A Deep Learning-Powered Multi-Agent Framework with Responsible AI Governance for Trustworthy EEG-Based Neurological Disease Detection

• Rakeshkumarreddy Ambati

  Author

Deep learning architectures have further accelerated the analysis of neurological disorders through EEG signals at a rate far above the clinical diagnosis. In this paper, we present NeuroMCP-Agent, which is a novel deep learning-based multi-agent framework for XAI-Enabled neuro-disease detection. Our method makes it possible to trust EEG-driven detection of seven neurological conditions, namely, Parkinson PD, epilepsy, schizophrenia, chronic stress, Alzheimer's disease, autism spectrum disorder ASD, and major depression. Moreover, more than a billion people around the globe are affected by these neurological diseases. Thus, making the effort impactful.  We propose a novel deep learning hybrid stacking ensemble, which is produced by using a combination of 15 different deep learning and classical classifiers such as ExtraTrees, Random Forest, Gradient Boosting, XGBoost, LightGBM, AdaBoost, MLP and SVM. Along with 47 EEG feature extractions and 15x Data augmentation. A new RAI governance system, which contains 46 modules and over 1300 types of analyses, is also introduced. It offers comprehensive governance solutions for data life cycles, model internals, deep learning diagnostics, computer vision, NLP analyses, RAG pipelines and AI security. Application of 12-Pillar Trustworthy AI establishes a framework for trust calibration, governance, portability and robustness throughout the various stages of life. The proposed framework is evaluated through LOSO-CV and a robust 5-fold stratified CV through BCIs (1000 iterations). The results are statistically significant (p<0.01, Wilcoxon signed-rank test). NeuroMCP-Agent ushers in a novel era of reliable, high-accuracy medical artificial intelligence and competent deep learning methods of RAI.

• PDF