Parallel Processing of Intra-cranial Electroencephalogram Readings on Distributed Memory Systems

This article presents an approach for parallel processing of electroencephalogram readings over distributed memory systems. This is a complex problem that deals with a significantly large amount of data, especially considering that the volume of electroencephalogram readings has been growing for the last few years due to their handling in medical and health applications. Different parallelization and workload distribution techniques applied to processing intra-cranial electroencephalogram readings are studied, in order to efficiently detect whether a patient may suffer a seizure or not. More precisely, two separate approaches are presented: a first one describing a traditional Message Passing Interface implementation for cluster systems, and a second implementation using Apache Hadoop, more adapted to large-scale processing in cloud systems. The experimental evaluation performed on standard datasets demonstrates that it is possible to remarkably speedup electroencephalogram processing by applying efficient data distribution strategies. The parallel/distributed approach allows accelerating the execution time up to 22x when compared with the sequential version.