Assessment of prostate cancer prognostic Gleason grade group using zonal-specific features extracted from...

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Machine Learning in Medicine

Background:

• Prostate cancer (PCa) aggressiveness grading is crucial for treatment decisions.
• Accurate assessment of PCa Grade Groups (GG) traditionally relies on histopathology.
• Non-invasive imaging biomarkers are needed for objective PCa grading.

Purpose of the Study:

• To automatically assess prostate cancer (PCa) lesion aggressiveness.
• To utilize zonal-specific image features from diffusion weighted imaging (DWI) and T2W MRI.
• To classify PCa lesions into prognostic Grade Groups (GG) using machine learning.

Main Methods:

• Regions of interest were extracted from DWI (peripheral zone) and T2W MRI (transitional zone, anterior fibromuscular stroma) of 112 PCa lesions from 99 patients.
• 38 histogram and texture features were extracted and used with a k-nearest neighbor classifier.
• A semi-exhaustive feature search and threefold stratified cross-validation were employed for classification into GG.

Main Results:

• Area Under the Curve (AUC) values for peripheral zone classification ranged from 0.87 (GG1) to 0.98 (GG3).
• AUC values for transitional zone and anterior fibromuscular stroma ranged from 0.83 (GG1+2) to 0.94 (GG3).
• The model demonstrated high accuracy in differentiating PCa lesions across various Grade Groups.

Conclusions:

• Zonal-specific imaging features from DWI and T2W MRI show significant potential for automated PCa aggressiveness assessment.
• These MRI-derived features can effectively differentiate between PCa lesions of varying aggressiveness (GG).
• The findings suggest a promising non-invasive approach for PCa grading, complementing histopathological evaluation.