Development and validation of a risk prediction model for prostate cancer integrating quantitative MRI and clinical data

Objectives To develop and validate a predictive model for prostate cancer (PCa)by integrating quantitative MRI and clinical data. Methods A retrospective study was conducted on 290 patients clinically diagnosed with prostate lesions between March 2023 and October 2024. All patients underwent multimodal quantitative MRI preoperatively, including T2-weighted imaging (T2WI), T1-weighted imaging (T1WI), diffusion-weighted imaging (DWI, b=0, 500 mm 2/s ), synthetic relaxometry (MAGiC), iterative decomposition of water and fat with echo asymmetry and least-squares estimation quantitation (IDEAL-IQ), and amide proton transfer (APT) sequences. Based on pathological results, patients were stratified into PCa group (n=140) and Non-PCa group (n=150). All quantitative parameters derived from multimodal MRI were measured independently by a dedicated team of radiologists, and their reproducibility was assessed through intra−class correlation coefficients (ICCs). Differences in all MRI quantitative parameters and clinical data between the groups were compared. All patients were randomly divided into a training group (203 cases) and a validation group (87 cases) at a ratio of 7:3. The Least Absolute Shrinkage and Selection Operator (LASSO) and Logistic regression were used to screen independent predictive parameters for PCa, constructing a predictive model, internal validation and bootstrap validation were performed on the model. The model was evaluated using receiver operating characteristic curves (ROC), goodness-of-fit tests (Hosmer-Lemeshow statistic), calibration curves, and decision curve analysis (DCA), and presented in the form of nomograms. Results 1) Two radiologists showed good consistency (ICCs > 0.75). 2) There were significant differences between the PCa and non-PCa groups for all variables except age( P < 0.05). 3) Four independent predictors were selected through LASSO-penalized logistic regression to construct the model: the Apparent Diffusion Coefficient (ADC) from DWI, T2 from MAGiC, MTRasym(3.5ppm) from APT, and Prostate-Specific Antigen (PSA). The model exhibited excellent performance, achieving an AUC of 0.986 (95% CI: 0.944-0.993); sensitivity: 92.3%; specificity: 93.3%; the optimism−adjusted area under the curve was 0.988 (bootstrap 95 % CI: 0.986–0.991), satisfactory calibration (Hosmer-Lemeshow test, P > 0.05), and a substantial net benefit on decision curve analysis. Conclusion The prediction model based on quantitative MRI and clinical parameters has good performance in discrimination, stability, consistency and clinical applicability. Compared to existing diagnostic models, it offers distinct advantages in data objectivity, reproducibility, non-invasiveness, cost-effectiveness, and operational simplicity.