Computer Aided Tool for diagnosing Epilepsy using Kolmogorov Complexity and Approximate Entropy

Abstract

One of the most common neurological disorders in human beings is Epilepsy which is known to cause unprovoked seizures and convulsions. Electroencephalogram (EEG) readings which capture the signals that are transmitted between the neurons across various parts of the brain can help in diagnosing Epileptic seizures which is different from normal controls due to topological, structural, and network changes. Features like Approximate Entropy and Kolmogorov Complexity are extracted from the readings captured by EEG electrodes. These readings act as inputs to the five-layered Back Propagation Multi-Layer Perceptron Neural Network in performing training and testing in order to classify the patients suffering from Epilepsy and normal controls. Initially, this methodology is applied to readings from all the 14 electrodes that are available from the database resulting in Accuracy of 96.5 %, Precision of 98.1 %, Sensitivity of 95%, and Specificity of 98% with Area Under the Curve (AUC) of 0.964. Since the data from 14 electrodes consume a lot of storage space and time for calculation and analysis, the subsets of EEG electrodes F7, F8, FC5, FC6, T7, T8 which are placed over the temporal region of the brain which is mainly affected during seizures is considered and when the same methodology is applied on it, results in Accuracy of 97 %, Precision of 95.5 %, Sensitivity of 99 %, and Specificity of 94.5 % with AUC 0.967. In both the cases, their classification performance is almost equal but the storage space and the time taken for calculation in the second case are comparatively lesser than the first case due to less number of EEG electrodes involved. This can help in the faster diagnosis of Epilepsy in patients.