Comparison of diffusion-relaxation correlation spectrum imaging and conventional quantitative MRI parameters in...

To investigate the diagnostic value of diffusion-relaxation correlation spectrum imaging (DR-CSI) for differentiating benign prostatic hyperplasia (BPH) from prostate cancer (PCa), further evaluate its diagnostic performance for transition-zone lesions, and assess the robustness of DR-CSI across different classification algorithms.

This prospective study included 142 patients with pathologically confirmed benign prostatic hyperplasia (BPH; n = 83) or prostate cancer (PCa; n = 59). Of these, 85 patients were included in an exploratory transition-zone subgroup analysis, comprising 54 patients with BPH and 31 with transition-zone PCa. Quantitative parameters, including the apparent diffusion coefficient (ADC), T2, and DR-CSI compartmental signal fractions (fA-fD), were obtained. Binary logistic regression was used to construct the ADC + T2 and DR-CSI combined models. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curves, and areas under the curve (AUCs) were compared using the DeLong test. Internal validation was performed using 1000 stratified bootstrap resamples, and model calibration was assessed using calibration curves and Brier scores. To assess the robustness of DR-CSI across different classification algorithms, fA, fC, and fD were used as input features for logistic regression, LASSO-logistic regression, and support vector machine (SVM) models, with model performance evaluated using repeated stratified five-fold cross-validation with 10 repetitions.

The DR-CSI combined model achieved an AUC of 0.964 for differentiating BPH from PCa, with a sensitivity of 91.53% and a specificity of 91.57%, demonstrating significantly better diagnostic performance than the ADC + T2 model (AUC = 0.891, P = 0.0185). Bootstrap internal validation yielded a bias-corrected AUC of 0.961 for the DR-CSI model, with a mean optimism of 0.003 and a Brier score of 0.075. In the transition-zone subgroup, the DR-CSI combined model achieved an AUC of 0.949, higher than that of the ADC + T2 model (AUC = 0.869). Sensitivity analyses across different classification algorithms showed mean cross-validated AUCs of 0.964 ± 0.029, 0.964 ± 0.031, and 0.962 ± 0.031 for logistic regression, LASSO-logistic regression, and SVM, respectively, in the overall cohort, and 0.947 ± 0.060, 0.942 ± 0.062, and 0.942 ± 0.059, respectively, in the transition-zone subgroup. All three classification algorithms demonstrated high and comparable discriminative performance.

DR-CSI effectively differentiates BPH from PCa, and its combined model demonstrates superior diagnostic performance compared with the ADC + T2 model. Exploratory subgroup analysis further indicates that DR-CSI provides good discrimination between BPH and transition-zone PCa, suggesting its potential as a complementary quantitative imaging approach for the assessment of transition-zone lesions.