Region-based Prediction of Extraprostatic Extension Using a Machine Learning Approach Integrating MRI Radiomics and Clinical Data

Motivation: Current machine learning methods for predicting prostate cancer (PCa) extraprostatic extension (EPE) do not leverage spatial information, which could enhance prediction accuracy. Goal(s): To improve EPE prediction in PCa by integrating region-specific spatial information from mpMRI with clinical data, exploring how lesion location within the prostate affects predictive performance. Approach: We combined mpMRI radiomics features with clinical data, applied feature selection to optimize performance, and trained a classification model to predict EPE status in subjects undergoing radical prostatectomy. Results: Regional information can enhance EPE prediction, particularly for index csPCa lesions in the anterior or posterior prostate regions, which showed the highest performance metrics. Impact: This study highlights the potential of incorporating spatial characteristics of csPCa to enhance EPE prediction through precision imaging, integrating region-specific mpMRI radiomics features with clinical and histopathological parameters to guide more precise treatment decisions and interventions.