Related Experiment Video
Updated: Dec 20, 2025
06:08
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
940
An Automated Two-step Pipeline for Aggressive Prostate Lesion Detection from Multi-parametric MR Sequence.
Josh Sanyal1, Imon Banerjee1,2,3,4, Lewis Hahn3
1Department of Biomedical Data Science, Stanford University, Stanford, CA.
Summary
An automated approach using multi-parametric MRI can accurately assess prostate cancer aggressiveness at the pixel level. This method aids in reducing overdiagnosis and overtreatment by providing interpretable, localized analyses for clinicians.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Prostate cancer overdiagnosis and overtreatment stem from challenges in assessing tumor aggressiveness.
- Multi-parametric MRI (mpMRI) offers detailed lesion characteristics but manual interpretation is expert-dependent, time-consuming, and prone to inter-observer variability.
Purpose of the Study:
- To develop a fully automated computational approach for pixel-level prostate cancer aggressiveness assessment in mpMRI.
- To enhance diagnostic accuracy and reduce inter-observer variability in prostate cancer grading.
Main Methods:
- Combined traditional computer vision and deep learning algorithms for automated analysis.
- Utilized specific MRI sequences, including Apparent Diffusion Coefficient (ADC) and Diffusion-Weighted Imaging (DWI).
- Focused on pixel-level classification without requiring manual feature extraction, segmentation, or prior lesion detection.
Main Results:
- Achieved an Area Under the Receiver Operating Characteristic Curve (ROC-AUC) of 0.86 for diagnosing aggressive prostate lesions.
- Achieved an ROC-AUC of 0.88 for diagnosing non-aggressive prostate lesions.
- Demonstrated interpretable pixel-level classifications enabling localized lesion analysis.
Conclusions:
- The proposed automated model effectively assesses prostate cancer aggressiveness from mpMRI data.
- This approach has the potential to improve diagnostic consistency and support clinical decision-making, thereby mitigating overtreatment.
- Pixel-level classification provides localized insights crucial for precise cancer management.
