Intelligent and Automatic Treatment Planning System for Low-Dose-Rate Brachytherapy of Malignant Hepatic Tumors

The inefficient manual treatment planning of low-dose-rate brachytherapy for the malignant hepatic tumor is still the dominant clinical application. The purpose of our work is to develop an intelligent and efficient treatment planning system (TPS) that can segment multiple organ targets accurately and quickly, generate optimal seed implantation plans with fewer manual interactions, in the meantime, automatically avoid vital organ puncture. The TPS consists of three main modules, which are the abdominal multi-organ segmentation module, inverse dose planning module, and puncture pathway selection module. In the segmentation module, we adopt the latest deep learning-based model, which can automatically segment the hepatic tumor and 13 abdominal organs by training on public datasets and the datasets we collected. In the dose planning model, a novel parameterization strategy for the implantation plan is proposed. The parameterization strategy dramatically decreases the number of parameters needed to define the implantation plan and enables the fast simulated annealing algorithm to explore a possible solution. The puncture pathway selection is coupled within the optimization algorithm by rejecting clinically unacceptable ones when generating new potential solutions, which enables automatical puncture pathway selection during inverse planning. Hepatic malignant tumors cases are used to test the performance of our TPS, and the optimization time and cost value of each FSA iteration were recorded. The proposed method achieved ideal dose distribution and high conformity in the clinical practice within 1-4 minutes according to the size of the computational phantom and the number of used seeds. In addition, the consistency in the repetition test and the decreasing tendency of cost values from each iteration demonstrated the convergence of the algorithm. More importantly, our TPS can generate an ideal implantation plan in minutes without any medical physicists’ involvement.