A Tandem Model To Categorise Patients Requiring Clinical Corrections From EEG Signals

Abstract

Electroencephalograms (EEGs), which measure brain activity, are now the method most commonly used by doctors to diagnose epilepsy due to its inexpensive cost, ease of generation, and superior temporal resolution. The Electroencephalogram (EEG), which records brain electrical activity, is now the method that doctors use the most frequently to diagnose neurological illnesses. In this article, we provide an automated technique for detecting abnormal signals using recordings of raw EEG signals. A one-dimensional convolutional neural network (CNN) serves as the front end of the proposed system's preprocessing, and a support vector machine (SVM) serves as the back end. The method efficiently categorises unprocessed EEG signals without the extra work of feature extraction. The early detection of abnormal signals with the help of accurate predictive model with improved performance assisting psychiatrists in their diagnosis and so that the treatment could be started as soon as possible. An optimal combination of machine & deep learning models is designed to detect patients requiring clinical correction or not. HTER is the prime metric amongst other evaluated metrics as Accuracy, F1-score, Precision, Recall. The model runs over 10- fold cross validation for HTER for robust performance. This paper evaluates performance of Deep model with the proposed tandem model on various datasets on 5 performance metrics with 96.05% accuracy, 95.61%F1-score, 74% precision, 71% recall, 0.25±0.01 as HTER avg over 16 electrodes. The proposed framework's performance is validated using the public benchmark dataset Alzheimer's dataset, SNMC dataport. The experiment findings demonstrate that the developed framework outperforms established methods for EEG signal classification in terms of accuracy and F1-score, including numerous Machine Learning and Deep Learning algorithms.