This is a single - center cross - sectional retrospective study. The study included patients who underwent an ultrasound-guided transperineal prostate biopsy at Anqing Municipal Hospital between January 2024 and February 2025. All clinicopathological data were collected from the hospital information system. Following the application of exclusion criteria, 160 patients were enrolled into the modeling group from 240 individuals between January 2024 and December 2024, while 40 patients were selected for the external validation group from 60 cases between December 2024 and February 2025. Based on pathological outcomes in this modeling group, there were 83 cases (51.9%) in the csPCa group and 77 cases (48.1%) in the non - csPCa group, in the external validation group, there were 22 cases (55%) in the CsPCa group and 18 cases (45%) in the non - csPCa group. The exclusion criteria of the study were as follows: patients with incomplete clinical, MRI, or ultrasound data; patients who did not undergo biopsy within 7 days after MRI; positive cases without detected positive cores in the region of interest (ROI) and lacking elastography images. This research was approved by the Medical Ethics Committee of Anqing Municipal Hospital. This study was conducted following the guiding principles of the Declaration of Helsinki, and all data were collected in compliance with the Health Insurance Portability and Accountability Act (HIPAA). As a retrospective study, it adheres to the 1975 Declaration of Helsinki, and a waiver of informed consent was obtained. The flowchart of the study can be seen in Figure 1.

Clinical baseline data collection: Relevant clinical baseline data of patients before puncture were retrospectively collected from the hospital’s electronic medical record system, such as age, free prostate-specific antigen (fPSA), total prostate-specific antigen (tPSA), and the ratio of free prostate-specific antigen to total prostate-specific antigen (f/tPSA), and the patients who were included in the study underwent serology for tumor markers of males after hospital admission. examination and complete magnetic resonance examination to ensure the timeliness and accuracy of the data. Total PSA and free PSA were measured using the Cobas e602 fully automated immunoassay analyzer (Roche, Switzerland).

MRI data were obtained using a 3.0 T Skyra ultra-high field magnetic resonance scanner (Siemens, Germany) with an 18-channel body matrix coil. Patients partially emptied their bladder before scanning, which began 1 cm superior to the pubic symphysis, covering at least the prostate and seminal vesicles. Imaging included T2-weighted imaging (T2WI) sequence (including the cross section, sagittal plane, and coronal plane), diffusion-weighted imaging (DWI) generated by single-shot spin echo planar imaging (SS-EPI) sequence, and dynamic contrast-enhanced MRI (DCE-MRI) was generated by fat-suppressed T1-weighted imaging (T1WI) sequence with 3D spoiled gradient echo, after rapid intravenous injection with gadopentetate dimeglumine (0.2 mmol/kg). Suspicious lesions were identified and marked by specialized radiologists, who recorded the maximum diameter of the lesions and assigned scores according to the Prostate Imaging-Reporting and Data System (PI-RADS) version 2.1. Additionally, glandular zone delimitation was performed, and maximum dimensions in three orthogonal planes were measured.

Prostate biopsy was performed as soon as possible after MRI scanning (all patients underwent biopsy within one week of MRI). Equipment used: Mindray Nuewa R9s color Doppler ultrasound diagnostic system with shear wave elastography (STE) technology, measurement analysis software, and ELC13-4U endocavity biplane probe. Curved array probe: Frequency 3.5–9.5 MHz, scanning angle 217°. Linear array probe: Frequency 3.2–12.8 MHz, scanning angle 40°.

The patient was placed in the lithotomy position and underwent routine disinfection. A bi - planar probe was inserted through the anus to conduct a transrectal ultrasound (TRUS) examination. The ultrasound image manifestations of the prostate in the horizontal and sagittal planes were observed. Attention was paid to the location and characteristics of abnormal echoes. Manual cognitive fusion was carried out in combination with MRI images to preliminarily determine the location of the lesion. If multiple suspicious lesions were detected by mpMRI, the lesion with the highest PI - RADS score was considered. The STE program was then initiated. The initial range of the probe was set to 0–70 kPa. The probe was kept gently touching the rectal wall as much as possible, and the sampling frame was adjusted to the maximum to include the left and right lobes of the prostate as much as possible. The region of interest was located according to the hardness indication and MRI images. After the color filling in the region of interest was satisfactory, the image was frozen. Calcified areas were avoided. The region of interest was selected using the tracing method, and the average Young’s modulus value (Emean) and the maximum Young’s modulus value (Emax) of the region of interest were measured. The measurement was repeated three times in the same region, and the average of the three measurements was taken as the final result. Subsequently, under the guidance of transrectal bi - planar ultrasound, a 12 - needle systematic prostate biopsy was performed. One to two additional targeted biopsies were taken from the region of interest. The puncture specimens from each needle were fixed and sent for biopsy. The histopathological analysis report of each sample included positive cores and the corresponding Gleason grades. The cases were grouped according to the pathological results of the puncture in the region of interest. In some cases, no positive cores were detected in the region of interest, but positive cores were found in other samples. These cases were not included in this study due to the lack of elastography data. Clinically significant prostate cancer (csPCa) was defined as a Gleason score of ≥3 + 4 = 7. All the above operations were performed by physicians at or above the intermediate level in this specialty.

The prostate volume (PV) and the transition zone volume (TZV) are calculated using the ellipsoid formula (volume = length × width × height × π/6, in cubic centimeters). The peripheral zone volume (PZV) is calculated as the difference between the PV and the TZV. The lesion density is calculated as the ratio of the longest diameter of the lesion (in millimeters) to the prostate volume (in cubic centimeters). The TZ-ratio is defined as the ratio of the transition zone volume to the prostate volume.

All statistical analyses were performed using Excel 2019 (Microsoft Corporation, Redmond, WA, USA), IBM SPSS Statistics 26.0 (IBM Corp, Armonk, NY, USA), and R 4.4.0 (R Foundation for Statistical Computing, Vienna, Austria). Continuous variables with normal distribution were expressed as mean ± standard deviation (mean [SD]), while non-normally distributed variables were presented as medians with 25th and 75th percentiles. Categorical variables were reported as counts and percentages (%). Baseline characteristics were compared across groups using χ² tests or Fisher’s exact tests for categorical variables. For normally distributed continuous variables, one-way analysis of variance (ANOVA) was applied; non-normally distributed variables were analyzed using the Kruskal-Wallis H test or Mann-Whitney U test. Spearman correlation coefficients were calculated to assess relationships between variables. Univariate and multivariate logistic regression models were constructed to identify independent predictors of csPCA Covariates with significant associations in univariate analysis (p < 0.05) were included in the multivariate model. Forest plots were used to visualize adjusted odds ratios (ORs) and 95% confidence intervals (CIs). A predictive nomogram was developed based on significant variables from multivariate analysis. Discrimination was evaluated using receiver operating characteristic (ROC) curves, with the area under the curve (AUC) reported. Calibration was assessed via Hosmer-Lemeshow goodness-of-fit testing (p > 0.05 indicating adequate fit), and clinical utility was validated through decision curve analysis (DCA). The Delong test was used to compare ROC curves and evaluate differences in the AUC among various models. All statistical tests were two-tailed, with p < 0.05 considered statistically significant.

A total of 160 subjects were included in the modeling group of this study. Clinical baseline data of the patients were retrospectively collected (Table 1). According to the pathological results, the csPCa group (n=83, 51.9%) had a median age of 74.00 (68.00, 77.00), and the non-csPCa group (n=77, 48.1%) had a median age of 67.00 (62.00, 73.00). In the comparison of baseline characteristics, several variables showed significant differences, including age; tPSA; fPSA; f/t PSA; Emax; Emean; PV; TZV; TZ-ratio; Lesion Density; PI-RADS score (P<0.05). Compared with the non-csPCa group, the csPCa group had a higher median age (74.00 years vs. 67.00 years), higher levels of tPSA and fPSA, but a lower f/t PSA. Among elastography-related indicators, Emax and Emean in the csPCa group were significantly higher than those in the non-csPCa group. PV, TZV, and TZ-ratio were higher in the non-csPCa group. In addition, the csPCa group had a higher lesion density and a relatively higher PI-RADS score, and there was also a significant difference in PZV between the two groups (P = 0.032).

The TZ-ratio is a measure of the volume of the transition zone between the prostate and the PV, while the peripheral zone volume is calculated by taking TZV out of PV. Both PV and TZV were excluded from the univariate analysis due to their linear relationship and standard deviations. In the logistic regression analysis, variables were assigned specific cut-off values. (Table 2, Supplementary Table 1). The model identified several independent diagnostic factors for csPCa, including age, f/t PSA, Emax, TZ-ratio, lesion density, and PI-RADS score. These coefficients indicate their impact on the probability of diagnosing csPCa (Supplementary Table 2, Figure 2). Collinearity diagnosis using Variance Inflation Factor (VIF) demonstrated no significant multicollinearity among the six independent factors, with VIF values of 1.2, 1.1, 1.1, 1.1, 1.1, and 1.1 respectively (all <10) (Supplementary Table 3).

The logistic regression equation is as follows: Logit (p)= -4.723 + 0.951× Age – 0.669 × f/tPSA + 0.766 × Emax – 0.729 × TZ ratio + 0.747 × Lesion density + 1.323× PI-RADS score. Nomograms were constructed using R software and the rms package to visualize the model, as shown in Figure 3. From the figure, one can obtain the corresponding score for each indicator, and the predicted probability corresponding to the total points represents the probability of being diagnosed with csPCa.

Based on the sources of baseline factors with statistical significance in univariate analysis, other predictive models were constructed, namely Model 1 (Clinical Model): Age + tPSA +fPSA + f/t PSA; Model 2 (Ultrasonic Elasticity Model): Emax+ Emean; Model 3 (Magnetic Resonance Model): TZ-ratio + lesion density + PI-RADS score; and Model 4 (Independent Risk Factor Model): PI-RADS score + Age + f/t PSA + Emax+ TZ-ratio + lesion density. Compared with other models, the independent risk factor model showed a higher AUC value, indicating better diagnostic accuracy, and all differences were statistically significant (all P<0.05), (Table 3, Table 4; Figure 4).

The modeling cohort was randomly divided into a training set and a testing set at a 6:4 ratio are presented in Supplementary Table 4. No significant differences were observed between the two groups (P>0.05).For internal validation, the Hosmer-Lemeshow goodness-of-fit test showed excellent calibration in both the training set (χ²=5.418, P = 0.712) and the testing set (χ²=2.930, P = 0.939), with P>0.05 indicating no significant deviation from perfect calibration. Calibration curves demonstrated high consistency between predicted and observed probabilities (Supplementary Figures 1A, B). Decision Curve Analysis (DCA) revealed the Nomogram’s favorable clinical utility, with substantial net benefit across a wide range of threshold probabilities (Supplementary Figures 2A, B). Receiver Operating Characteristic (ROC) curve analysis yielded an AUC of 0.952 in the training set and 0.922 in the testing set (Supplementary Figures 3A, B). Delong’s test comparing the two ROC curves showed no significant difference (P = 0.07), confirming the model’s stable performance across datasets.

Clinical baseline data in the external validation cohort are presented in Supplementary Table 5. The Hosmer-Lemeshow test results (χ²=7.779; P = 0.455) indicated high consistency between predicted and observed values via the calibration curve (P = 0.455>0.05), confirming the model’s robust predictive performance. The DCA revealed the Nomogram’s favorable clinical utility. The external validation cohort achieved an AUC of 0.924, demonstrating excellent discriminatory ability for csPCa (Supplementary Figures 4A–C).