EEG entropies as estimators for the diagnosis of encephalopathy

Brain consists of a network of millions of neurons and the neural activities of the brain are clearly pictured in its signal, electroencephalogram (EEG). Many pathological conditions of brain can be studied in detail by inspecting the EEG signal in detail rather than just visual inspection. Non linear analysis has proved to be an efficient method for exploring the subtle information embedded in EEG. Approximate entropy and sample entropy are utilized in this study for comparing EEGs of patients with a neurological disease called encephalopathy, with normal EEGs. Both entropies were found to be significantly less (p < 0.01; independent sample t test) for encephalopathy group than normal healthy controls. Support vector machine, multilayer perceptron and random forest classifiers have been employed for identifying disease groups based on the EEG entropies and their performance were evaluated. Random forest classifier gave the maximum accuracy of 90% while multilayer perceptron and SVM classifier gave an accuracy of 87% and 84% respectively. The optimum performance was obtained by combining both approximate entropies and sample entropies as features to the classifiers, than using individual set of features. Thus, this work emphasizes that entropies of EEG are good bio-markers for the diagnosis of encephalopathy and that non linear analysis techniques should be employed for analyzing EEG signals.