Computational Acceleration for Medical Treatment Planning

In medicine, there is a clear trend towards individualized therapies, for cancer and other diseases. Individualized treatment planning for cancer, particularly in radiotherapy and light therapies, is a complex optimization problem. As analytical inverse planning solutions do not exist for light therapies, a large number of light delivery configurations must be evaluated to find one that best conforms to the clinical target (e.g., a tumour). An integral part of this optimization is the accurate computation of light dose, ideally using Monte Carlo (MC) simulations for realistic, 3-D modelling. This text explores two hardware-accelerated solutions to overcome the general speed limitation of MC simulations: (1) designing custom hardware on field-programmable gate arrays, and (2) creating highly parallel software on graphics processing units (GPUs). Notably, a speedup of over 1000x was achieved on four GPUs compared to a state-of-the-art CPU. As the Monte Carlo method is used in many fields such as radiation medicine, this text also includes the GPU MC code package and is of interest to scientists, engineers, and medical professionals exploring real-time treatment planning solutions.

William C. Y. Lo joined the Harvard-MIT HST MD program after obtaining the BASc and MSc degrees from the University of Toronto. Dr. Rose is a Professor in Electrical and Computer Engineering at the University of Toronto. Dr. Lilge is a Professor in Medical Biophysics at the University of Toronto and Senior Scientist at Princess Margaret Hospital.