A Fast Deep Learning Approach for Beam Orientation Optimization for Prostate Cancer IMRT Treatments

Abstract

We propose a fast beam orientation selection method, based on deep neural networks (DNN), capable of developing a plan comparable to those by the state-of-the-art column generation method. The novelty of Our model lies in its supervised learning structure, the DNN architecture, and ability to learn from anatomical features to predict dosimetrically suitable beam orientations without using the dosimetric information from the candidate beams, a time consuming and computationally expensive process. This may save hours of computation. A supervised DNN is trained to mimic the column generation algorithm, which iteratively chooses beam orientations by calculating beam fitness values based on the KKT optimality conditions. The dataset contains 70 prostate cancer patients. The DNN trained over 400 epochs, each with 2500 steps, using the Adam optimizer and a 6-fold cross-validation technique. The average and standard deviation of training, validation, and testing loss functions among the 6-folds were at most 1.44%. The differences in the dose coverage of PTV between plans generated by column generation and by DNN were 0.2%. The average dose differences received by organs at risk were between 1 and 6 percent: bladder had the smallest average difference, then rectum, left and right femoral heads. The dose received by body had an average difference of 0.1%. In the training phase of the proposed method, the model learns the suitable beam orientations based on the anatomical features of patients and omits time intensive calculations of dose influence matrices for all possible candidate beams. Solving the Fluence Map Optimization to get the final treatment plan requires calculating dose influence matrices only for the selected beams. The proposed DNN is a fast beam orientation selection method based that selects beam orientations in seconds and is therefore suitable for clinical routines.