MASK-FREE RADIOTHERAPY DOSE PREDICTION VIA MULTI-TASK LEARNING

In radiotherapy, an ideal treatment plan always takes the treatment planner several days to tweak repeatedly, owing to the complex dose distribution. To facilitate this process, many deep-learning-based studies have been devoted to automatic prediction of dose distribution. However, besides the computed tomography (CT) image, these methods usually require extra segmentation masks of target tumor and organs at risk as input, and such annotations are quite timeconsuming to acquire. In this paper, we propose a mask-free dose prediction model based on the multi-task learning, which only needs CT images as input and outputs dose distribution maps efficiently. Specifically, considering the high correlation between the tumor anatomical structure and the dose distribution, we develop a multi-task architecture consisting of 1) a primary dose prediction task aiming to generate accurate dose distribution map, and 2) an auxiliary segmentation task applied to provide anatomical information for the primary task. Experiments on an in-house dataset demonstrate the effectiveness of our method.