Machine Learning Techniques Reveal Aberrated Multidimensional EEG Characteristics in Patients with Depression

Abstract

Depression has become one of the most common mental illnesses, causing serious physical and mental harm. However, there remain unclear and uniform physiological indicators to support the diagnosis of clinical depression. This study aimed to use machine learning techniques to investigate the abnormal multidimensional EEG features in patients with depression. Resting-state EEG signals were recorded from 41 patients with depression and 34 healthy controls. Multiple dimensional characteristics were extracted, including power spectral density (PSD), fuzzy entropy (FE), and phase lag index (PLI). These three different dimensional characteristics with statistical differences between two groups were ranked by three machine learning algorithms. Then, the ranked characteristics were placed into the classifiers according to the importance of features to obtain the optimal feature subset with the highest classification accuracy. The results showed that the optimal feature subset contained 86 features with the highest classification accuracy of 98.54% ± 0.21%. According to the statistics of the optimal feature subset, PLI had the largest number of features among the three categories, and the number of beta features was bigger than other rhythms. Moreover, compared to the healthy controls, the PLI values in the depression group increased in theta and beta rhythms, but decreased in alpha1 and alpha2 rhythms. The PSD of theta and beta rhythms were significantly greater in depression group than that in healthy controls, and the FE of beta rhythm showed the same trend. These findings indicate that the distribution of abnormal multidimensional features is potentially useful for the diagnosis of depression and understanding of neural mechanisms.

Keywords

• depression
• electroencephalogram (EEG)
• machine learning
• feature selection
• functional connectivity
• power spectral density