Computer-aided Ischemic Stroke Classification from EEG Data Using a Single-tiered Spiking Neural Network Framework

摘要

Ischemic stroke is one of the most common cerebrovascular conditions, and constitutes a significant portion of global mortality rates. Early diagnoses are vital for successful recoveries, but with conventional diagnostic imaging techniques and computer systems, radiologists misdiagnose more than 20% of all ischemic strokes. AI methods have been developed for automated neurological disorder prediction using electroencephalographic (EEG) data, but artificial and recurrent classifiers have still seen mediocre performance. Spiking neural networks (SNNs), however, have demonstrated their capacity in personalized spatio-and spectro-temporal non-stationary time series data modeling and pattern recognition. This study developed a SNN model that can detect cerebral ischaemia from temporal EEG data with an accuracy of 94.45%, far exceeding the performance of traditional stroke models operating on computerized tomography data. The approach outlined also required a significantly low number of training samples, having been trained and evaluated on an EEG corpus with recordings from 46 stroke patients and 46 healthy individuals, in addition to being scalable to other neurodegenerative diseases and mental illnesses. Ultimately, the diagnostic precision of the SNN can be adapted in professional environments to replace or be in conjunction with other computerized medical systems to improve ischemic stroke prognosis and recovery.