Related Experiment Video
Updated: Jun 15, 2026
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
Evaluation of a Cascaded Deep Learning-based Algorithm for Prostate Lesion Detection at Biparametric MRI
Yue Lin1, Enis C Yilmaz1, Mason J Belue1
1From the Molecular Imaging Branch (Y.L., E.C.Y., M.J.B., S.A.H., T.E.P., K.M.M., N.S.L., P.L.C., B.T.), Center for Interventional Oncology (L.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; NVIDIA, Santa Clara, Calif (J.T., D.Y., Z.X., D.X.); Department of Radiology, Clinical Center, National Institutes of Health, Bethesda, Md (L.H., C.G., B.J.W.); and Department of Radiology, Singapore General Hospital, Singapore (Y.M.L.).
An artificial intelligence (AI) algorithm demonstrated high accuracy in detecting prostate cancer (PCa) on biparametric MRI scans, performing comparably to experienced radiologists in identifying clinically significant lesions.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Multiparametric MRI (mpMRI) enhances prostate cancer (PCa) detection but faces interreader variability.
- Artificial intelligence (AI) models offer potential for consistent mpMRI interpretation.
- Large datasets and rigorous testing are crucial for AI model development in medical imaging.
Purpose of the Study:
- To assess an AI algorithm's capability for detecting and segmenting intraprostatic lesions on biparametric MRI.
- To compare the AI algorithm's performance against radiologist interpretations and histopathologic findings.
Main Methods:
- A secondary analysis of a prospective registry involving patients with suspected PCa.
- Evaluation of a cascaded deep learning algorithm on biparametric MRI scans.
- Comparison of AI performance (sensitivity, PPV, Dice similarity coefficient) with radiologist readings and biopsy outcomes.
Main Results:
- The AI algorithm detected 96% of participants with clinically significant PCa, comparable to radiologists (98%).
- The algorithm showed high sensitivity across different International Society of Urological Pathology (ISUP) grade groups (84%-98%).
- Lesion-level detection sensitivity was 55% with a PPV of 57%; lesion segmentation Dice similarity coefficient was 0.29.
Conclusions:
- The AI algorithm exhibits performance comparable to experienced radiologists in detecting cancer-suspicious lesions on biparametric MRI.
- The AI algorithm reliably identifies clinically significant prostate cancer lesions, aiding in diagnosis and management.
