Exploring Deep Convolution Neural Networks with Transfer Learning for Rectum Toxicity Prediction in Cervical Cancer Radiotherapy

Abstract

Despite benefits of chemoradiation for cervical cancer, the local control of advanced stage bulky cervical tumors(LACC) is determined by delivering adequate Brachytherapy dose.This is essentially limited by the toxicity of surrounding organ at risk (OARs) especially the rectum. Better understanding of the OARs dose distribution-toxicity relationship is critical for safe dose escalation to improve tumor local control and treatment outcome. We sought to clarify the rectal dose spatial distribution- brachy toxicity relationship by employing 1) an accurate deformable registration to obtain precise accumulated dose map and 2) a deep convolutional neural network (CNN) with transfer learning to establish a dose distribution-toxicity prediction model.