Parallel 3D brain modeling & feature extraction: ADNI dataset case study

In neuroimaging science, virtual brain modeling and feature extraction over medical scans may take very long execution time. In cases where the dataset contains too many MRI acquisitions, the total process may not be accomplished sequentially. In this study, the ADNI dataset samples containing 2292 brain imaging scans are virtually modeled for feature extraction. Using single computer, modeling one sample takes approximately 7 hours and all samples could be completed after about 2 years. In order to model all samples in a reasonable time, a parallel solution is proposed. An algorithm with Message Passing Interface library is developed. The proposed parallel algorithm runs on a Beowulf cluster of computers. The master computer in the cluster partitions the work pool, samples of the ADNI dataset, to parallel slave cluster computers. Data parallelism is achieved when each core runs Virtual 3D Modeling process for its own partial samples. All samples in the ADNI dataset are modeled using the 160-core cluster in Eskişehir Osmangazi University Beowulf Cluster Lab. At the end of the whole modeling process, the proposed parallel algorithm has a speedup factor that completes all workload in 7 days only.