The design & fabrication of an epileptic seizure detection watch
This paper describes the design and fabrication of an epileptic seizure detection watch for the timely detection of Generalized Tonic-Clonic (GTC) seizures; using skin conductance (SC) signals. The watch’s circuit was designed in EasyEDA and implemented on a Breadboard to showcase the dispatch of a seizure event alert to a phone via a Bluetooth module; in the event of an ongoing seizure and vice versa. Due to the unavailability of SC signal databases, Electroencephalography (EEG) signals, acquired from a physiological database known as PhysioNet were used in showcasing the signal processing of incoming SC signals, temporal and spectral feature extraction of these signals, and the classification of these signals using a trained machine learning algorithm. Twenty-five machine learning algorithms provided by the MATLAB Classification Learner App were trained using 80 EEG signals (both seizure and non-seizure) and only two algorithms, namely the Medium Tree and Linear Support Vector Machine (SVM) had the highest training prediction accuracy. However, in determining their prediction accuracy with two different data sets, the Medium Tree model had the highest cumulative prediction accuracy of 76.7%; as compared to the Linear SVM model which had a cumulative prediction accuracy of 73.3%. Based on these results, the Medium Tree model was recommended as a good seizure detection algorithm to prevent fatal and non-fatal injuries; and even Sudden Unexpected Death in Epilepsy (SUDEP).
Capstone submitted to the Department of Engineering, Ashesi University in partial fulfilment of the requirements for the award of Bachelor of Science degree in Electrical & Electronic Engineering, April 2019.
epileptic seizure, detection watch, seizure detection watch, Generalized Tonic-Clonic seizure, EasyEDA, Electroencephalography, Classification Learner App, Support Vector Machine