A Quantitative Nomogram Integrating IVIM-derived True Diffusion Coefficient and PSAD for Predicting Clinically...

To develop and validate a quantitative nomogram integrating the intravoxel incoherent motion (IVIM)-derived true diffusion coefficient (D) and prostate-specific antigen density (PSAD) for predicting clinically significant prostate cancer (csPCa) in patients with elevated serum PSA, and to compare its diagnostic performance with that of Prostate Imaging Reporting and Data System (PI-RADS)-based assessment.

This combined retrospective and prospective study enrolled 341 patients (total PSA > 4 ng/mL) who underwent prostate magnetic resonance imaging incorporating IVIM-diffusion-weighted imaging and subsequent pathological confirmation. The retrospective cohort (n = 253) was randomly split into training (n = 176) and internal validation (n = 77) sets; an independent prospective cohort (n = 88) served as the temporal validation set. csPCa was defined as a Gleason score ≥7. Independent predictors were identified from IVIM-derived parameters (D, D*, f), conventional apparent diffusion coefficient (ADC), and PSA-derived indices using multivariate logistic regression. Models of increasing complexity were constructed, and the optimal parameter combination was determined. Diagnostic performance was compared using the area under the receiver operating characteristic curve (AUC), the DeLong test, net reclassification improvement, integrated discrimination improvement (IDI), and decision curve analysis (DCA). PI-RADS alone and PI-RADS + PSAD served as reference comparators.

PSAD, ADC, and D were identified as independent predictors of csPCa. The PSAD + D model achieved AUCs of 0.920, 0.900, and 0.916 in the training, validation, and prospective sets, respectively; PSAD + ADC + D yielded AUCs of 0.923, 0.877, and 0.917; and PI-RADS + PSAD achieved 0.903, 0.919, and 0.936. The DeLong test revealed no significant differences among the three models in any dataset (all false discovery rate-corrected p > 0.05). IDI analysis demonstrated that incorporating ADC into the PSAD + D model resulted in significantly negative discrimination improvement across all three datasets.

The PSAD + D nomogram achieves diagnostic performance comparable to PI-RADS + PSAD for predicting csPCa while eliminating reader dependency through objective, highly reproducible parameters (intraclass correlation coefficients: 0.93-0.96). The paradoxical performance attenuation upon ADC addition demonstrates that parameter parsimony, rather than dimensionality maximization, optimizes predictive accuracy in quantitative imaging models.