A Novel method for Schizophrenia classification using nonlinear features and neural networks

원문 초록

One notable method for recording brainwaves to identify neurological problems is electroencephalography (hereafter EEG). A trained neuro physician can learn more about how the brain functions through the use of EEGs. However conventionally, EEGs are only used to examine neurological problems (Eg. Seizures). But abnormal links to neurological circuits can also exist in psychological illnesses like Schizophrenia. Hence EEGs can be an alternate source of data for detection and classification of psychological disorders. A study on the classification of EEG data obtained from healthy individuals and individuals experiencing schizophrenia is conducted. The inherent nonlinear nature of brain waves are made use for the dimensionality reduction of the data. Nonlinear parameters such as Lyapunov exponent (LE) and Hurst exponent (HE) were selected as essential features. The EEG data was obtained from the openly available EEG database of MV. Lomonosov Moscow State university. To perform Noise reduction of the data, a more recently developed Tunable Q factor based wavelet transform (TQWT) is used . Finally for the classification, the 16 channel EEG time series is converted into spatial heatmaps using the aforementioned features. A convolutional neural network (CNN) is designed and trained with the modified data format for classification