Dynamic Bayesian spectrum optimization and LSTM integration for accurate EEG-based dementia detection

Dementia is a neurodegenerative disorder marked by cognitive decline affecting daily activities, including various conditions, however, early detection and classification of dementia using EEG signals is crucial. Hence, a novel "Dynamic Bayesian Morlet-Hilbert Consensus Spectrum Optimization" is proposed to enhance the accuracy of dementia diagnosis, specifically for Alzheimer's Disease and Mild Cognitive Impairment. Additionally, Cholinergic System Dysfunction, characterized by non-uniform brain degeneration, complicates EEG feature extraction due to heterogeneous neural activity patterns, and existing models struggle to accurately interpret these patterns. Thus, a novel Adaptive Morlet Consensus Beamforming Transform is introduced to capture heterogeneous neural activity patterns and adapt to complex brain regions, thereby improving the identification of biomarkers associated with cholinergic system dysfunction. Furthermore, existing models often overlook the striatum due to its intricate structure and the cholinergic system's impact on cognitive functions, making it challenging to accurately assess cholinergic dysfunction. So, a novel Multilayer Adaptive Dynamic Causal Frequency Decomposition is introduced to detect subtle variations in brain function, improving the identification of early-stage dementia-related changes linked to cholinergic system dysfunction. Moreover, the striatum, a complex structure with subregions, exhibits varying impacts of cholinergic dysfunction, causing significant variability in cognitive and motor functions. Therefore, a novel Bayesian Hilbert-ElasticNet Spectrum Optimization is introduced to effectively capture the spectral and temporal dynamics of EEG signals to classify subtle variations in dementia. The results demonstrate that the methodology significantly improves classification performance, achieving an accuracy of 98.23%, along with precision and F1-Score rates of 98.84% and 97.52%, respectively, while minimizing RMSE and enhancing processing efficiency.