Prostate MR AI (VA10A)

Prostate MR AI (VA10A) is a software plug-in that assists radiologists in interpreting prostate MRI examinations acquired according to the PI-RADS standard. It analyzes T2-weighted and diffusion-weighted MRI images to automatically segment the prostate gland and detect suspicious lesions in treatment-naïve men aged 40 and older. The device provides lesion contours, severity scores (3–5 scale), and a level-of-suspicion metric to support radiologist decision-making; however, clinical management decisions should not be based solely on the algorithm's output.

Prostate MR AI is a software-only device using an artificial intelligence algorithm trained on prostate MRI image series acquired per PI-RADS standard, with corresponding radiological and biopsy findings as ground truth. It automatically performs prostate segmentation (peripheral zone and non-peripheral zone), lesion detection, and PI-RADS classification. Unlike the predicate device Transpara™ (which marks suspicious locations on breast mammograms), Prostate MR AI outputs lesion contours and a suspicion map without a standalone graphical user interface; instead, results integrate into a hosting workflow application.

ISO 14971 (risk management), IEC 62304 (medical device software lifecycle), and IEC 82304-1 (health software lifecycle). The submission also references FDA Guidance "Content of Premarket Submissions for Device Software Functions" (June 2023) for Basic Documentation Level compliance.

The predicate device Transpara™ is a CAD/CADx software that assists radiologists in detecting and characterizing abnormalities on medical images using AI algorithms; Prostate MR AI performs the same core functions—detection, assessment, and characterization—with substantially similar clinical workflow (concurrent reading aid, not replacement of clinician judgment). Although differences exist in anatomical region (prostate vs. breast), imaging modality (MRI vs. mammography), and output format (lesion contours/suspicion map vs. marks and region scores), these are justified by the reference device ProstatID™ (also prostate MRI CAD/CADx) which demonstrates that such differences do not raise new safety or effectiveness concerns. Multi-reader/multi-case clinical data show statistically significant improvement in radiologist performance with device aid (AUROC 0.6758 → 0.7010 unaided-to-aided; p=0.040), comparable to predicate study results; nonclinical testing met all acceptance criteria for segmentation (Dice score >0.9) and lesion detection (sensitivity ≥0.80, accuracy ≥0.8).