A novel approach for afloat EEG channel selection and fusion: application in EEG schizophrenia detection

Schizophrenia (SZ) is an enduring, intricate, and debilitating neuropsychological illness. Traditional approaches for diagnosing SZ face several challenges, including the arduous and time-consuming nature of the examination process. Recently, the development of a pipeline for automatic, objective detection of SZ using electroencephalography (EEG) has become of great interest. Scholars have utilized all EEG channels or conventional channel selection techniques where an optimal brain channel is selected for the entire recording period. However, brain neurons continuously receive information from the surroundings or internals. As a result, different brain areas are activated over time. This study presents a novel approach to updating optimal selected/fused brain channels that float over time. Four strategies are proposed for dynamically selecting or integrating brain channels. After reducing the number of EEG electrodes, either by channel selection or fusion, EEG dynamicity is characterized by some indices of Poincaré plot asymmetry. The feature vector is then utilized in the principal component analysis, and the resulting outcome is fed into various machine learning algorithms to complete the SZ detection scheme. Our results show that floating brain channel selection or fusion could result in 100% classification accuracy. The highest classification performances are achieved by utilizing dynamic channel fusion. The process entails the calculation of the summation of EEGs in individual hemispheres, which is subsequently followed by the computation of the absolute difference between the obtained signals. The accuracy of our proposed system is superior or comparable to state-of-the-art EEG SZ detection tools.