Performance Evaluation of TQWT and EMD for Automated Major Depressive Disorder Detection Using EEG Signals

According to the World Health Organization, around 260 million people suffer from major depressive disorder (MDD). For screening the MDD, the electroencephalogram (EEG) signal is used extensively. The manual diagnosis of MDD utilizing EEG signals is very tedious and may lead to human error-prone. Therefore, various automated MDD systems have been developed nowadays for accurate and fast detection. This proposed framework presents a novel automated MDD detection method using EEG signals based on two wavelet transform methods: the tuned Q-wavelet transform (TQWT) method and empirical mode decomposition (EMD). First, the EEG signals are decomposed using both the transform methods from each channel. Second, seven non-linear features are derived from each decomposed signal. Third, a student t-test is applied to determine statistically significant features. In the end, the selected features are passed to several machine learning classifiers to evaluate the performance on a single channel. The classification performance is also evaluated by concatenating the best performing channel’s features. The classifier’s performance is optimized using different feature ranking methods. It is observed that using three channels, the proposed framework from the TQWT method achieves the highest classification accuracy of 99.30% using an ensemble classifier and outperforms the other existing methodologies. Hence, the proposed framework can be used to detect MDD using EEG signals for clinical purposes.