A Deep Learning Dose Prediction Algorithm Based On Local Self-attention Transformers

Background:Cancer,also known as tumor,is a term for a group of diseases in which certain cells in an organ proliferate uncontrollably and invade adjacent tissues and organs.Cancer is a leading cause of death worldwide,with an estimated 19.3 million new cases and nearly 10 million deaths in 2020.Intensity-modulated radiation therapy(IMRT)has been the standard treatment protocol for various treatment sites as it allows a conformal dose distribution for target volumes with complex shapes.However,treatment planning for IMRT is a time-consuming and labor-intensive process,as planners need to repeatedly adjust a number of parameters in treatment planning system(TPS)to determine the intensities of beamlets.Purpose:To alleviate the tedious planning process of IMRT,this subject explored the application of new artificial intelligence techniques for radiation treatment planning and developed a novel deep-learning based dose prediction algorithm(TrDosePred).Methods:This study is based on head and neck cancer cases and the dataset from the Open Knowledge-Based Planning Challenge(OpenKBP).The OpenKBP dataset includes 340 patients treated by 6MV IMRT with nine equispaced coplanar beams.Each patient has at least one PTV,at most seven OARs and a dose distribution generated by a 3D generative adversarial network(GAN)and the Computational Environment for Radiotherapy Research(CERR).An algorithm based on the transformers has been explored for the dose prediction task in treatment planning.The proposed TrDosePred,which generated the dose distribution from a contoured CT image,was a U-shape network.A few of convolutions are introduced to interleave with the transformers to improve the optimization stability and peak performance.Data augmentation and ensemble approach were used to improve the generalization and robustness.The performance of TrDosePred was evaluated with two mean absolute error(MAE)based scores utilized by OpenKBP challenge(i.e.,Dose score and DVH score)and compared to the top three approaches of the challenge.In addition,several state-of-the-art methods were re-implemented and compared to TrDosePred.Results:The TrDosePred ensemble achieved the dose score of 2.426 Gy and the DVH score of 1.592 Gy on the test dataset.In terms of DVH metrics,on average,the relative MAE against the clinical plans was 2.25%for targets and 2.17%for organs at risk.Conclusions:A transformer-based framework TrDosePred was developed for dose prediction.The results showed a comparable or superior performance as compared to the previous state-of-the-art approaches,demonstrating the potential of transformer to boost the treatment planning procedures.