Comparison of EEG indices after sleep deprivation in healthy subjects with EEG indices of patients with psychotic disorders

Abstract

Introduction: Sleep deprivation can lead to impairment in physical and mental function, unconsciousness, and damage to health. According to research, controlled sleep deprivation can cause behavioral and cognitive impairments on healthy people that are very similar to the symptoms of psychosis. Knowing that each of the behavioral and cognitive manifestations is due to the function of brain neurons and the result of the neuronal activity of the brain is the EEG signal. Therefore, it is expected that the similarity between healthy sleep-deprived people and psychotic patients is not limited to cognitive and behavioral disorders, and this similarity will cause similar changes in the signals of these two groups, which studies are showing some similarities. In this study, the EEG signal alteration of these two groups was investigated and the two groups were automatically classified using the extracted features. Materials and methods: The EEG signal power spectrum of 27 healthy individuals after normal sleep and after sleep deprivation and 27 psychotic patients were extracted after preprocessing. Using a convolutional neural network, these data, which are divided into 5-second windows, are classified and the results of this classification are evaluated by the k-fold cross-validation method. The result is compared with some traditional machine learning methods. The average power density spectrum density was also compared using Tukey's statistical method. Findings: The average accuracy value for data classification is 98.01% with a standard deviation of 1.2% and its sensitivity for psychotic patients is 99% and 97% for healthy sleep-deprived subjects, which is the highest accuracy among all machine learning and neural network methods. A comparison of the power spectrum of the signals of the two groups shows a significant difference in the power of the delta, theta, and beta subbands. In the female group, the delta, theta, and beta subbands showed a significant difference in power, while the males showed a significant difference only in the theta and beta subbands, as well as the whole frequency band. Conclusion: The EEG signal power spectrum characteristic is a suitable feature for classifying healthy sleep-deprived people from psychotic patients, despite the similarity in behavioral and cognitive manifestations, which its power has been proven using a powerful convolutional network. It is concluded that psychotic patients generally have more power in the delta, theta, and beta subbands than healthy sleep-deprived individuals, and there is no significant difference in the alpha and gamma subbands and the overall signal frequency range.