Development and Validation of an Explainable Radiomics Model to Predict High-Aggressive Prostate Cancer: A...

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Artificial Intelligence in Medicine

Background:

• Prostate cancer (PCa) aggressiveness, categorized by Grade Group (GG), significantly impacts patient prognosis and mortality.
• Distinguishing between low-aggressive (GG ≤ 2) and high-aggressive (GG ≥ 3) PCa is crucial for effective treatment planning.

Purpose of the Study:

• To develop and externally validate a radiomic model for noninvasive classification of low- vs. high-aggressive prostate cancer.
• To leverage biparametric magnetic resonance imaging (bpMRI) for improved PCa risk stratification.

Main Methods:

• Retrospective analysis of 283 patients across four centers.
• Feature extraction from apparent diffusion coefficient (ADC) maps and T2-weighted (T2w) MRI sequences.
• Development and validation of a Naïve Bayes classifier using cross-validation and external validation sets.

Main Results:

• The radiomic model, based on ten features, achieved an AUC of 0.75 in the construction set and 0.73 in the external validation set.
• Shapley additive explanation (SHAP) values and partial dependence plots (PDP) provided interpretability for the radiomics signature.
• The Naïve Bayes classifier demonstrated robust performance in distinguishing PCa aggressiveness.

Conclusions:

• The developed radiomic model shows potential for noninvasively classifying prostate cancer aggressiveness.
• This approach could assist clinicians in managing men with suspected PCa, potentially improving decision-making.
• Further validation and integration into clinical decision support systems are recommended.