Machine Learning Optimization for Prostate Brachytherapy Treatment Planning

Abstract

Prostate Low-Dose-Rate brachytherapy (LDR) is one of the most effective treatments for localized prostate cancer. Machine Learning (ML), the application of statistics to complex computational problem solving, was applied to prostate LDR brachytherapy treatment planning. Planning time, pre-implant dosimetry, and various measures of clinical implant quality for ML plans were compared against plans created by expert brachytherapists. The average planning time to create an ML plan was 0.84 _ 0.57 min compared to over 17.88 _ 8.76 min for an experienced brachytherapists. Dosimetry was not significantly different for ML and expert brachytherapist plans. Clinical implant quality for the ML plans were ranked as nearly equivalent to the brachytherapist treatment plans in all qualitative categories evaluated. The results of this thesis demonstrate that it is possible to generate high quality prostate brachytherapy treatment plans with comparable quality to those of a human expert using a custom ML algorithm.