EEG based detection of schizophrenia using asymmetry of entropy and CNN-LSTM model

Schizophrenia is a severe neuropsychiatric disorder with a significant impact on individual's real-life functioning. It is characterized by abnormal asymmetry in the neural activities of the brain reflecting functional and cognitive impairment. The irregularities in the neural dynamics are well captured by Electroencephalogram (EEG) based complexity measures. In this work, automated detection of schizophrenia is attempted using EEG based asymmetric entropy analysis and convolutional neural networks (CNN) integrated with Long Short Term Memory (LSTM) classification model. The asymmetric entropy feature maps are extracted from EEG frequency bands across all channel pairs using approximate, differential, sample, Shannon and spectral co-occurrence matrix entropies and are subjected to classification using pre-trained Inception-V3 CNN-LSTM model and the performance measures are evaluated. It is found that the magnitude values of approximate and sample entropies are found to be high when compared to other entropies and exhibit significant discrimination between normal and schizophrenic subjects. It is also found that statistically significant inception features derived from the inter-channel asymmetric feature maps yield high values of accuracy, precision, and F1 score across various frequency bands. It is further observed that high classification accuracy of 94.11% and precision of 100% are obtained for delta band. The classification model utilizing inter-channel asymmetries could capture the functional alterations due to the pathological condition and helps in accurate detection of schizophrenia.