Related Experiment Video
Updated: Oct 18, 2025
06:08
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
480
A Cascaded Deep Learning-Based Artificial Intelligence Algorithm for Automated Lesion Detection and Classification on
Sherif Mehralivand1, Dong Yang2, Stephanie A Harmon1
1Molecular Imaging Branch, National Cancer Institute, National Institutes of Health, 10 Center Dr., MSC 1182, Building 10, Room B3B85, Bethesda, Maryland.
Academic Radiology
|October 2, 2021
Summary
This study developed an AI system using deep learning to detect and classify prostate cancer lesions on MRI scans. The AI showed good detection but moderate classification performance, aiding radiologists in prostate cancer diagnosis.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Prostate MRI is crucial for detecting significant prostate cancer but faces diagnostic performance variability.
- Artificial intelligence (AI) offers potential to enhance radiologist accuracy in prostate lesion detection and classification.
Purpose of the Study:
- To develop and evaluate a cascaded deep learning system for detecting and classifying prostate lesions on biparametric MRI.
- To assist radiologists in interpreting prostate MRI using PI-RADS (Prostate Imaging Reporting and Data System) categories.
Main Methods:
- A cascaded deep learning system utilizing 3D U-Net and residual neural networks was trained on biparametric MRI scans from two institutions.
- The system was trained for lesion detection, segmentation, and PI-RADS classification (categories 2-5 and BPH).
- Performance was assessed using sensitivity, PPV, FDR, accuracy, and Dice similarity coefficient on an independent test set.
Main Results:
- The AI system achieved a lesion-level sensitivity of 56.1% and a positive predictive value (PPV) of 62.7%.
- False discovery rate (FDR) was 37.3%, with a median Dice similarity coefficient (DSC) of 0.359 for lesion segmentation.
- Overall PI-RADS classification accuracy was 30.8%.
Conclusions:
- The developed cascaded deep learning architecture demonstrates capability in detecting and classifying prostate cancer-suspicious lesions on MRI.
- The system provides good detection metrics and reasonable classification performance, offering potential as a radiologist aid.
