GPU-accelerated solutions to forward problem of TMS

Abstract A forward model characterizes the relationship between source and a measurement in computational form. In TMS, such models are used for planning targeting stimulation. The contribution of head volume conductor to is typically solved using BEM or FEM. Here we consider context real-time TMS navigation performance benefits C++ GPU (Cuda) code over MATLAB. MATLAB highly efficient matrix computations but rather slow other computations, while GPUs excel massively parallel problems. We implemented linear Galerkin (LG) solvers Cuda languages benchmarked them against our optimized solver [1] constructing model. test had realistic gyral structure contained 21000 surface nodes 10200 cortical dipole triplets. On standard 2019 PC entry-level (Nvidia GTX 1060), built 75 seconds full potentials all triplets 15 C++/Cuda, compared 20 minutes 104 (not GPU). When this was simulation with 42-dipole coil, electric field 49 coil positions per second (cps); GPU-accelerated were as fast. With 15000-dipole model, C++/Cuda operated at speed 44 cps, 50 times slower (< 0.1 cps). navigation, use allows build whole 4-compartment defining regions interest during stimulation session, instead building offline saving loading 1.8-2.5 GB files. Further, enables practically any Stenroos Koponen, NeuroImage Research Category Technology Methods Basic Research: 10. Transcranial Magnetic Stimulation (TMS) Keywords: GPU,