Improved empirical wavelet transform combined with particle swarm optimization-support vector machine for EEG-based depression recognition

Depression not only inflicts physical harm on patients and diminishes their quality of life, but also imposes a significant burden on families and society. Current diagnostic methods are predominantly employed post-onset, leading to a lack of early intervention opportunities for patients. Therefore, there is a pressing need to develop techniques for detecting early signs of depression to enable timely intervention and potentially improve recovery rates. In this paper, we propose an improved method for the early objective diagnosis of depression utilizing an empirical wavelet transform (EWT) technique enhanced by a particle swarm optimization-support vector machine (PSO-SVM) algorithm. Our approach specifically focuses on the Fpz channel in the prefrontal lobe of the brain, which most accurately reflects the electrical anomalies associated with depression among 128 channels of resting-state electroencephalogram (EEG). The EWT is refined based on the Morlet wavelet, which allows for the precise decomposition of EEG rhythms. From these decompositions, we effectively extract six depression-related EEG features: frequency band power, frequency band power ratio, Shannon entropy, permutation entropy, LZ complexity, and variance. Afterward, these distinguishing characteristics are harnessed to detect depression through the optimized PSO-SVM algorithm. Our approach exhibited a accuracy rate of 81.25% on the MODMA publicly accessible dataset, thereby validating its proficiency in assisting in the diagnosis of depression via the analysis of the EEG Alpha band.