Cluster computing-based EEG sub-band signal extraction with channel-wise and time-slice-wise data partitioning technique

Cluster computing provides an effective approach in implementing parallel and distributed processing using existing computing resources of any institute or organization. In this objective, the present work is focused on the investigation of cluster computing to enhance the execution time of electroencephalography (EEG) signal processing, for the extraction of alpha, beta, theta, and delta components from a 256-channel Steady-state visually Evoked Potential (SSVEP) dataset of 11 subjects. Here, we have compared the execution time of two cluster computing frameworks with proposed channel-wise and Time-slice-wise data partitioning approaches and evaluate the signal extraction performance of FastICA, Infomax, and Picard methods individually. In this study, we found that the speed-up of the Transmission Control Protocol (TCP) based cluster is significantly improved, compared to the User Datagram Protocol (UDP) based cluster, achieved more than 90% percentage time saving with compare to sequential processing in the channel-wise and Time-slice-wise data partitioning with TCP and UDP based cluster. Additionally, the Infomax and Picard methods archived excellent signal extraction quality, while the FastICA method exhibits required lower execution time and achieve a high speed-up. It is also observed that the Time-slice-wise data partitioning and channel-wise partition strategy, although effective in parallel execution, but Time-slice-wise data partitioning needs more execution time compared to the channel partition strategy. Here, the present work is an attempt to utilize the potential benefits of cluster computing frameworks for EEG signal processing.