Parallelization of Classification Algorithms for Medical Imaging on a Cluster Computing System

Abstract In this paper we examine the Single Program Multiple Data (SPMD) parallel implementation of image classification algorithms on a cluster comprised of personal computers. The small-scale cluster environment employed utilizes two quite different Application Programming Interfaces (APls) for inter-process communications, message passing and virtual shared memory. We quantitatively compare both of these communication approaches in conjunction with a small scale cluster for medical image classification by presenting the SPMD parallelization of three well-known context-independent image classification algorithms: Nearest Mean, Maximum Likelihood and K Nearest Neighbors. These classic approaches are applied to massive medical images and the resulting average speedup using both message passing and virtual shared memory inter-process communications is presented.