Automated diagnosis of bipolar depression through Welch periodogram and machine learning techniques

Bipolar depression is a chronic mood disorder that causes severe shifts in mood, behaviors and energy levels. Very few studies to date have utilized these electroencephalogram (EEG) band abnormalities for the diagnosis of bipolar depression. Considering that bipolar depression affects millions of people around the world and its correct early diagnosis is of great importance in the treatment process. Considering the importance of EEG frequency bands in investigating the neuropathology of neuropsychiatric disorders, in this study, we intend to use a Welch periodogram and support vector machine (SVM) for the automatic diagnosis of bipolar depression from EEG signals. Therefore, in this research, 20 patients with bipolar depression aged 20 to 45 years and 25 healthy individuals with the same age range were subjected to electroencephalography in a resting state. After rejecting the artifact and reducing the noise in the preprocessing stage, the Welch periodogram was applied to the EEG and five statistical features (power, mean, variance, skewness and Shannon entropy) were calculated in delta, theta, alpha, beta and gamma frequency bands. SVM with linear, polynomial, RBF and sigmoid kernels, LDA, KNN, SOM and RBF were used as classifiers. The results showed that the features selected by statistical analysis and SOM neural network could achieve good accuracy, sensitivity and specificity of 97.62, 98.70 and 97.02%, respectively, in diagnosing bipolar depression. Considering that our simple system gives promising results in diagnosing patients with bipolar depression from EEG, there is scope for further work with a larger sample size and different ages and genders.