Advanced EEG signal processing with deep autoencoders and hybrid Mamba classifier for accurate classification of chronic neuropathic pain etiologies

Abstract

This study introduces a groundbreaking framework that leverages deep autoencoders and a novel hybrid Mamba classifier to enhance the objective classification of chronic neuropathic pain (CNP) etiologies using EEG signals, addressing a critical gap in pain diagnostics. Chronic neuropathic pain is a multifaceted condition characterized by diverse symptoms and etiologies, making accurate diagnosis challenging due to reliance on subjective assessments. The primary aim of this research is to develop a data-driven, scalable solution capable of classifying six distinct CNP categories, including diabetes-related neuropathy, spinal cord injury (SCI), and trigeminal neuralgia, with exceptional precision. Our unique contribution lies in the integration of deep learning for EEG feature extraction and the hybrid Mamba classifier, combining the strengths of traditional and advanced machine learning techniques for unparalleled accuracy. Using a dataset of 36 patients, EEG signals were preprocessed through artifact removal, segmentation, and balancing via the SMOTE algorithm. The model achieved superior performance metrics, including 99% accuracy, precision, recall, F1-score, and Matthews Correlation Coefficient (MCC), along with an AUC of 0.99 across all categories, significantly outperforming traditional models like SVM (AUC 0.97) and logistic regression (AUC 0.83). By identifying distinct EEG patterns linked to different pain types, this approach not only ensures diagnostic reliability but also supports personalized treatment planning. These findings underscore the transformative potential of integrating EEG-based biomarkers with advanced computational techniques, setting a new standard in neurophysiological pain diagnostics. Future work will focus on expanding datasets, incorporating multimodal data, and enabling real-time applications to further enhance clinical impact and generalizability.