Effect of combining features generated through non-linear analysis and wavelet transform of EEG signals for the diagnosis of encephalopathy. (20th November 2021)

• Record Type: Journal Article
• Authors: Jacob, Jisu Elsa; Chandrasekharan, Sreejith; Nair, Gopakumar Kuttappan; Cherian, Ajith; Iype, Thomas
• Is Part Of: Neuroscience letters. Volume 765(2021)
• Publication Date: 2021-11-20
• Subjects: Electroencephalogram -- Encephalopathy -- Discrete wavelet transform -- Gini impurity score -- Support vector machine -- Random forest; Neurology -- Periodicals; Neurology -- Periodicals; Research -- Periodicals; Neurologie -- Périodiques; Neuroanatomie -- Périodiques; Neuropharmacologie -- Périodiques; Neurophysiologie -- Périodiques; Neurology; Periodicals; Electronic journals; 617.48
• DOI: 10.1016/j.neulet.2021.136269
• Languages: English
• ISSNs: 0304-3940

Abstract

Highlights: EEG signals portray dynamic, complex, chaotic and nonlinear nature of the brain. EEG signals are represented with chaotic features, fractal dimensions and entropy features. Combining EEG features compresses information and improve discriminative power. Gini importance score comparison is useful to ensure removal of redundant features. Random Forest classifier provides an optimum model for encephalopathy diagnosis. Abstract: Electroencephalogram (EEG) signals portray hidden neuronal interactions in the brain and indicate brain dynamics. These signals are dynamic, complex, chaotic and nonlinear, the nature of which is represented with features - fractal dimensions, entropies and chaotic features. This study aims at examining the discriminative power of individual features and their combination in the diagnosis of a neuro-pathological condition called encephalopathy. Feature combination is accomplished with the help of feature selection using Gini impurity score that improves discriminative power and keeps redundancy minimal. Further, three widely used non-parametric classifiers which are known to be effective with wavelet features on EEG signals — Support Vector Machine, Random Forest, Multilayer Perceptron — are employed for disease classification. The models created by the combination of aforementioned stages are analysed and evaluated with performance scores, leading to an optimal model for automated diagnostic applications.