SRD) was defined as residual tumor with a maximum cross-sectional area ≥ 5 mm. For the purpose of this study, both pCR and MRD were categorized as a complete response (CR), whereas SRD was categorized as a partial response (PR) [7, 18]. 2.6 Follow-up and Clinical Endpoints Patients were regularly followed up through outpatient clinic visits and telephone consultations. During follow-up, prostate-specific antigen (PSA) levels were monitored, and imaging studies—including computed tomography (CT) scans of the chest, abdomen, and pelvis, as well as pelvic magnetic resonance imaging (MRI) and bone scans—were performed when clinically indicated. The definitions of castration-resistant prostate cancer (CRPC) and the postoperative PSA monitoring protocol adhered to the 2021 European Association of Urology (EAU) guidelines. Recurrence-free survival (RFS) was defined as the interval from the date of radical prostatectomy (RP) to the development of CRPC.CRPC was defined as disease progression in patients with prostate cancer despite ongoing androgen deprivation therapy (ADT) that maintained serum testosterone at castrate levels (<50 ng/dL or <1.7 nmol/L). Disease progression was determined by either of the following criteria: 1. PSA progression: Serum PSA was measured at 1-week intervals for three consecutive tests. PSA progression was defined as a confirmed rise in PSA exceeding 50% above the nadir (lowest value reached) and reaching an absolute PSA level of at least 2 ng/mL. 2. Radiographic progression: Defined as the appearance of new lesions, including the detection of at least two new bone metastases on a bone scan, or new soft-tissue lesions according to the Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. Postoperatively, serum PSA was first measured at 6 to 8 weeks, then every 3 months during the first year, every 6 months during the second year, and annually thereafter. 2.7 Radiomics Feature Extraction and Selection Radiomics feature extraction was performed using Python (version 3.7.3) and the PyRadiomics package (version 3.0). Features were extracted from the original apparent diffusion coefficient (ADC) images and from images processed with two filter types: Laplacian of Gaussian (LoG) and wavelet transforms. The extracted feature categories included: (1) shape-based features, (2) first-order statistical features, (3) gray-level run length matrix (GLRLM) features, (4) gray-level co-occurrence matrix (GLCM) features, (5) gray-level dependence matrix (GLDM) features, (6) gray-level size zone matrix (GLSZM) features, and (7) neighboring gray-tone difference matrix (NGTDM) features. Since all MRI images were acquired from a single center, preprocessing included resampling the ADC images to an isotropic voxel resolution of 1×1×1 mm³ using cubic B-spline interpolation. The gray-level intensities of the ADC images were discretized into 5 bins. Features were subsequently extracted from the three-dimensional (3D) tumor segmentations across all patients. Feature selection was conducted on the training dataset (TD). To eliminate scale differences and ensure comparability, all features were normalized using Z-score transformation. Features with low reproducibility were excluded from further analysis. Inter- and intra-observer reproducibility was assessed using the intraclass correlation coefficient (ICC), calculated with the R psych package (version 2.4.3). Radiomics features with an ICC ≥ 0.75 were considered reliable and retained for subsequent analysis. The feature selection process comprised three stages: 1. Univariate screening: Features associated with complete response (CR) were identified using the Mann-Whitney U test (p < 0.05). 2. Redundancy reduction: Highly correlated features (Spearman’s |r| ≥ 0.90) were identified. For any pair of highly correlated features, the one with the larger mean absolute correlation with all other features was removed. 3. Relevance selection: The Boruta algorithm was applied to identify and retain the most informative and non-redundant features [22]. The final set of selected radiomics features was used for model construction. 2.8 Model Evaluation and Survival Analysis Five distinct models were constructed: one clinical-MRI model, three radiomics models, and one combined model. First, the association between clinical and MRI features and the status of complete response (CR) was evaluated using univariable logistic regression. Features significantly associated with CR were subsequently included in a multivariable logistic regression to build the clinical-MRI model. Three radiomics prediction models were then developed using the optimal set of eight radiomics features with different classifiers: Logistic Regression (LR), Random Forest (RF), and Support Vector Machine (SVM). For the SVM classifier, a radial basis function (RBF) was employed as the kernel. Model training utilized 10-fold cross-validation with 5 repetitions to identify the best-performing classifier; the optimal parameters for SVM were determined to be C=12 and gamma=0.0012. The radiomics signature derived from the best-performing classifier was defined as a new composite feature. This signature was then integrated with the significant clinical-MRI features to construct the final combined model. All models were developed using features selected exclusively from the training set (TD) and subsequently validated on the independent validation set (IVD). The predictive performance of the five models was evaluated on the validation dataset. Receiver operating characteristic (ROC) curves were generated. The optimal cut-off point, determined by maximizing Youden’s index in the TD, was applied to the IVD. DeLong’s test was used to assess the statistical significance of differences between ROC curves. The area under the curve (AUC), sensitivity, specificity, and accuracy were calculated. The 95% confidence intervals (CIs) for the AUCs were estimated using the bootstrap resampling method with 1000 repetitions. Decision curve analysis (DCA) was performed to evaluate the clinical net benefit of the models across a range of threshold probabilities. Furthermore, to explore the prognostic utility of the combined model, patients were stratified into low- and high-probability CR groups based on the optimal cut-off value determined by the maximum selection rank statistics method for this model. Kaplan-Meier survival analysis was then conducted, and the log-rank test was used to compare disease-free survival (DFS) between the two probability groups in both the TD and the IVD. 2.9 Statistical Analysis All statistical analyses were performed using SPSS software (version 25.0) and Jupyter Notebook (version 2.15.0). Differences in clinical and MRI features between groups or datasets were compared using Fisher’s exact test or the Chi-square test for categorical variables, and the independent t-test or the Mann-Whitney U test for continuous variables, as appropriate. A two-sided p-value < 0.05 was considered statistically significant. To ensure high-standard reporting and scientific rigor for this radiomics study, the Radiomics Quality Score (RQS) — a checklist developed based on expert consensus by Lambin et al. — was applied [23]. 3 RESULTS Patient Baseline Characteristics A total of 140 patients were included in this study and categorized into a complete response (CR) group (n=44) and an incomplete response (IR) group (n=96) based on their treatment response. To ensure a balanced distribution of patients from both response groups between the training and validation sets, stratified random sampling was employed. Patients from each subgroup were randomly allocated to the training set and the internal validation set at a ratio of 7:3. Consequently, 31 patients (70.5%) from the CR group and 67 patients (69.8%) from the IR group were assigned to the training set, while 13 (29.5%) and 29 (30.2%) patients from the respective groups were allocated to the internal validation set. This resulted in a final cohort of 98 patients in the training dataset (TD) and 42 patients in the internal validation dataset (IVD). Figure 1-1 illustrates the patient recruitment pathway. The CR rates were 31.6% (31 of 98) in the TD and 31.0% (13 of 42) in the IVD. Figure 1-2 depicts the study flowchart. Table 1 summarizes the clinical and MRI characteristics of all enrolled patients with high-risk prostate cancer. Table 1. Clinical and MRI characteristics of patients with prostate cancer. Characteristic TD IVD CR PR P CR PR P Age 70.06 ±6.01 71.63±6.10 0.239 70.46 士6.41 68.52 + 6.53 0.375 PSA（＜20 vs ≥ 20 ng/ml） 21/31(67.7%) 48/67(71.6%) 0.877 4/13(30.8%) 21/29(72.4%) 0.028 Tumor size（＜1.5cm vs ≥1.5cm） 17/31(54.8%) 43/67(64.2%) 0.509 5/13(38.5%) 18/29(62.1%) 0.278 SVI 5/31(16.1%) 24/67(35.8%) 0.080 0/13(0.0%) 13/29(44.8%) 0.011 ECE 16/31(51.6%) 39/67(58.2%) 0.694 0/13(0.0%) 24/29(82.8%) <0.001 Bone metastasis 8/31(25.8%) 16/67(23.9%) 1.000 0/13(0.0%) 7/29(24.1%) 0.136 Cribriform 1/31 (3.2%) 30/67(44.8%) <0.001 0/13(0.0%) 4/29(13.8%) 0.401 Ductal 1/31 (3.2%) 11/67(16.4%) 0.128 0/13(0.0%) 4/29(13.8%) 0.401 ERG 3/31(9.7%) 5/67(7.5%) 1.000 0/13(0.0%) 5/29(17.2%) 0.280 Abbreviations: PSA, prostate-specific antigen; SVI, seminal vesicle invasion; CR, complete response; PR, partial response; ECE, extracapsular extension; TD, training dataset; IVD, internal validation dataset. Within the training set, positive MRI-detected seminal vesicle invasion (SVI) was more frequently observed in the partial response (PR) group, showing a trend toward significance (16.1% vs. 35.8%, p = 0.080). Patients with cribriform adenocarcinoma on pathology were significantly more likely to have a PR (1/31 [3.2%] vs. 30/67 [44.8%], P 20 ng/mL (4/13 [30.8%] vs. 21/29 [72.4%], p = 0.028), SVI (0/13 [0.0%] vs. 13/29 [44.8%], p = 0.011), or extracapsular extension (ECE) (0/13 [0.0%] vs. 24/29 [82.8%], p < 0.001) were more prone to exhibit a PR. Furthermore, no statistically significant differences were observed between the PR and complete response (CR) groups in either dataset regarding tumor size, age, bone metastasis status, presence of ductal adenocarcinoma, or ERG rearrangement status (all p > 0.05). Performance of the Clinical-MRI Model In the TD, univariable Cox regression analysis revealed that cribriform adenocarcinoma and MRI-detected seminal vesicle invasion (SVI) were significantly associated with PR. No statistically significant differences were found between the groups regarding age, PSA level, ERG positivity, tumor size, bone metastasis status, or the presence of ductal adenocarcinoma. Subsequent stepwise multivariable analysis identified cribriform adenocarcinoma (HR = 0.021, 95% CI: 0.002–0.228; p = 0.004) and SVI (HR = 0.132, 95% CI: 0.023–0.749; p = 0.034) as independent predictors. Consequently, a clinical-MRI model for predicting CR was constructed utilizing cribriform adenocarcinoma and SVI (Table 2). The model’s performance was further evaluated using ROC curve analysis, yielding an AUC of 0.723 in the TD and 0.784 in the IVD. Table 2. Logistic regression analysis for predicting complete response (CR) in the training set. Characteristic Univariate OR(95%CI) p Multivariate OR(95%CI) p Age 0.987(0.932-1.045) 0.656 / / PSA（＜20 vs ≥ 20ng/ml） 0.515(0.245-1.084) 0.08 0.648(0.263-1.595) 0.345 Tumor size（＜1.5cm vs ≥1.5cm） 0.574(0.279-1.182) 0.132 / / SVI 0.204(0.074-0.566) 0.002 0.251(0.070-0.899) 0.034 ECE 0.299(0.142-0.630) 0.002 0.956(0.338-2.698) 0.932 Bone metastasis 0.705(0.287-1.731) 0.446 / / Cribriform 0.629(0.164-2.410) 0.002 0.049(0.006-0.386) 0.004 Ductal 0.126(0.016-0.983) 0.048 0.143(0.016-1.298) 0.084 ERG 0.629(0.164-2.410) 0.499 / / Construction and Validation of the Radiomics Model Based on inter- and intra-observer assessments, 807 out of 851 radiomics features demonstrated intraclass correlation coefficients (ICC) > 0.75. Features with low reproducibility (ICC < 0.75 for either intra- or inter-observer analysis) were excluded. The features with ICC < 0.75 are listed in Table S2. Subsequently, 609 features showed significant differences via the Mann-Whitney U test. After applying Spearman correlation analysis to remove highly correlated features, 137 features remained. The Boruta algorithm was then used for final feature selection, resulting in 8 key features. These 8 final radiomics features were: 1. wavelet-HLL_firstorder_Mean; 2. wavelet-LHL_firstorder_Maximum; 3. wavelet-HHL_glszm_SizeZoneNonUniformity; 4. wavelet-HHH_firstorder_Maximum; 5. wavelet-LHH_glrlm_HighGrayLevelRunEmphasis; 6. wavelet-LHL_glszm_SizeZoneNonUniformity; 7. wavelet-LHH_glcm_Autocorrelation; 8. wavelet-HLH_firstorder_Median. A radiomics model was built based on these features. All 8 features showed significant differences between the PR and CR groups (all p-values < 0.05) and were used for subsequent model construction (Figure 2-1). The importance of the selected features is shown in Figure 2-2, and their correlation matrix is depicted in Figure 2-3. Finally, radiomics models were developed using logistic regression (LR), random forest (RF), and support vector machine (SVM) algorithms based on this specific feature set. The performance of the three radiomics models is shown in Table 3. In the training set, the logistic regression (LR) model achieved an AUC of 0.836 (95% CI: 0.758-0.914), the random forest (RF) model an AUC of 0.866 (95% CI: 0.796-0.936), and the support vector machine (SVM) model an AUC of 0.872 (95% CI: 0.804-0.940). The radiomics model constructed using SVM demonstrated the best performance. In the validation set, among the three classifiers, the SVM-based radiomics model also performed best in predicting CR, with an AUC of 0.923 (95% CI: 0.882-1.012) ( Figure 3 ). Table 3 summarizes the predictive performance of the different models. Training Dataset Validation Dataset Model AUC 95% CI AUC 95% CI Clinical 0.723 (0.622-0.825) 0.784 (0.644-0.923) LR 0.836 (0.758-0.914) 0.899 (0.807-0.992) RF 0.866 (0.796-0.936) 0.850 (0.735-0.965) SVM 0.872 (0.804-0.940) 0.923 (0.843-1.003) COMB 0.909 (0.852-0.965) 0.947 (0.882-1.012) Performance of the Combined Model in Predicting Complete Response and Prognostic Assessment A combined model was constructed by integrating the clinical model (based on the presence of cribriform adenocarcinoma and MRI-detected seminal vesicle invasion) with the radiomics signature generated by the SVM algorithm. For predicting CR, the combined model achieved an area under the curve (AUC) of 0.909 (95% CI: 0.852–0.965) in the training set and 0.947 (95% CI: 0.882–1.012) in the validation set. Decision curve analysis (DCA), a statistical method to evaluate a model's utility in facilitating clinical decision-making, demonstrates clinical value only when the model's net benefit exceeds that of the "treat all" and "treat none" strategies across a range of threshold probabilities. In our study, the net benefit curve of the combined model was higher than both the "treat all" and "treat none" curves when the threshold probability ranged from 0.35 to 1.0 in the training cohort and from 0.47 to 1.0 in the validation cohort ( Figure 4 ). This indicates that the combined model provided higher net benefit within these threshold ranges, suggesting its potential to support clinical decision-making at higher risk thresholds. Collectively, these results indicate that the combined model exhibits high predictive performance for NHT efficacy and demonstrates favorable clinical utility. DFS Follow-up Duration Statistics: In the training set, the median follow-up time was 34.0 months (range: 1.0-101.0 months) for the complete response (CR) group and 19.0 months (range: 0.2-85.0 months) for the partial response (PR) group. In the validation set, the median follow-up time was 33.0 months (range: 8.0-53.0 months) for the CR group and 15.0 months (range: 1.0-56.0 months) for the PR group（ Table 4 ）. Kaplan-Meier survival curves for both datasets revealed a significant difference in disease-free survival (DFS) between the CR and PR groups (all p < 0.05) ( Figure 5 ). While the number of patients progressing to castration-resistant prostate cancer (CRPC) showed no significant difference between the CR and PR groups when analyzing the training and validation sets separately, a pooled analysis of both datasets demonstrated a statistically significant difference in CRPC progression between the two groups (p = 0.038). Patients in the CR group had significantly longer DFS than those in the PR group (p < 0.05), further indicating that the treatment response following neoadjuvant therapy holds predictive value for patient prognosis. Table 4 . Prognostic disparities between the training and validation datasets. Training Dataset Validation Dataset CR PR P CR PR P CRPC-NO 27 50 12 20 CRPC-YES 4 17 0.2566 1 9 0.2113 DFS-Media 34 19 33 15 DFS-Range 1-101 0.2-85 0.042 8-53 1-56 0.017 4 DISCUSSION Our study demonstrates the value of radiomics in predicting the response to neoadjuvant hormonal therapy (NHT) in patients with high-risk prostate cancer (PCa). We further found that an MRI-based radiomics model outperformed the clinical-MRI model and could accurately predict treatment response. In this study, the combined model, developed using a Support Vector Machine (SVM) algorithm and incorporating clinical MRI features, showed the best performance in predicting NHT response. Compared to the clinical-MRI model and the individual radiomics models in both the training and internal validation datasets, the combined model exhibited a higher area under the curve (AUC) in ROC analysis, indicating superior predictive efficacy. Prostate-specific antigen (PSA) is currently a commonly used tool for assessing NHT outcomes and treatment response in PCa patients. However, its most significant limitation is its low specificity, as PSA is an organ-specific rather than a tumor-specific biomarker [17]. In this study, PSA was categorized as a binary variable (<20 ng/mL vs. ≥20 ng/mL), and no significant difference was observed between patients with different treatment responses. Ke ZB, Chen SM, et al. utilized [68Ga]Ga-PSMA-11 PET/CT to predict treatment response after NHT in high-risk non-metastatic PCa patients. Their multivariate logistic regression analysis indicated that the maximum standardized uptake value (SUVmax) after neoadjuvant combined hormonal therapy (NCHT) prior to radical prostatectomy was an independent predictor of favorable pathological response [10]. However, their study had a limited sample size and did not incorporate radiomics analysis. Wu XH et al. previously explored the value of radiomics in predicting NHT response in high-risk PCa patients [16]. However, their radiomics model construction employed only a single feature selection method (LASSO) and did not further investigate the optimal classifier for building the model, and the sample size was under 100 patients. The choice of classifier can impact model performance; however, no consensus exists on the optimal selection. In our study, the SVM model outperformed the RF and LR models in treatment response prediction. The SVM algorithm is particularly adept at identifying subtle patterns in complex datasets due to its ability to minimize classification errors on unseen data without requiring prior assumptions about the data's probability distribution [14]. Given the complexity and non-linearity between radiomic features and tumor response, this grants the algorithm an advantage in modeling moderately non-linear relationships [15]. Nonetheless, the efficacy of different classifiers may vary across clinical scenarios. Future studies exploring the optimal classifier for specific clinical applications are warranted. Our study also further analyzed the prognostic value of the combined model. Kaplan-Meier survival curves in both datasets revealed a significant difference in disease-free survival (DFS) between the complete response (CR) group and the incomplete response (non-CR) group (both p < 0.05). The CR group demonstrated longer DFS compared to the non-CR group (p < 0.05), further indicating that different post-neoadjuvant therapy responses can predict patient outcomes after surgery. This study has several limitations. Manual segmentation of regions of interest (ROIs) is a time-consuming process requiring accurate identification of MRI lesions, which can be challenging for clinicians lacking experience in interpreting prostate MRI. Moreover, ROI delineation varies between physicians, highlighting a lack of standardized quality control. Therefore, developing an automated or semi-automated tool to optimize the ROI segmentation process is necessary. Additionally, prospective studies are needed to validate our findings. 5 CONCLUSIONS In summary, based on a large sample size, conventional imaging sequences, and multi-dataset validation, this study constructed a robust and generalizable radiomics model for predicting complete treatment response, showing potential value for prognostic risk stratification. Adhering strictly to radiomics quality control standards, this study achieved a Radiomics Quality Score (RQS) of 20 points(Table S3), confirming the rigor of the methodology and the reliability of the conclusions. The proposed predictive model holds promise for clinical application. Declarations Acknowledgements Not applicable. Funding Not applicable. Author information Hai Zhou mainly contributed to this work. Authors and Affiliations Department of Urology,Fudan University affiliated Huadong hospital, China, shanghai Hai Zhou, Zhi-Yu Qian, Zhi-Hao Chen, Jian-Jin Fan, Yong-Xin Mao ,Le-Yan Xu, Yi-Jun He, Jin-Xiong Zhang ,Jian- Hong Wu, Lu Sheng Contributions Hai Zhou : Formal analysis ; methodology ; visualization ; writing – original draft . Zhi-Yu Qian : Writing – original draft . Zhi-Hao Chen: Formal analysis . Jian-Jin Fan , Yong-Xin Mao , Le-Yan Xu: Data curation . Yi-Jun He , Jin-Xiong Zhang, Jian- Hong Wu : Conceptualization. Lu Sheng: Project administration.( All authors reviewed the manuscript.) Corresponding authors Correspondence to Lu Sheng Ethics declarations Ethics approval and consent to participate This study was approved by the Ethics Committee of Fudan University affiliated Huadong hospital , and all patients provided written informed consent. Consent for publication Not applicable. Competing interests The authors declare no competing interests. DATA AVAILABILITY STATEMENT The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. References Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2021. CA Cancer J Clin. 2021;71:7–33. Chang AJ, Autio KA, Roach M 3rd, Scher HI. High-risk prostate cancer-classification and therapy. Nat Rev Clin Oncol. 2014;11:308_23. Pan J, Chi C, Qian H, Zhu Y, Shao X, Sha J, et al. Neoadjuvant chemohormonal therapy combined with radical prostatectomy and extended PLND for very high risk locally advanced prostate cancer: a retrospective comparative study. Urol Oncol. 2019;37:991–8. Devos G, Devlies W, De Meerleer G, et al. Neoadjuvant hormonal therapy before radical prostatectomy in high-risk prostate cancer. Nat Rev Urol . 2021; 18 (12): 739-762. Mottet N, van den Bergh R, Briers E. European association of urology guidelines 2019 Arnhem European Association of Urology Office:196–299 Shelley MD, Kumar S, Wilt T, Staffurth J, Coles B, et al A systematic review and meta-analysis of randomised trials of neo-adjuvant hormone therapy for localised and locally advanced prostate carcinoma Cancer Treat Rev. 2009;35:9–17 Pignot G, Maillet D, Gross E, Barthelemy P, Beauval JB, et al Systemic treatments for high-risk localized prostate cancer Nat Rev Urol. 2018;15:498–510 Liu W, Yao Y, Liu X, Liu Y, Zhang G-M. Neoadjuvant hormone therapy for patients with high-risk prostate cancer: a systematic review and meta-analysis. Asian J Androl . 2021; 23 (4): 429-436. Dolezel M, Odrazka K, Vanasek J, et al. Neoadjuvant hormonal therapy in prostate cancer—impact of PSA level before radiotherapy. J BUON . 2013; 18 (4): 949-953. Ke Z-B, Chen S-M, Chen J-Y, et al. Head-to-head comparisons of [68Ga]Ga-PSMA-11 PET/CT, multiparametric MRI, and prostate-specific antigen for the evaluation of therapeutic responses to neoadjuvant chemohormonal therapy in high-risk non-metastatic prostate cancer patients: a prospective study. Eur J Nucl Med Mol Imaging . 2023; 50 (4): 1240-1251. Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol . 2017; 14 (12): 749-762. Hu T, Gong J, Sun Y, Li M, Cai C, Li X, Cui Y, Zhang X, Tong T. Magnetic resonance imaging-based radiomics analysis for prediction of treatment response to neoadjuvant chemoradiotherapy and clinical outcome in patients with locally advanced rectal cancer: A large multicentric and validated study. MedComm (2020). 2024 Jun 20;5(7):e609. Abdollahi H, Mofid B, Shiri I, et al. Machine learning-based radiomic models to predict intensity-modulated radiation therapy response, Gleason score and stage in prostate cancer. Radiol Med . 2019; 124 (6): 555-567. Huang S, Cai N, Pacheco PP, Narrandes S, Wang Y, Xu W. Applications of support vector machine (SVM) learning in cancer genomics. Cancer Genomics Proteomics . 2018; 15 (1): 41-51. Emblem KE, Pinho MC, Zollner FG, et al. A generic support vector machine model for preoperative glioma survival associations. Radiology . 2015; 275 (1): 228-234. Wu XH, Ruan ZT, Ke ZB, Lin F, Chen JY, Xue YT, Lin B, Chen SH, Chen DN, Zheng QS, Xue XY, Wei Y, Xu N. Magnetic resonance imaging-based radiomics nomogram for the evaluation of therapeutic responses to neoadjuvant chemohormonal therapy in high-risk non-metastatic prostate cancer. Cancer Med. 2024 Jul;13(14):e70001. Filella X, Foj L. Emerging biomarkers in the detection and prognosis of prostate cancer. Clin Chem Lab Med. 2015;53:963–73. Wang Z, Wang J, Li D, Wu R, Huang J, Ye L, Tuo Z, Yu Q, Shao F, Wusiman D, Cho WC, Koh SB, Xiong W, Feng D. Novel hormone therapies for advanced prostate cancer: Understanding and countering drug resistance. J Pharm Anal. 2025 Sep;15(9):101232. Petraki CD, Sfikas CP. Histopathological changes induced by therapies in the benign prostate and prostate adenocarcinoma. Histol Histopathol . 2007; 22 (1): 107-118. Thalgott M, Horn T, Heck MM, Maurer T, Eiber M, Retz M, et al. Long-term results of a phase II study with neoadjuvant docetaxel chemotherapy and complete androgen blockade in locally advanced and high-risk prostate cancer. J Hematol Oncol. 2014;7:20. Cheng Q, Butler W, Zhou Y, Zhang H, Tang L, Perkinson K, Chen X, Jiang XS, McCall SJ, Inman BA, Huang J. Pre-existing Castration-resistant Prostate Cancer-like Cells in Primary Prostate Cancer Promote Resistance to Hormonal Therapy. Eur Urol. 2022 May;81(5):446-455. Miron B, Kursa WRR. Feature selection with the boruta package. J Statist Software . 2010; 36 (11): 13. Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol . 2017; 14 (12): 749-762. Additional Declarations No competing interests reported. Supplementary Files SupplementaryFigureandtable.docx Cite Share Download PDF Status: Published Journal Publication published 29 Apr, 2026 Read the published version in BMC Cancer → Version 1 posted Editorial decision: Revision requested 02 Mar, 2026 Reviews received at journal 01 Mar, 2026 Reviews received at journal 21 Feb, 2026 Reviewers agreed at journal 19 Feb, 2026 Reviewers agreed at journal 10 Feb, 2026 Reviewers invited by journal 10 Feb, 2026 Editor invited by journal 30 Jan, 2026 Editor assigned by journal 26 Jan, 2026 Submission checks completed at journal 26 Jan, 2026 First submitted to journal 20 Jan, 2026 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-8655062","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":590186395,"identity":"985a7b87-0c27-4f9b-aa05-bff8e9adf2e7","order_by":0,"name":"Hai Zhou","email":"","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":false,"prefix":"","firstName":"Hai","middleName":"","lastName":"Zhou","suffix":""},{"id":590186396,"identity":"b431335c-c52f-46d9-a99f-2d62b58c3367","order_by":1,"name":"Zhi-Yu Qian","email":"","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":false,"prefix":"","firstName":"Zhi-Yu","middleName":"","lastName":"Qian","suffix":""},{"id":590186397,"identity":"5bb9c579-5e58-4052-8c58-fe4a6e7d7c21","order_by":2,"name":"Zhi-Hao Chen","email":"","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":false,"prefix":"","firstName":"Zhi-Hao","middleName":"","lastName":"Chen","suffix":""},{"id":590186400,"identity":"5e54ce9a-3534-4f66-9806-5a386f228e6d","order_by":3,"name":"Jian-Jin Fan","email":"","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":false,"prefix":"","firstName":"Jian-Jin","middleName":"","lastName":"Fan","suffix":""},{"id":590186401,"identity":"334319d8-26d0-4eca-9eac-21d52c406d6d","order_by":4,"name":"Yong-Xin Mao","email":"","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":false,"prefix":"","firstName":"Yong-Xin","middleName":"","lastName":"Mao","suffix":""},{"id":590186403,"identity":"2574b15b-7cec-49bc-a70f-98cee0863eb8","order_by":5,"name":"Le-Yan Xu","email":"","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":false,"prefix":"","firstName":"Le-Yan","middleName":"","lastName":"Xu","suffix":""},{"id":590186405,"identity":"26c6b2d4-bf7f-4a8f-9de2-430b5aff8715","order_by":6,"name":"Yi-Jun He","email":"","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":false,"prefix":"","firstName":"Yi-Jun","middleName":"","lastName":"He","suffix":""},{"id":590186406,"identity":"504f9d9b-ffb0-4094-9a4d-37dd34eee390","order_by":7,"name":"Jin-Xiong Zhang","email":"","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":false,"prefix":"","firstName":"Jin-Xiong","middleName":"","lastName":"Zhang","suffix":""},{"id":590186407,"identity":"0aca8b36-a108-4a6c-8318-70f66fa501fc","order_by":8,"name":"Jian- Hong Wu","email":"","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":false,"prefix":"","firstName":"Jian-","middleName":"Hong","lastName":"Wu","suffix":""},{"id":590186408,"identity":"3160cafd-308f-43b9-bfc7-4a96691d29d7","order_by":9,"name":"Lu Sheng","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2ElEQVRIiWNgGAWjYBACfvb+hw8+VEjIybM3HyBOi2TPGWbDGWdsjA17jiUQp8Xghg+bNG9LWiLDjRwDIl12g/ewAW/D4QTGhpyPN94w2MnpNhDQwTi7L/GB5I7DeewMZzdbzmFINjY7QEALs8wBYwPDM4eLGRt7t0nzMBxI3EZIC5tEgplEYtvhxIbDPM+I08IjkWMmcbAtLbHhGA8bcVokeI4lGzaAA5nN2HKOARF+sT/efPDxH1BUyj9+eONNhZ0cQS1oVhIbNUhaSNUxCkbBKBgFIwIAAK2zR/USFRDRAAAAAElFTkSuQmCC","orcid":"","institution":"Fudan University affiliated Huadong hospital","correspondingAuthor":true,"prefix":"","firstName":"Lu","middleName":"","lastName":"Sheng","suffix":""}],"badges":[],"createdAt":"2026-01-21 04:38:57","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8655062/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8655062/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12885-026-16015-0","type":"published","date":"2026-04-29T15:57:37+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":102621482,"identity":"b1467013-7b2c-4335-8f5a-f6dd259b82ec","added_by":"auto","created_at":"2026-02-13 16:51:06","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":192977,"visible":true,"origin":"","legend":"\u003cp\u003eFlow diagram of this study.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/b920bf360ca18109d5f198b6.png"},{"id":102621423,"identity":"90371b62-72a4-46ec-a316-56d12d30374c","added_by":"auto","created_at":"2026-02-13 16:51:03","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":243536,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFIGURE 1-2\u003c/strong\u003e Workflow of this study.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/661d0faec94535f200e5c74d.png"},{"id":102621418,"identity":"af4c016a-8b86-47c0-95b0-499e65ccb4c0","added_by":"auto","created_at":"2026-02-13 16:51:03","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":127084,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 2-1\u003c/strong\u003e​displays box plots of the eight selected radiomics features, all showing significant differences between the CR and PR groups in the training set.\u003c/p\u003e","description":"","filename":"21.png","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/7e01d13b622c23f5ca12fb72.png"},{"id":102621416,"identity":"be6068dc-fd6a-475f-9d3d-bca96a1d29c4","added_by":"auto","created_at":"2026-02-13 16:51:03","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":133504,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 2-2. The importance of the selected 8 features with the most predictive value\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"22.png","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/8ade2e7f0557dac6929877fe.png"},{"id":102621422,"identity":"8c605938-1538-438f-9ab5-f119344dccf2","added_by":"auto","created_at":"2026-02-13 16:51:03","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":173987,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 2-3\u003c/strong\u003e. The correlation matrix of the selected features.\u003c/p\u003e","description":"","filename":"23.png","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/9f5d257168a624e0bfcdf4eb.png"},{"id":102621417,"identity":"73c09698-d33b-4f9f-8555-42505bcfa950","added_by":"auto","created_at":"2026-02-13 16:51:03","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":117952,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 3. Receiver operating characteristic (ROC) curves of the five models for predicting complete response.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAbbreviations:\u003c/strong\u003e​AUC, area under the curve; COMB, combined model; LR, logistic regression; RF, random forest; SVM, support vector machine.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/c3a8e4bb75cf01509094aa79.png"},{"id":102621419,"identity":"78111525-efbd-4443-ba37-cb09a4be5057","added_by":"auto","created_at":"2026-02-13 16:51:03","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":124005,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 4. Decision curve analysis of the five models for predicting complete response.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/64e1e82e7e221f08c0144fd5.png"},{"id":102621424,"identity":"76561d86-9318-4954-8e16-b59f467975fc","added_by":"auto","created_at":"2026-02-13 16:51:03","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":67249,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 5.\u003c/strong\u003e Kaplan-Meier survival curve analysis for complete response (CR) and partial response (PR) in the training and validation sets.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/6259bd02663ea3d6dd750eb5.png"},{"id":108437606,"identity":"9eadf4ab-cdf8-4a0c-997b-f64499596261","added_by":"auto","created_at":"2026-05-04 16:00:15","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1389134,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/06430392-6e7a-44ef-afec-3976efbacd15.pdf"},{"id":102621421,"identity":"6878d030-f68b-45a8-9441-61f488ff01a3","added_by":"auto","created_at":"2026-02-13 16:51:03","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":908798,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryFigureandtable.docx","url":"https://assets-eu.researchsquare.com/files/rs-8655062/v1/4b2acf9e77753f632a833b1b.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Predicting Response to Neoadjuvant Hormonal Therapy and Prognostic Outcomes in High-Risk Prostate Cancer Using MRI-Based Radiomics","fulltext":[{"header":"1 INTRODUCTION","content":"\u003cp\u003eProstate cancer (PCa) is the most common malignancy and the second leading cause of cancer-related death among men worldwide [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. High-risk prostate cancer (HRPCa) represents a risk category associated with a high probability of disease progression or recurrence [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. In China, the majority of PCa cases are diagnosed at advanced stages, with a particularly high prevalence of the high-risk subgroup [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Accounting for 15%\u0026ndash;20% of clinically localized PCa cases, HRPCa is characterized by an increased likelihood of biochemical recurrence (BCR), metastatic progression, and cancer-specific mortality [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Despite ongoing clinical efforts, a consensus on the optimal treatment strategy for men with HRPCa remains elusive [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Consequently, novel therapeutic strategies, including multimodal approaches, are needed.\u003c/p\u003e \u003cp\u003eNeoadjuvant hormonal therapy (NHT) combined with radical prostatectomy (RP) or radiotherapy (RT) may improve outcomes in PCa [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Specifically, neoadjuvant androgen deprivation therapy (ADT) prior to RP, compared to RP alone, can reduce rates of pathological T3 stage (downstaging), positive surgical margins, and lymph node invasion [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Although NHT before RP has demonstrated significant improvements in pathological outcomes, it has not yet been shown to confer a definitive survival benefit for patients with high-risk disease [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAccurately identifying patients who may respond to NHT is clinically important. Previous studies have proposed several strategies for this purpose, such as monitoring prostate-specific antigen (PSA) kinetics [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] or using prostate-specific membrane antigen positron emission tomography/computed tomography (PSMA PET/CT) [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. However, these approaches are largely based on a single modality, lack robust quantifiable metrics, and have demonstrated limited reliability and accuracy.\u003c/p\u003e \u003cp\u003eRadiomics, which involves the high-throughput extraction of quantitative features from medical images, holds promise for providing non-invasive insights into tumor heterogeneity and underlying pathophysiology [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Prior research has indicated the potential value of radiomics features derived from multiparametric magnetic resonance imaging (mpMRI) in predicting response to NHT and in assessing treatment response across various cancers, including breast, colorectal, and prostate cancer [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Nevertheless, in the context of PCa, existing radiomics studies are often limited by small sample sizes and the absence of independent validation cohorts.\u003c/p\u003e \u003cp\u003eTo address these limitations, we conducted this study utilizing a larger sample size with an internal validation cohort. Our objectives are: (1) to develop and evaluate the efficacy of an mpMRI-based radiomics model, employing different classifiers, for predicting a complete pathological response (pCR) to NHT in patients with HRPCa prior to treatment; and (2) to construct and validate a combined model that integrates clinical MRI features with the optimal radiomics signature to improve predictive performance and explore its potential prognostic value.\u003c/p\u003e"},{"header":"2 MATERIALS AND METHODS","content":"\u003cp\u003e2.1 Patient Cohort\u003c/p\u003e\n\u003cp\u003eThis retrospective study was approved by the Ethics Committee of Huadong Hospital Affiliated with Fudan University, and written informed consent was obtained from all participants. We enrolled 140 patients diagnosed with high-risk prostate cancer (PCa) who received neoadjuvant hormonal therapy (NHT) followed by radical prostatectomy (RP) at our center between January 2017 and January 2025. All patients had a pathological diagnosis of prostatic adenocarcinoma via prostate biopsy prior to NHT initiation. PCa staging was based on the 2017 TNM classification. According to the 2021 European Association of Urology (EAU) guidelines, high-risk PCa encompasses locally high-risk disease (PSA \u0026gt;20 ng/mL, ISUP grade group 4 or 5, or clinical stage cT2c) and locally advanced disease (clinical stage cT3\u0026ndash;4 or cN+, irrespective of PSA level and ISUP grade). Inclusion criteria were: (1) biopsy-confirmed high-risk PCa, and (2) availability of prostate multiparametric magnetic resonance imaging (mpMRI) performed prior to NHT initiation. Exclusion criteria were: (1) presence of metastatic PCa at diagnosis, (2) absence of pre-NHT MRI data, and (3) failure to undergo RP (e.g., patients who underwent transurethral resection of the prostate [TURP] instead). After excluding [X] patients meeting the exclusion criteria, 140 eligible subjects with complete clinicopathological data were finally analyzed. They were divided into a training cohort (n=95) and an internal validation cohort (n=45).\u003c/p\u003e\n\u003cp\u003e2.2 Clinical and MRI Feature Assessment\u003c/p\u003e\n\u003cp\u003eClinical characteristics, including patient age and pre-NHT prostate-specific antigen (PSA) level, were retrieved from medical records. MRI morphological features\u0026mdash;tumor size (measured on T2-weighted sagittal images), MRI-detected extracapsular extension (ECE), and seminal vesicle invasion (SVI)\u0026mdash;were independently evaluated by two associate chief urologists (Reader 1 and Reader 2, each with \u0026gt;10 years of experience in prostate MRI interpretation). In cases of disagreement, a third senior urologist (Reader 3, with \u0026gt;20 years of experience) provided a final assessment. All readers were blinded to clinical outcomes and postoperative pathology. The specific criteria for assessing MRI morphological features, along with representative MR images illustrating ECE and SVI, are provided in Figure S1.\u003c/p\u003e\n\u003cp\u003e2.3 MRI Acquisition and Tumor Segmentation\u003c/p\u003e\n\u003cp\u003eAll patients underwent mpMRI on a SIEMENS Verio 3.0 T scanner within two weeks before NHT initiation. Patients were scanned in the supine position with the prostate at the isocenter. They were instructed to empty their bowels and maintain a moderately full bladder prior to imaging. The scanning protocol included axial, sagittal, and coronal T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and the derived apparent diffusion coefficient (ADC) maps. Detailed sequence parameters are listed in Table S1.\u003c/p\u003e\n\u003cp\u003eUsing 3D Slicer software (version 3.4.3), a uroradiologist manually delineated the region of interest (ROI) slice-by-slice along the tumor border on T2WI and high b-value DWI (b = 800 s/mm\u0026sup2;) images, covering the entire tumor volume. The ROIs were then propagated to the corresponding ADC maps. During segmentation on DWI, anatomical reference from T2W images was carefully integrated to ensure accuracy.\u003c/p\u003e\n\u003cp\u003e2.4 Reproducibility Assessment of Radiomics Feature Extraction\u003c/p\u003e\n\u003cp\u003eTo evaluate inter- and intra-observer reproducibility, a random subset of 50 patients was selected. For intra-observer assessment, Reader 1 performed manual ROI segmentation on the MRI scans and repeated the process one month later, blinded to the initial contours. For inter-observer assessment, Reader 2 independently delineated the ROIs for the same patients. Both readers were blinded to pathological and clinical outcomes. The intraclass correlation coefficient (ICC) was calculated for each extracted radiomics feature. Features demonstrating an ICC (inter- or intra-observer) \u0026ge; 0.75 were considered to have good reproducibility and were retained for subsequent analysis.\u003c/p\u003e\n\u003cp\u003e2.5 NHT Regimen and Treatment Response Assessment\u003c/p\u003e\n\u003cp\u003eThe NHT regimen comprised subcutaneous injections of either goserelin or leuprorelin (3.6 mg) every 28 days, combined with daily oral bicalutamide (50 mg). All enrolled patients completed three cycles of NHT prior to surgery. Radical prostatectomy (RP) with standard pelvic lymph node dissection (PLND) was performed within 3\u0026ndash;4 weeks after completing NHT.\u003c/p\u003e\n\u003cp\u003ePostoperative pathological assessment served as the gold standard for evaluating treatment response. Pathological complete response (pCR) was characterized by features such as reduced glandular volume, decreased glandular density, increased periglandular stromal density, and near-complete degeneration of cancer cells. Minimal residual disease (MRD) was defined as residual tumor with a maximum cross-sectional area \u0026lt; 5 mm, while significant residual disease (SRD) was defined as residual tumor with a maximum cross-sectional area \u0026ge; 5 mm. For the purpose of this study, both pCR and MRD were categorized as a complete response (CR), whereas SRD was categorized as a partial response (PR) [7, 18].\u003c/p\u003e\n\u003cp\u003e2.6 Follow-up and Clinical Endpoints\u003c/p\u003e\n\u003cp\u003ePatients were regularly followed up through outpatient clinic visits and telephone consultations. During follow-up, prostate-specific antigen (PSA) levels were monitored, and imaging studies\u0026mdash;including computed tomography (CT) scans of the chest, abdomen, and pelvis, as well as pelvic magnetic resonance imaging (MRI) and bone scans\u0026mdash;were performed when clinically indicated. The definitions of castration-resistant prostate cancer (CRPC) and the postoperative PSA monitoring protocol adhered to the 2021 European Association of Urology (EAU) guidelines. Recurrence-free survival (RFS) was defined as the interval from the date of radical prostatectomy (RP) to the development of CRPC.CRPC was defined as disease progression in patients with prostate cancer despite ongoing androgen deprivation therapy (ADT) that maintained serum testosterone at castrate levels (\u0026lt;50 ng/dL or \u0026lt;1.7 nmol/L). Disease progression was determined by either of the following criteria:\u003c/p\u003e\n\u003cp\u003e1. PSA progression: Serum PSA was measured at 1-week intervals for three consecutive tests. PSA progression was defined as a confirmed rise in PSA exceeding 50% above the nadir (lowest value reached) and reaching an absolute PSA level of at least 2 ng/mL.\u003c/p\u003e\n\u003cp\u003e2. Radiographic progression: Defined as the appearance of new lesions, including the detection of at least two new bone metastases on a bone scan, or new soft-tissue lesions according to the Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1.\u003c/p\u003e\n\u003cp\u003ePostoperatively, serum PSA was first measured at 6 to 8 weeks, then every 3 months during the first year, every 6 months during the second year, and annually thereafter.\u003c/p\u003e\n\u003cp\u003e2.7 Radiomics Feature Extraction and Selection\u003c/p\u003e\n\u003cp\u003eRadiomics feature extraction was performed using Python (version 3.7.3) and the PyRadiomics package (version 3.0). Features were extracted from the original apparent diffusion coefficient (ADC) images and from images processed with two filter types: Laplacian of Gaussian (LoG) and wavelet transforms. The extracted feature categories included: (1) shape-based features, (2) first-order statistical features, (3) gray-level run length matrix (GLRLM) features, (4) gray-level co-occurrence matrix (GLCM) features, (5) gray-level dependence matrix (GLDM) features, (6) gray-level size zone matrix (GLSZM) features, and (7) neighboring gray-tone difference matrix (NGTDM) features.\u003c/p\u003e\n\u003cp\u003eSince all MRI images were acquired from a single center, preprocessing included resampling the ADC images to an isotropic voxel resolution of 1\u0026times;1\u0026times;1 mm\u0026sup3; using cubic B-spline interpolation. The gray-level intensities of the ADC images were discretized into 5 bins. Features were subsequently extracted from the three-dimensional (3D) tumor segmentations across all patients.\u003c/p\u003e\n\u003cp\u003eFeature selection was conducted on the training dataset (TD). To eliminate scale differences and ensure comparability, all features were normalized using Z-score transformation. Features with low reproducibility were excluded from further analysis. Inter- and intra-observer reproducibility was assessed using the intraclass correlation coefficient (ICC), calculated with the R psych package (version 2.4.3). Radiomics features with an ICC \u0026ge; 0.75 were considered reliable and retained for subsequent analysis.\u003c/p\u003e\n\u003cp\u003eThe feature selection process comprised three stages:\u003c/p\u003e\n\u003cp\u003e1. Univariate screening: Features associated with complete response (CR) were identified using the Mann-Whitney U test (p \u0026lt; 0.05).\u003c/p\u003e\n\u003cp\u003e2. Redundancy reduction: Highly correlated features (Spearman\u0026rsquo;s |r| \u0026ge; 0.90) were identified. For any pair of highly correlated features, the one with the larger mean absolute correlation with all other features was removed.\u003c/p\u003e\n\u003cp\u003e3. Relevance selection: The Boruta algorithm was applied to identify and retain the most informative and non-redundant features [22].\u003c/p\u003e\n\u003cp\u003eThe final set of selected radiomics features was used for model construction.\u003c/p\u003e\n\u003cp\u003e2.8 Model Evaluation and Survival Analysis\u003c/p\u003e\n\u003cp\u003eFive distinct models were constructed: one clinical-MRI model, three radiomics models, and one combined model. First, the association between clinical and MRI features and the status of complete response (CR) was evaluated using univariable logistic regression. Features significantly associated with CR were subsequently included in a multivariable logistic regression to build the clinical-MRI model. Three radiomics prediction models were then developed using the optimal set of eight radiomics features with different classifiers: Logistic Regression (LR), Random Forest (RF), and Support Vector Machine (SVM). For the SVM classifier, a radial basis function (RBF) was employed as the kernel. Model training utilized 10-fold cross-validation with 5 repetitions to identify the best-performing classifier; the optimal parameters for SVM were determined to be C=12 and gamma=0.0012. The radiomics signature derived from the best-performing classifier was defined as a new composite feature. This signature was then integrated with the significant clinical-MRI features to construct the final combined model. All models were developed using features selected exclusively from the training set (TD) and subsequently validated on the independent validation set (IVD).\u003c/p\u003e\n\u003cp\u003eThe predictive performance of the five models was evaluated on the validation dataset. Receiver operating characteristic (ROC) curves were generated. The optimal cut-off point, determined by maximizing Youden\u0026rsquo;s index in the TD, was applied to the IVD. DeLong\u0026rsquo;s test was used to assess the statistical significance of differences between ROC curves. The area under the curve (AUC), sensitivity, specificity, and accuracy were calculated. The 95% confidence intervals (CIs) for the AUCs were estimated using the bootstrap resampling method with 1000 repetitions. Decision curve analysis (DCA) was performed to evaluate the clinical net benefit of the models across a range of threshold probabilities.\u003c/p\u003e\n\u003cp\u003eFurthermore, to explore the prognostic utility of the combined model, patients were stratified into low- and high-probability CR groups based on the optimal cut-off value determined by the maximum selection rank statistics method for this model. Kaplan-Meier survival analysis was then conducted, and the log-rank test was used to compare disease-free survival (DFS) between the two probability groups in both the TD and the IVD.\u003c/p\u003e\n\u003cp\u003e2.9 Statistical Analysis\u003c/p\u003e\n\u003cp\u003eAll statistical analyses were performed using SPSS software (version 25.0) and Jupyter Notebook (version 2.15.0). Differences in clinical and MRI features between groups or datasets were compared using Fisher\u0026rsquo;s exact test or the Chi-square test for categorical variables, and the independent t-test or the Mann-Whitney U test for continuous variables, as appropriate. A two-sided p-value \u0026lt; 0.05 was considered statistically significant. To ensure high-standard reporting and scientific rigor for this radiomics study, the Radiomics Quality Score (RQS) \u0026mdash; a checklist developed based on expert consensus by Lambin et al. \u0026mdash; was applied [23].\u003c/p\u003e"},{"header":"3 RESULTS","content":"\u003ch2\u003e\u003cstrong\u003ePatient Baseline Characteristics\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eA total of 140 patients were included in this study and categorized into a complete response (CR) group (n=44) and an incomplete response (IR) group (n=96) based on their treatment response. To ensure a balanced distribution of patients from both response groups between the training and validation sets, stratified random sampling was employed. Patients from each subgroup were randomly allocated to the training set and the internal validation set at a ratio of 7:3. Consequently, 31 patients (70.5%) from the CR group and 67 patients (69.8%) from the IR group were assigned to the training set, while 13 (29.5%) and 29 (30.2%) patients from the respective groups were allocated to the internal validation set. This resulted in a final cohort of 98 patients in the training dataset (TD) and 42 patients in the internal validation dataset (IVD). Figure 1-1 illustrates the patient recruitment pathway. The CR rates were 31.6% (31 of 98) in the TD and 31.0% (13 of 42) in the IVD. Figure 1-2 depicts the study flowchart. Table 1 summarizes the clinical and MRI characteristics of all enrolled patients with high-risk prostate cancer.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1.\u0026nbsp;\u003c/strong\u003eClinical and MRI characteristics of patients with prostate cancer.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCharacteristic\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 190px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTD\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 235px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eIVD\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eCR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003ePR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eCR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003ePR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e70.06 \u0026plusmn;6.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e71.63\u0026plusmn;6.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.239\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e70.46 士6.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e68.52 + 6.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.375\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePSA（＜20 vs \u0026ge; 20 ng/ml）\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e21/31(67.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e48/67(71.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.877\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e4/13(30.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e21/29(72.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.028\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTumor size（＜1.5cm vs \u0026ge;1.5cm）\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e17/31(54.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e43/67(64.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.509\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e5/13(38.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e18/29(62.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.278\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVI\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e5/31(16.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e24/67(35.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.080\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0/13(0.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e13/29(44.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.011\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eECE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e16/31(51.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e39/67(58.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.694\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0/13(0.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e24/29(82.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBone metastasis\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e8/31(25.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e16/67(23.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0/13(0.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e7/29(24.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.136\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCribriform\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1/31 (3.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e30/67(44.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0/13(0.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e4/29(13.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.401\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDuctal\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1/31 (3.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e11/67(16.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0/13(0.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e4/29(13.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e0.401\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 124px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eERG\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e3/31(9.7%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e5/67(7.5%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1.000\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0/13(0.0%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e5/29(17.2%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.280\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eAbbreviations:\u003c/strong\u003e PSA, prostate-specific antigen; SVI, seminal vesicle invasion; CR, complete response; PR, partial response; ECE, extracapsular extension; TD, training dataset; IVD, internal validation dataset.\u003c/p\u003e\n\u003cp\u003eWithin the training set, positive MRI-detected seminal vesicle invasion (SVI) was more frequently observed in the partial response (PR) group, showing a trend toward significance (16.1% vs. 35.8%, p = 0.080). Patients with cribriform adenocarcinoma on pathology were significantly more likely to have a PR (1/31 [3.2%] vs. 30/67 [44.8%], P \u0026lt; 0.001). In the internal validation set, patients with a PSA level \u0026gt;20 ng/mL (4/13 [30.8%] vs. 21/29 [72.4%], p = 0.028), SVI (0/13 [0.0%] vs. 13/29 [44.8%], p = 0.011), or extracapsular extension (ECE) (0/13 [0.0%] vs. 24/29 [82.8%], p \u0026lt; 0.001) were more prone to exhibit a PR. Furthermore, no statistically significant differences were observed between the PR and complete response (CR) groups in either dataset regarding tumor size, age, bone metastasis status, presence of ductal adenocarcinoma, or ERG rearrangement status (all p \u0026gt; 0.05).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003ePerformance of the Clinical-MRI Model\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eIn the TD, univariable Cox regression analysis revealed that cribriform adenocarcinoma and MRI-detected seminal vesicle invasion (SVI) were significantly associated with PR. No statistically significant differences were found between the groups regarding age, PSA level, ERG positivity, tumor size, bone metastasis status, or the presence of ductal adenocarcinoma. Subsequent stepwise multivariable analysis identified cribriform adenocarcinoma (HR = 0.021, 95% CI: 0.002\u0026ndash;0.228; p = 0.004) and SVI (HR = 0.132, 95% CI: 0.023\u0026ndash;0.749; p = 0.034) as independent predictors. Consequently, a clinical-MRI model for predicting CR was constructed utilizing cribriform adenocarcinoma and SVI (Table 2). The model\u0026rsquo;s performance was further evaluated using ROC curve analysis, yielding an AUC of 0.723 in the TD and 0.784 in the IVD.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2. Logistic regression analysis for predicting complete response (CR) in the training set.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCharacteristic\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUnivariate OR(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ep\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMultivariate OR(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e0.987(0.932-1.045)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.656\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e/\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e/\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePSA（＜20 vs \u0026ge; 20ng/ml）\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e0.515(0.245-1.084)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.648(0.263-1.595)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.345\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTumor size（＜1.5cm vs \u0026ge;1.5cm）\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e0.574(0.279-1.182)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.132\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e/\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e/\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVI\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.204(0.074-0.566)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.002\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.251(0.070-0.899)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.034\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eECE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.299(0.142-0.630)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.956(0.338-2.698)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.932\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBone metastasis\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.705(0.287-1.731)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.446\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e/\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e/\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCribriform\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e0.629(0.164-2.410)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.049(0.006-0.386)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDuctal\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e0.126(0.016-0.983)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.048\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.143(0.016-1.298)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.084\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eERG\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.629(0.164-2.410)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.499\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e/\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e/\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch2\u003e\u003cstrong\u003eConstruction and Validation of the Radiomics Model\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eBased on inter- and intra-observer assessments, 807 out of 851 radiomics features demonstrated intraclass correlation coefficients (ICC) \u0026gt; 0.75. Features with low reproducibility (ICC \u0026lt; 0.75 for either intra- or inter-observer analysis) were excluded. The features with ICC \u0026lt; 0.75 are listed in Table S2. Subsequently, 609 features showed significant differences via the Mann-Whitney U test. After applying Spearman correlation analysis to remove highly correlated features, 137 features remained. The Boruta algorithm was then used for final feature selection, resulting in 8 key features. These 8 final radiomics features were: 1. wavelet-HLL_firstorder_Mean; 2. wavelet-LHL_firstorder_Maximum; 3. wavelet-HHL_glszm_SizeZoneNonUniformity; 4. wavelet-HHH_firstorder_Maximum; 5. wavelet-LHH_glrlm_HighGrayLevelRunEmphasis; 6. wavelet-LHL_glszm_SizeZoneNonUniformity; 7. wavelet-LHH_glcm_Autocorrelation; 8. wavelet-HLH_firstorder_Median. A radiomics model was built based on these features. All 8 features showed significant differences between the PR and CR groups (all p-values \u0026lt; 0.05) and were used for subsequent model construction (Figure 2-1). The importance of the selected features is shown in Figure 2-2, and their correlation matrix is depicted in Figure 2-3. Finally, radiomics models were developed using logistic regression (LR), random forest (RF), and support vector machine (SVM) algorithms based on this specific feature set.\u003c/p\u003e\n\u003cp\u003eThe performance of the three radiomics models is shown in Table 3. In the training set, the logistic regression (LR) model achieved an AUC of 0.836 (95% CI: 0.758-0.914), the random forest (RF) model an AUC of 0.866 (95% CI: 0.796-0.936), and the support vector machine (SVM) model an AUC of 0.872 (95% CI: 0.804-0.940). The radiomics model constructed using SVM demonstrated the best performance. In the validation set, among the three classifiers, the SVM-based radiomics model also performed best in predicting CR, with an AUC of 0.923 (95% CI: 0.882-1.012) (\u003cstrong\u003eFigure 3\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u003c/strong\u003e summarizes the predictive performance of the different models.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 211px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTraining Dataset\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 238px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eValidation Dataset\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003e95% CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e95% CI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eClinical\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e0.723\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003e(0.622-0.825)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.784\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e(0.644-0.923)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e0.836\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003e(0.758-0.914)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.899\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e(0.807-0.992)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRF\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.866\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e(0.796-0.936)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.850\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e(0.735-0.965)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVM\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.872\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003e(0.804-0.940)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e0.923\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e(0.843-1.003)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCOMB\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e0.909\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e(0.852-0.965)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.947\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 119px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e(0.882-1.012)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch2\u003e\u003cstrong\u003ePerformance of the Combined Model in Predicting Complete Response and Prognostic Assessment\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eA combined model was constructed by integrating the clinical model (based on the presence of cribriform adenocarcinoma and MRI-detected seminal vesicle invasion) with the radiomics signature generated by the SVM algorithm. For predicting CR, the combined model achieved an area under the curve (AUC) of 0.909 (95% CI: 0.852\u0026ndash;0.965) in the training set and 0.947 (95% CI: 0.882\u0026ndash;1.012) in the validation set. Decision curve analysis (DCA), a statistical method to evaluate a model\u0026apos;s utility in facilitating clinical decision-making, demonstrates clinical value only when the model\u0026apos;s net benefit exceeds that of the \u0026quot;treat all\u0026quot; and \u0026quot;treat none\u0026quot; strategies across a range of threshold probabilities. In our study, the net benefit curve of the combined model was higher than both the \u0026quot;treat all\u0026quot; and \u0026quot;treat none\u0026quot; curves when the threshold probability ranged from 0.35 to 1.0 in the training cohort and from 0.47 to 1.0 in the validation cohort (\u003cstrong\u003eFigure 4\u003c/strong\u003e). This indicates that the combined model provided higher net benefit within these threshold ranges, suggesting its potential to support clinical decision-making at higher risk thresholds. Collectively, these results indicate that the combined model exhibits high predictive performance for NHT efficacy and demonstrates favorable clinical utility.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDFS Follow-up Duration Statistics:\u003c/strong\u003e In the training set, the median follow-up time was 34.0 months (range: 1.0-101.0 months) for the complete response (CR) group and 19.0 months (range: 0.2-85.0 months) for the partial response (PR) group. In the validation set, the median follow-up time was 33.0 months (range: 8.0-53.0 months) for the CR group and 15.0 months (range: 1.0-56.0 months) for the PR group（\u003cstrong\u003eTable 4\u003c/strong\u003e）. Kaplan-Meier survival curves for both datasets revealed a significant difference in disease-free survival (DFS) between the CR and PR groups (all p \u0026lt; 0.05) (\u003cstrong\u003eFigure 5\u003c/strong\u003e). While the number of patients progressing to castration-resistant prostate cancer (CRPC) showed no significant difference between the CR and PR groups when analyzing the training and validation sets separately, a pooled analysis of both datasets demonstrated a statistically significant difference in CRPC progression between the two groups (p = 0.038). Patients in the CR group had significantly longer DFS than those in the PR group (p \u0026lt; 0.05), further indicating that the treatment response following neoadjuvant therapy holds predictive value for patient prognosis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4\u003c/strong\u003e. Prognostic disparities between the training and validation datasets.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 231px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eTraining Dataset\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eValidation Dataset\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003eCR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 82px;\"\u003e\n \u003cp\u003ePR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eCR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003ePR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCRPC-NO\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 82px;\"\u003e\n \u003cp\u003e50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCRPC-YES\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 82px;\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e0.2566\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e0.2113\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDFS-Media\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e34\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e19\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e33\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e15\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDFS-Range\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e1-101\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.2-85\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.042\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e8-53\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1-56\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.017\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"4 DISCUSSION","content":"\u003cp\u003eOur study demonstrates the value of radiomics in predicting the response to neoadjuvant hormonal therapy (NHT) in patients with high-risk prostate cancer (PCa). We further found that an MRI-based radiomics model outperformed the clinical-MRI model and could accurately predict treatment response. In this study, the combined model, developed using a Support Vector Machine (SVM) algorithm and incorporating clinical MRI features, showed the best performance in predicting NHT response. Compared to the clinical-MRI model and the individual radiomics models in both the training and internal validation datasets, the combined model exhibited a higher area under the curve (AUC) in ROC analysis, indicating superior predictive efficacy.\u003c/p\u003e\n\u003cp\u003eProstate-specific antigen (PSA) is currently a commonly used tool for assessing NHT outcomes and treatment response in PCa patients. However, its most significant limitation is its low specificity, as PSA is an organ-specific rather than a tumor-specific biomarker [17]. In this study, PSA was categorized as a binary variable (\u0026lt;20 ng/mL vs. ≥20 ng/mL), and no significant difference was observed between patients with different treatment responses. Ke ZB, Chen SM, et al. utilized [68Ga]Ga-PSMA-11 PET/CT to predict treatment response after NHT in high-risk non-metastatic PCa patients. Their multivariate logistic regression analysis indicated that the maximum standardized uptake value (SUVmax) after neoadjuvant combined hormonal therapy (NCHT) prior to radical prostatectomy was an independent predictor of favorable pathological response [10]. However, their study had a limited sample size and did not incorporate radiomics analysis. Wu XH et al. previously explored the value of radiomics in predicting NHT response in high-risk PCa patients [16]. However, their radiomics model construction employed only a single feature selection method (LASSO) and did not further investigate the optimal classifier for building the model, and the sample size was under 100 patients.\u003c/p\u003e\n\u003cp\u003eThe choice of classifier can impact model performance; however, no consensus exists on the optimal selection. In our study, the SVM model outperformed the RF and LR models in treatment response prediction. The SVM algorithm is particularly adept at identifying subtle patterns in complex datasets due to its ability to minimize classification errors on unseen data without requiring prior assumptions about the data's probability distribution [14]. Given the complexity and non-linearity between radiomic features and tumor response, this grants the algorithm an advantage in modeling moderately non-linear relationships [15]. Nonetheless, the efficacy of different classifiers may vary across clinical scenarios. Future studies exploring the optimal classifier for specific clinical applications are warranted.\u003c/p\u003e\n\u003cp\u003eOur study also further analyzed the prognostic value of the combined model. Kaplan-Meier survival curves in both datasets revealed a significant difference in disease-free survival (DFS) between the complete response (CR) group and the incomplete response (non-CR) group (both p \u0026lt; 0.05). The CR group demonstrated longer DFS compared to the non-CR group (p \u0026lt; 0.05), further indicating that different post-neoadjuvant therapy responses can predict patient outcomes after surgery.\u003c/p\u003e\n\u003cp\u003eThis study has several limitations. Manual segmentation of regions of interest (ROIs) is a time-consuming process requiring accurate identification of MRI lesions, which can be challenging for clinicians lacking experience in interpreting prostate MRI. Moreover, ROI delineation varies between physicians, highlighting a lack of standardized quality control. Therefore, developing an automated or semi-automated tool to optimize the ROI segmentation process is necessary. Additionally, prospective studies are needed to validate our findings.\u003c/p\u003e"},{"header":"5 CONCLUSIONS","content":"\u003cp\u003eIn summary, based on a large sample size, conventional imaging sequences, and multi-dataset validation, this study constructed a robust and generalizable radiomics model for predicting complete treatment response, showing potential value for prognostic risk stratification. Adhering strictly to radiomics quality control standards, this study achieved a Radiomics Quality Score (RQS) of 20 points(Table S3), confirming the rigor of the methodology and the reliability of the conclusions. The proposed predictive model holds promise for clinical application.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHai Zhou mainly contributed to this work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors and Affiliations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDepartment of Urology,Fudan University affiliated Huadong hospital, China, shanghai\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHai Zhou, Zhi-Yu Qian,\u0026nbsp;Zhi-Hao\u0026nbsp;Chen, Jian-Jin Fan, Yong-Xin Mao ,Le-Yan Xu, \u0026nbsp;Yi-Jun He, Jin-Xiong Zhang ,Jian- Hong Wu,\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eLu Sheng\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHai Zhou\u003c/strong\u003e: Formal analysis ; methodology ; visualization ; writing – original draft . \u003cstrong\u003eZhi-Yu Qian\u003c/strong\u003e:\u0026nbsp;Writing – original draft .\u003cstrong\u003e\u0026nbsp;Zhi-Hao Chen:\u003c/strong\u003e Formal analysis . \u003cstrong\u003eJian-Jin Fan ,\u003c/strong\u003e \u003cstrong\u003eYong-Xin Mao\u003c/strong\u003e,\u003cstrong\u003e\u0026nbsp;Le-Yan Xu:\u003c/strong\u003e Data curation . \u003cstrong\u003eYi-Jun He\u003c/strong\u003e, \u003cstrong\u003eJin-Xiong Zhang,\u003c/strong\u003e \u003cstrong\u003eJian- Hong Wu\u003c/strong\u003e:\u0026nbsp;Conceptualization.\u003cstrong\u003eLu Sheng:\u003c/strong\u003e Project administration.( \u0026nbsp;All authors reviewed the manuscript.)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCorresponding authors\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorrespondence to \u003cstrong\u003eLu Sheng\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics declarations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Ethics Committee of \u003cstrong\u003eFudan University affiliated Huadong hospital\u003c/strong\u003e , and all patients provided written informed consent.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDATA AVAILABILITY STATEMENT\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSiegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2021. CA Cancer J Clin. 2021;71:7\u0026ndash;33.\u003c/li\u003e\n\u003cli\u003eChang AJ, Autio KA, Roach M 3rd, Scher HI. High-risk prostate cancer-classification and therapy. Nat Rev Clin Oncol. 2014;11:308_23.\u003c/li\u003e\n\u003cli\u003ePan J, Chi C, Qian H, Zhu Y, Shao X, Sha J, et al. Neoadjuvant chemohormonal therapy combined with radical prostatectomy and extended PLND for very high risk locally advanced prostate cancer: a retrospective comparative study. Urol Oncol. 2019;37:991\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eDevos G, Devlies W, De Meerleer G, et al. Neoadjuvant hormonal therapy before radical prostatectomy in high-risk prostate cancer. \u003cem\u003eNat Rev Urol\u003c/em\u003e. 2021; \u003cstrong\u003e18\u003c/strong\u003e(12): 739-762.\u003c/li\u003e\n\u003cli\u003eMottet N, van den Bergh R, Briers E. European association of urology guidelines 2019 Arnhem European Association of Urology Office:196\u0026ndash;299\u003c/li\u003e\n\u003cli\u003eShelley MD, Kumar S, Wilt T, Staffurth J, Coles B, et al A systematic review and meta-analysis of randomised trials of neo-adjuvant hormone therapy for localised and locally advanced prostate carcinoma Cancer Treat Rev. 2009;35:9\u0026ndash;17\u003c/li\u003e\n\u003cli\u003ePignot G, Maillet D, Gross E, Barthelemy P, Beauval JB, et al Systemic treatments for high-risk localized prostate cancer Nat Rev Urol. 2018;15:498\u0026ndash;510\u003c/li\u003e\n\u003cli\u003eLiu W, Yao Y, Liu X, Liu Y, Zhang G-M. Neoadjuvant hormone therapy for patients with high-risk prostate cancer: a systematic review and meta-analysis. \u003cem\u003eAsian J Androl\u003c/em\u003e. 2021; \u003cstrong\u003e23\u003c/strong\u003e(4): 429-436.\u003c/li\u003e\n\u003cli\u003eDolezel M, Odrazka K, Vanasek J, et al. Neoadjuvant hormonal therapy in prostate cancer\u0026mdash;impact of PSA level before radiotherapy. \u003cem\u003eJ BUON\u003c/em\u003e. 2013; \u003cstrong\u003e18\u003c/strong\u003e(4): 949-953.\u003c/li\u003e\n\u003cli\u003eKe Z-B, Chen S-M, Chen J-Y, et al. Head-to-head comparisons of [68Ga]Ga-PSMA-11 PET/CT, multiparametric MRI, and prostate-specific antigen for the evaluation of therapeutic responses to neoadjuvant chemohormonal therapy in high-risk non-metastatic prostate cancer patients: a prospective study. \u003cem\u003eEur J Nucl Med Mol Imaging\u003c/em\u003e. 2023; \u003cstrong\u003e50\u003c/strong\u003e(4): 1240-1251.\u003c/li\u003e\n\u003cli\u003eLambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. \u003cem\u003eNat Rev Clin Oncol\u003c/em\u003e. 2017; \u003cstrong\u003e14\u003c/strong\u003e(12): 749-762.\u003c/li\u003e\n\u003cli\u003eHu T, Gong J, Sun Y, Li M, Cai C, Li X, Cui Y, Zhang X, Tong T. Magnetic resonance imaging-based radiomics analysis for prediction of treatment response to neoadjuvant chemoradiotherapy and clinical outcome in patients with locally advanced rectal cancer: A large multicentric and validated study. MedComm (2020). 2024 Jun 20;5(7):e609.\u003c/li\u003e\n\u003cli\u003eAbdollahi H, Mofid B, Shiri I, et al. Machine learning-based radiomic models to predict intensity-modulated radiation therapy response, Gleason score and stage in prostate cancer. \u003cem\u003eRadiol Med\u003c/em\u003e. 2019; \u003cstrong\u003e124\u003c/strong\u003e(6): 555-567.\u003c/li\u003e\n\u003cli\u003eHuang S, Cai N, Pacheco PP, Narrandes S, Wang Y, Xu W. Applications of support vector machine (SVM) learning in cancer genomics. \u003cem\u003eCancer Genomics Proteomics\u003c/em\u003e. 2018; \u003cstrong\u003e15\u003c/strong\u003e(1): 41-51.\u003c/li\u003e\n\u003cli\u003eEmblem KE, Pinho MC, Zollner FG, et al. A generic support vector machine model for preoperative glioma survival associations. \u003cem\u003eRadiology\u003c/em\u003e. 2015; \u003cstrong\u003e275\u003c/strong\u003e(1): 228-234.\u003c/li\u003e\n\u003cli\u003eWu XH, Ruan ZT, Ke ZB, Lin F, Chen JY, Xue YT, Lin B, Chen SH, Chen DN, Zheng QS, Xue XY, Wei Y, Xu N. Magnetic resonance imaging-based radiomics nomogram for the evaluation of therapeutic responses to neoadjuvant chemohormonal therapy in high-risk non-metastatic prostate cancer. Cancer Med. 2024 Jul;13(14):e70001.\u003c/li\u003e\n\u003cli\u003eFilella X, Foj L. Emerging biomarkers in the detection and prognosis of prostate cancer. Clin Chem Lab Med. 2015;53:963\u0026ndash;73. \u003c/li\u003e\n\u003cli\u003eWang Z, Wang J, Li D, Wu R, Huang J, Ye L, Tuo Z, Yu Q, Shao F, Wusiman D, Cho WC, Koh SB, Xiong W, Feng D. Novel hormone therapies for advanced prostate cancer: Understanding and countering drug resistance. J Pharm Anal. 2025 Sep;15(9):101232.\u003c/li\u003e\n\u003cli\u003ePetraki CD, Sfikas CP. Histopathological changes induced by therapies in the benign prostate and prostate adenocarcinoma. \u003cem\u003eHistol Histopathol\u003c/em\u003e. 2007; \u003cstrong\u003e22\u003c/strong\u003e(1): 107-118.\u003c/li\u003e\n\u003cli\u003eThalgott M, Horn T, Heck MM, Maurer T, Eiber M, Retz M, et al. Long-term results of a phase II study with neoadjuvant docetaxel chemotherapy and complete androgen blockade in locally advanced and high-risk prostate cancer. J Hematol Oncol. 2014;7:20.\u003c/li\u003e\n\u003cli\u003eCheng Q, Butler W, Zhou Y, Zhang H, Tang L, Perkinson K, Chen X, Jiang XS, McCall SJ, Inman BA, Huang J. Pre-existing Castration-resistant Prostate Cancer-like Cells in Primary Prostate Cancer Promote Resistance to Hormonal Therapy. Eur Urol. 2022 May;81(5):446-455. \u003c/li\u003e\n\u003cli\u003eMiron B, Kursa WRR. Feature selection with the boruta package. \u003cem\u003eJ Statist Software\u003c/em\u003e. 2010; \u003cstrong\u003e36\u003c/strong\u003e(11): 13.\u003c/li\u003e\n\u003cli\u003eLambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. \u003cem\u003eNat Rev Clin Oncol\u003c/em\u003e. 2017; \u003cstrong\u003e14\u003c/strong\u003e(12): 749-762.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Neoadjuvant Hormonal Therapy, complete pathological response, Prostate cancer, Radiomics","lastPublishedDoi":"10.21203/rs.3.rs-8655062/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8655062/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003eThis study aimed to develop and validate an MRI-based radiomics model for predicting complete pathological response (pCR) to neoadjuvant hormonal therapy (NHT) and to assess its prognostic value in patients with high-risk prostate cancer (HRPCa).\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003e This retrospective study included 140 patients with HRPCa who underwent NHT followed by radical prostatectomy at Huadong Hospital. Radiomic features were extracted from preoperative apparent diffusion coefficient (ADC) maps derived from multiparametric MRI. Feature selection was performed using appropriate statistical methods, and predictive models were constructed using various machine learning classifiers. The performance of the radiomics signature and a combined model integrating clinical MRI and pathological features were evaluated.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eFor pCR prediction, one clinical MRI feature (seminal vesicle invasion), one pathological feature (cribriform adenocarcinoma), and eight radiomics features were ultimately selected. The combined model, which incorporated these features with a radiomics signature generated by a Support Vector Machine (SVM) classifier, demonstrated excellent discriminative ability. It achieved areas under the curve (AUC) of 0.909 (95% CI: 0.852\u0026ndash;0.965) in the training cohort and 0.947 (95% CI: 0.882\u0026ndash;1.012) in the internal validation cohort. Decision curve analysis confirmed its clinical net benefit. Furthermore, Kaplan-Meier analysis revealed that patients predicted by the model to have a higher probability of pCR experienced significantly better disease-free survival compared to those with a lower predicted probability.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eBased on a substantial sample size and rigorous internal validation, the proposed MRI-based radiomics model achieved a high Radiomics Quality Score (RQS). It shows strong potential for the non-invasive prediction of treatment response and prognosis in HRPCa, demonstrating significant clinical value.\u003c/p\u003e","manuscriptTitle":"Predicting Response to Neoadjuvant Hormonal Therapy and Prognostic Outcomes in High-Risk Prostate Cancer Using MRI-Based Radiomics","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-13 16:50:58","doi":"10.21203/rs.3.rs-8655062/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-02T05:33:42+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-02T03:24:04+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-21T20:10:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"91712939189908871567518502306914821540","date":"2026-02-19T13:03:04+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"257376027854701793190356418195829677506","date":"2026-02-10T19:05:07+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-10T08:55:04+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-01-30T06:48:33+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-01-26T09:59:10+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-01-26T09:56:15+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Cancer","date":"2026-01-21T04:32:00+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fc7256c6-3bd8-44bd-aed9-2c3bb10ea00c","owner":[],"postedDate":"February 13th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-05-04T15:59:41+00:00","versionOfRecord":{"articleIdentity":"rs-8655062","link":"https://doi.org/10.1186/s12885-026-16015-0","journal":{"identity":"bmc-cancer","isVorOnly":false,"title":"BMC Cancer"},"publishedOn":"2026-04-29 15:57:37","publishedOnDateReadable":"April 29th, 2026"},"versionCreatedAt":"2026-02-13 16:50:58","video":"","vorDoi":"10.1186/s12885-026-16015-0","vorDoiUrl":"https://doi.org/10.1186/s12885-026-16015-0","workflowStages":[]},"version":"v1","identity":"rs-8655062","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8655062","identity":"rs-8655062","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}