Journal: Nature communications
This study presents ProAI, an automated decision aid designed to assess patient-level risk of clinically significant prostate cancer (csPCa) using biparametric MRI.
Key points include:
- Training and Validation: ProAI was trained and validated on 7,849 exams from multiple centers and public datasets.
- Diagnostic Accuracy: It demonstrated high diagnostic accuracy with a pooled external AUC of 0.93, comparable to PI-RADS, but with improved consistency.
- Multi-Reader Study: Use of ProAI increased clinician accuracy from 80% to 86% and reduced reading times.
- Prospective Deployment: In nearly 2,000 exams, ProAI maintained strong performance (AUC 0.92) and reduced radiology workload by 32%.
- Generalizability: The tool generalized well to external cohorts, supporting its potential to standardize reporting and improve efficiency in prostate cancer diagnosis.