Related Experiment Video
Updated: Jun 20, 2026
Use of MRI-ultrasound Fusion to Achieve Targeted Prostate Biopsy
Published on: April 9, 2019
External Validation of a Previously Developed Deep Learning-based Prostate Lesion Detection Algorithm on Paired
Enis C Yilmaz1, Stephanie A Harmon1, Yan Mee Law1
1From the Molecular Imaging Branch (E.C.Y., S.A.H., M.J.B., Y.L., D.G.G., K.B.O., N.S.L., P.E., P.L.C., B.T.), Biometric Research Program, Division of Cancer Treatment and Diagnosis (E.P.H.), Center for Interventional Oncology (L.A.H., C.G., B.J.W.), Department of Radiology, Clinical Center (L.A.H., C.G., B.J.W.), Laboratory of Pathology (A.T., M.J.M.), and Urologic Oncology Branch (S.G., P.A.P.), National Cancer Institute, National Institutes of Health, 10 Center Dr, MSC 1182, Bldg 10, Rm B3B85, Bethesda, MD 20892; Department of Radiology, Singapore General Hospital, Singapore (Y.M.L.); and NVIDIA Corporation, Santa Clara, Calif (D.Y., Z.X., J.T., D.X.).
An artificial intelligence (AI) model showed modest performance in detecting prostate cancer lesions on external biparametric MRI (bpMRI) scans, with improved detection on in-house scans. Keywords: AI, prostate cancer, bpMRI.
Area of Science:
- Radiology and Imaging Science
- Artificial Intelligence in Medicine
- Urologic Oncology
Background:
- Biparametric MRI (bpMRI) is increasingly used for prostate cancer detection.
- Artificial intelligence (AI) models show potential for improving diagnostic accuracy in radiology.
- Evaluating AI performance across different imaging datasets is crucial for clinical implementation.
Purpose of the Study:
- To assess an AI model's capability in detecting prostate cancer lesions on external and in-house bpMRI scans.
- To compare the AI model's detection performance between the two distinct bpMRI datasets.
- To identify factors influencing AI detection of overall and clinically significant prostate cancer (csPCa).
Main Methods:
- Retrospective analysis of 201 patients with paired external and in-house bpMRI scans.
- An established bpMRI-based AI lesion detection model was applied to both datasets.
- Detection rates were compared using permutation tests; influencing factors were analyzed with mixed-effects models.
Main Results:
- The AI model detected significantly more lesions on in-house bpMRI (56.0%) compared to external bpMRI (39.7%) (P < .001).
- Detection of clinically significant prostate cancer (csPCa)-positive lesions was also higher on in-house (79%) versus external scans (61%) (P < .001).
- On external scans, higher PI-RADS scores, larger lesion diameter, better image quality, and fewer lesions improved AI detection.
Conclusions:
- The AI model demonstrated modest performance for detecting prostate cancer lesions on external bpMRI.
- Performance was significantly better on in-house bpMRI scans, suggesting dataset variability impacts AI efficacy.
- Factors like PI-RADS score and lesion characteristics influence AI detection accuracy in prostate MRI.
