Parallel Implementation Of The Boundary Element Method For Electromagnetic Source Imaging Of The Human Brain

Human brain functions are based on the electrochemical activity and interaction of the neurons constituting the brain. Some brain diseases are characterized by abnormalities of this activity. Detection of the location and orientation of this electrical activity is called electro-magnetic source imaging (EMSI) and is of signicant importance since it promises to serve as a powerful tool for neuroscience. Boundary Element Method (BEM) is a method applicable for EMSI on realistic head geometries that generates large systems of linear equations with dense matrices. Generation and solution of these matrix equations are time and memory consuming due to the size of the matrices and high computational complexity of direct methods. This study presents a relatively cheap and e ective solution the this problem and reduces the processing times to clinically acceptable values using parallel cluster of personal computers on a local area network. For this purpose, a cluster of 8 workstations is used. A parallel BEM solver is implemented that distributes the model eciently to the processors. The parallel solver for BEM is developed using the PETSc library. The performance of the iv solver is evaluated in terms of CPU and memory usage for di erent number of processors. For a 15011 node mesh, a speed-up eciency of 97.5% is observed when computing transfer matrices. Individual solutions can be obtained in 520 ms on 8 processors with 94.2% parallellization eciency. It was observed that workstation clusters is a cost e ective tool for solving complex BEM models in clinically acceptable time. E ect of parallelization on inverse problem is also demonstrated by a genetic algorithm and very similar speed-up is obtained.