Abstract

It is a mental illness with cognitive process disruption, self-monitoring impairment, and functional brain connectivity abnormalities. Electroencephalography is a non-invasive and high temporal resolution tool for brain activity analysis; however, the existing EEG-based diagnostic frameworks cannot account for the complex spatio-temporal relationship and interpretability for clinical decision making. This paper presents an Internet of Medical Things-enabled explainable graph learning approach, namely Neuro Graph Former-X, for automatic diagnosis of Schizophrenia within smart health care environments. In the proposed solution, wearable EEG sensors are used for real-time measurement of brain electrical activities, and communication protocols such as 5G and LoRaWAN are utilized for monitoring purposes. Pre-processing is carried out via bandpass filtering, Z-score normalization, and temporal segmentation to eliminate noise and decrease bandwidth. Next, Discrete Wavelet Transform is implemented to extract multi-resolution features, and Phase Locking Value is used to build dynamic functional brain connectivity graphs. Then, these graphs are fed into the hybrid deep learning model comprising Graph Attention Networks to learn the spatial features, Temporal Convolution Networks for temporal modeling, and Transformer encoders to extract global dependencies. In addition to this, the explainable AI module, which uses neurocognitive attention and Integrated Gradients, is introduced for improving the explainability of the model. As per experimental results, it can be stated that the proposed model attains 96.84% Precision, 96.58% F1-Score, and 98.21% AUC, thereby performing better than existing machine learning and deep learning approaches. It has been proven from the results that the proposed approach is highly scalable and interpretable for schizophrenia diagnosis using IoT-enabled smart healthcare platforms.