Peritumoral Radiomics Strategy Based on Ensemble Learning for the Prediction of Gleason Grade Group of Prostate...

Area of Science:

• Radiology
• Machine Learning
• Oncology

Background:

• Prostate cancer (PCa) grading is crucial for treatment decisions.
• Differentiating low-Gleason grade group (L-GGG) from high-GGG (H-GGG) PCa non-invasively remains a challenge.

Purpose of the Study:

• To develop and evaluate a peritumoral radiomic-based machine learning model for differentiating L-GGG and H-GGG prostate lesions.
• To assess the model's performance in predicting PCa aggressiveness.

Main Methods:

• Retrospective study of 175 PCa patients (59 L-GGG, 116 H-GGG).
• Radiomic features extracted from T2WI, DWI, and ADC maps for original tumor, central-tumoral, and peritumoral regions of interest (ROIs).
• Peritumoral radiomics models developed for peripheral zone (PZ) and transitional zone (TZ) datasets; performance evaluated using ROC and precision-recall curves.

Main Results:

• The combined peritumoral radiomics model (T2+DWI+ADC) outperformed other models with an AUC of 0.850 and accuracy of 0.950.
• Peritumoral models showed superior performance over regional models (e.g., PZ: AUC 0.85 vs. 0.75; TZ: AUC 0.88 vs. 0.69).
• Models demonstrated higher efficacy in predicting PZ lesions compared to TZ lesions.

Conclusions:

• Peritumoral radiomics features demonstrate excellent performance in predicting Gleason grade group in PCa patients.
• This approach may serve as a valuable non-invasive tool for assessing PCa aggressiveness.