Machine Learning for the Prediction of Prostate Cancer Biopsy Based on 3D Dynamic Contrast-Enhanced Ultrasound Quantification

Abstract

Non-targeted transrectal-ultrasound-guided 12-core systematic biopsy (SBx) is the current guideline-recommended clinical pathway for prostate cancer (PCa) diagnosis, despite being associated with a risk of complications as well as un-derdiagnosis or overtreatment. Quantification algorithms for dynamic contrast-enhanced ultrasound (DCE-US) have shown good potential for PCa localisation in two dimensions (2D), and a few have recently been expanded to 3D. In this work, we present a 3D implementation of all estimators in the contrast ultrasound dispersion imaging (CUDI) family and exploit combinations of the extracted parameters to predict individual SBx-core outcomes. We show that machine-learning approaches can improve the classification performance compared to individual CUDI parameters and foresee potential for further development in image-based PCa localisation.