Prostate Cancer Risk Stratification and Scan Tailoring Using Deep Learning on Abbreviated Prostate MRI.

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Magnetic Resonance Imaging (MRI) is crucial for prostate cancer (PCa) management.
• Bi-parametric MRI (bpMRI) offers a faster, contrast-free alternative to multi-parametric MRI (mpMRI).
• Tailored MRI protocols based on individual risk may optimize resource utilization.

Purpose of the Study:

• Develop and evaluate a deep learning (DL) model for classifying clinically significant PCa (csPCa) using bpMRI.
• Assess the DL model's potential to optimize MRI protocol selection by recommending additional sequences only when beneficial.

Main Methods:

• A DL model was trained and validated on 26,129 prostate MRI studies.
• Evaluated on retrospective (n=151) and prospective (n=142) cohorts with ground-truth verification.
• Utilized a 3D ResNet-50 architecture for classification based on PI-RADS and Gleason scores.

Main Results:

• The DL model achieved an AUC of 0.83 in the prospective cohort (PI-RADS ≥3) and 0.86 in the retrospective cohort (Gleason ≥7).
• Demonstrated high sensitivity (93%) in both cohorts.
• Real-time implementation showed a processing latency of 14-16 seconds for protocol recommendations.

Conclusions:

• The developed DL model accurately identifies csPCa using bpMRI.
• The model integrates into clinical workflows for improved PCa detection and management.
• Potential for optimizing MRI protocol selection and resource allocation.