External validation on a Japanese cohort of a computer-aided diagnosis system aimed at characterizing ISUP ≥2 prostate cancers at multiparametric MRI
Résumé
Objective: To test, on a Japanese cohort, a computer-aided diagnosis (CADx) system trained on a French population to diagnose International Society of Urological Pathology (ISUP) ≥2 prostate cancer on multiparametric MRI, based on the apparent diffusion coefficient (ADC) and wash-in rate.
Methods: 68 consecutive patients who underwent prostate MRI before prostatectomy at a single Japanese institution were retrospectively included. MRIs were reviewed by an experienced (20 years) radiologist who delineated suspicious lesions and assigned to them a Prostate Imaging-Reporting and Data System version 2.1 (PI-RADSv2.1) score. The CADx score was computed from the delineated regions-of-interest. Using prostatectomy whole-mounts as reference, the PI-RADSv2.1 and CADx scores were compared at lesion level using areas under the receiver operating characteristics curves (AUC) and sensitivities and specificities obtained with predefined thresholds.
Results: The CADx and PI-RADSv2.1 AUCs were 80% (95% confidence interval (95%CI): 71-90) and 80% (95%CI: 71-89; p=0.886) respectively in PZ, and 79% (95%CI: 66-90) and 93% (95%CI: 82-96; p=0.051) respectively in TZ. The CADx diagnostic thresholds, that were expected to provide a 90% sensitivity, yielded sensitivities of 60% (95%CI: 38-83) in PZ and 42% (95%CI: 20-71) in TZ, for specificities of 95% (95%CI: 86-100) and 92% (95%CI: 73-100) respectively. Potential explanations for this difference include significantly higher ADC values in prostate tissues (except TZ cancers), and lower temporal resolution at dynamic contrast-enhanced imaging in the test cohort.
Conclusion: The CADx provided good overall results in this external validation cohort. However, its predefined diagnostic thresholds provided lower sensitivities and higher specificities than expected.