Transfer learning with deep convolutional neural network for automated detection of schizophrenia from EEG signals

Schizophrenia (SZ) is a severe disorder of the human brain which disturbs behavioral characteristics such as interruption in thinking, memory, perception, speech and other living activities. If the patient suffering from SZ is not diagnosed and treated in the early stages, damage to human behavioral abilities in its later stages could become more severe. Therefore, early discovery of SZ may help to cure or limit the effects. Electroencephalogram (EEG) is prominently used to study brain diseases such as SZ due to having high temporal resolution information, and being a noninvasive and inexpensive method. This paper introduces an automatic methodology based on transfer learning with deep convolutional neural networks (CNNs) for the diagnosis of SZ patients from healthy controls. First, EEG signals are converted into images by applying a time-frequency approach called continuous wavelet transform (CWT) method. Then, the images of EEG signals are applied to the four popular pre-trained CNNs: AlexNet, ResNet-18, VGG-19 and Inception-v3. The output of convolutional and pooling layers of these models are used as deep features and are fed into the support vector machine (SVM) classifier. We have tuned the parameters of SVM to classify SZ patients and healthy subjects. The efficiency of the proposed method is evaluated on EEG signals from 14 healthy subjects and 14 SZ patients. The experiments showed that the combination of frontal, central, parietal, and occipital regions applied to the ResNet-18-SVM achieved best results with accuracy, sensitivity and specificity of 98.60% +/- 2.29, 99.65% +/- 2.35 and 96.92% +/- 2.25, respectively. Therefore, the proposed method as a diagnostic tool can help clinicians in detection of the SZ patients for early diagnosis and treatment.