ResDense Fusion: enhancing schizophrenia disorder detection in EEG data through ensemble fusion of deep learning models

Schizophrenia, a complex and debilitating mental disorder, affects approximately 1% of the global population. Diagnosing schizophrenia is challenging due to its heterogeneous symptomatology and lack of objective biomarkers. Electroencephalography (EEG) has emerged as a promising modality for investigating the underlying neurophysiological mechanisms of schizophrenia. In this study, we introduce a novel deep learning model called ResDense Fusion, which leverages residual connections for feature extraction, tackling the issue of vanishing gradients in deep networks. Additionally, during classification, each layer within ResDense Fusion is connected to every other layer in a feed-forward manner within a dense block. ResDense Fusion model integrates the hierarchical feature extraction capabilities of ResNet with the dense feature reuse mechanism of DenseNet, aiming to capture both low-level and high-level representations from EEG signals. By combining these architectures, the model can effectively learn discriminative features relevant to schizophrenia pathology. We evaluated the performance of the ResDense Fusion model on a dataset comprising EEG recordings from schizophrenia patients and healthy controls. Implemented in Python and results demonstrate the efficacy of the proposed model, achieving an impressive accuracy of 96%. The findings of this study highlight the potential of deep learning approaches in harnessing EEG data for schizophrenia diagnosis. The ResDense Fusion model not only offers a powerful tool for detecting schizophrenia but also provides insights into neurobiological underpinnings of the disorder. Ultimately, the development of accurate and efficient diagnostic tools like the ResDense Fusion model has the potential to improve early detection, treatment planning, and patient outcomes in schizophrenia management.