Machine learning classifiers can predict Gleason pattern 4 prostate cancer with greater accuracy than experienced...

Area of Science:

• Urology
• Radiology
• Machine Learning
• Oncology

Background:

• Accurate prostate cancer grading is crucial for treatment decisions.
• Distinguishing Gleason pattern 4 is key for risk stratification.
• Multiparametric MRI (mpMRI) shows promise in characterizing prostate tumors.

Purpose of the Study:

• To evaluate machine learning (ML) classifiers for predicting Gleason pattern 4 in prostate tumors.
• To compare ML classifier performance against expert radiologists' opinions.
• To determine if zone-specific (peripheral zone and transition zone) models improve prediction accuracy.

Main Methods:

• Retrospective analysis of prospectively acquired 3-T mp-MRI data (2012-2015).
• Inclusion of 164 men with biopsy-confirmed prostate cancer (Gleason 3+3 or higher).
• Development and validation of zone-specific ML classifiers using quantitative MRI features and clinical data, compared against three radiologists.

Main Results:

• The best peripheral zone (PZ) classifier achieved an AUC of 0.83, with a sensitivity of 0.93 at 50% specificity, outperforming radiologists (0.72).
• The best transition zone (TZ) classifier achieved an AUC of 0.75, with a sensitivity of 0.88 at 50% specificity, also outperforming radiologists (0.82).
• Models utilized features such as prostate-specific antigen density, apparent diffusion coefficient (ADC), and maximum enhancement (ME) on DCE-MRI.

Conclusions:

• Zone-specific machine learning classifiers can accurately predict Gleason pattern 4 in prostate tumors.
• These ML models demonstrate superior performance compared to the subjective assessment of experienced radiologists.
• The developed classifiers offer potential utility in active surveillance protocols for guiding biopsy decisions.