EEG Functional Connectivity and Deep Learning for Automated Diagnosis of Alzheimer's disease and Schizophrenia

Abstract

Listen

Translate article

Electroencephalogram (EEG) functional connectivity analysis provides important clues about brain network abnormalities, an important approach to diagnose complex neurological diseases such as Alzheimer’s disease and schizophrenia. Advanced computational analysis can effectively analyze disorders with unique disruptions in neural connectivity. Deep learning (DL) is one of these, and has emerged as a powerful tool to facilitate automation in diagnostic processes and accurate classification by the use of DL models. The application of DL techniques and EEG functional connectivity metrics for the automated diagnosis of Alzheimer’s disease and schizophrenia is investigated in this study. For analysis, EEG data from patients with these disorders were used. To quantify the interregional synchronization of neural activity, functional connectivity metrics, such as coherence and phase locking value were extracted. Convolutional Neural Networks (CNN) and Long Short Term Memory (LSTM) networks based multi class classification framework was designed to detect patterns related with the disorders. Results demonstrated DL framework performance at 94% for Alzheimer’s disease and 91% for schizophrenia. The DL models were then found to robustly replicate such inter-regional disruptions, with connectivity patterns analyzed via connectivity maps, revealing distinct inter-regional patterns in both conditions. This has also been demonstrated by the superior performance of DL methods in processing EEG data with complex and high dimensionality, and in extracting informative features for diagnosis. Finally, EEG functional connectivity metrics and DL methods greatly increase diagnostic accuracy for Alzheimer’s disease and schizophrenia. These findings point towards the transformative power of AI driven solutions in clinical diagnostics to achieve scalability and efficiency in neurological disorder diagnosis. Future research should be directed towards gap expanding application level of these models to other neurological conditions, and refinement of frameworks that can be implemented in a clinical setting.