Personalized brachytherapy dose reconstruction using deep learning

Accurate calculation of the absorbed dose delivered to tumor and normal tissues improves treatment gain factor, which is major advantage brachytherapy over external radiation therapy. To address simplifications TG-43 assumptions that ignore dosimetric impact medium heterogeneities, we proposed a deep learning (DL)-based approach, accuracy while requiring reasonable computation time. We developed Monte Carlo (MC)-based personalized dosimetry simulator (PBrDoseSim), deployed generate patient-specific distributions. A neural network (DNN) was trained predict distributions derived from MC simulations, serving as ground truth. The paired channel input used for training composed distribution kernel in water along with full-volumetric density maps obtained CT images reflecting heterogeneity. predicted single-dwell kernels were good agreement MC-based reference, achieving mean relative absolute error (MRAE) (MAE) 1.16 ± 0.42% 4.2 2.7 × 10?4 (Gy.sec?1/voxel), respectively. MRAE volume histograms (DVHs) between DNN calculations clinical target 1.8 0.86%, 0.56 0.56%, 1.48 0.72% D90, V150, V100, For bladder, sigmoid, rectum, D5cc 1.7%, 1.9 1.3%, 2.1 DNN-based approach exhibited comparable performance method overcoming computational burden oversimplifications TG-43.