Explainable AI Radiomics in Prostate Cancer Aggressiveness Prediction using different quantitative Diffusion MRI...

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Prostate cancer (PCa) diagnosis requires accurate characterization for effective patient management.
• Distinguishing indolent from aggressive PCa early is a critical unmet need.
• Current methods may lead to variability in patient stratification and unnecessary procedures.

Purpose of the Study:

• To develop an automated method for classifying Gleason score (GS) in PCa using quantitative diffusion MRI radiomics.
• To assess the performance of T2-weighted and diffusion MRI models in predicting PCa aggressiveness (GS<7 vs. GS≥7).
• To reduce the rate of unnecessary prostate biopsies through improved early characterization.

Main Methods:

• Retrospective analysis of 202 histopathologically proven PCa patients.
• Quantitative diffusion MRI modeling and radiomics applied to T2 and diffusion data.
• Classification models trained and evaluated, with Shapley Additive Explanations (SHAP) for model interpretability.
• Intravoxel Incoherent Motion (IVIM) model used to derive parametric maps, including micro-perfusion fraction.

Main Results:

• The best performing model combined T2 imaging with the diffusion-derived micro-perfusion fraction from the IVIM model.
• This combined model achieved a mean accuracy of 80.91% and an Area Under the Curve (AUC) of 85.29%.
• Tissue structural information and blood microperfusion were identified as significant predictors of PCa aggressiveness.

Conclusions:

• Quantitative diffusion MRI radiomics, particularly when combined with T2 imaging and IVIM-derived parameters, offers a promising automated approach for PCa aggressiveness classification.
• This method has the potential to improve the accuracy of PCa staging and reduce inter-center variability.
• The findings support the use of advanced MRI techniques to guide clinical decisions and minimize invasive procedures like biopsies.