Multimodal Deep Learning Approaches for Early Detection and Classification of Psychiatric Disorders Using Neurophysiological Signal Analysis

Abstract

newline This study presents advanced artificial intelligence frameworks for the accurate prediction and classification of neurological and psychiatric disorders using multimodal neurophysiological data. The work introduces two novel deep learning architectures: PsyneuroNet and CardioNeuroFusionNet. newlineThe PsyneuroNet architecture employs cross-normalization across convolutional layers, a bidirectional long-short-term memory network (Bi-LSTM), and an enhanced inception-based convolutional neural network (ICNN). It demonstrates high prediction accuracies of 99.65%, 97.27%, and 98.08% on CHB-MIT, DEAP, and OpenNeuro datasets, respectively. PsyneuroNet effectively classifies various neurological and psychiatric conditions, including mental illnesses such as depression and sleep disorders, and psychiatric disorders such as schizophrenia and Alzheimer s disease, using electroencephalography (EEG) signals, particularly emphasizing the detection of comorbid symptoms within epilepsy patients. newlineCardioNeuroFusionNet advances diagnostic accuracy through multimodal data fusion, integrating EEG, magnetoencephalography (MEG), and electrocardiography (ECG) signals within an enhanced CNN-Bi-Transformer framework. This model notably surpasses existing AI techniques by an average accuracy improvement of 2.72%, achieving an accuracy of 98.54% and sensitivity of 97.77% on the Deep BCI Scalp Database and Kymata Atlas datasets. The integration of cardiac and neurological data offers deeper insights into the physiological interactions underlying mental health conditions, providing potential pathways to more personalized and accurate treatments. newlineBoth frameworks significantly contribute to the field by addressing common limitations of prior models, such as limited generalizability, lack of multimodal integration, and low interpretability. These advancements underline the efficacy of deep learning methodologies in providing precise, scalable, and interpretable tools for early and reliable diagnosis of mental illnesses, enhancing patient care outcomes and supporting clinical decision-making. newlineKeywords: Computer-Aided Diagnosis, Electroencephalography, Psychiatric Disorders, Convolutional Recurrent Neural Network, Neurocardiac data fusion. newline