Developing a Modified Deep Belief Network with metaheuristic optimization Algorithm for predicting Alzheimer disease using Electroencephalogram

Abstract

A neurological brain disorder that progresses over time is Alzheimer's disease. Alzheimer's disease can take years to identify, comprehend, and manifest—even in cases where signs are obvious. On the other hand, technological developments like imaging methods aid in early detection. But frequently, the results are unreliable, which delays the course of treatment. By dividing resting-state electroencephalography (EEG) signals into three groups—AD, healthy controls, and mild cognitive impairment (MCI)—this work offers a novel perspective on the diagnosis of Alzheimer's disease (AD). In order to overcome data limits and the over-fitting issue with deep learning models, we looked at augmenting the one-dimensional EEG data of 100 patients (49 AD participants, 37 MCI subjects, and 14 HC subjects) with overlapping sliding windows. Better results and early intervention could arise from this for persons afflicted with the illness. This research has the potential to significantly advance the early diagnosis of Alzheimer's disease and lay the groundwork for the creation of more precise and trustworthy diagnostic instruments for this debilitating condition. This study presents a Modified Deep Belief Network (MDBN) with a metaheuristic optimization method for detecting face expression and Alzheimer's disease using EEG inputs. The recommended method extracts significant features from EEG data in a novel way by applying the Improved Binary Salp Swarm Algorithm (IBSSA), which combines the MDBN and the metaheuristic optimization algorithm. The performance of the suggested technique MDBN-IBSSA for Alzheimer's disease diagnosis is evaluated using two publicly available datasets. The proposed technique's capacity to discriminate between healthy and ill patients is proved by the MDBN-IBSSA accuracy of 98.13%, f-Score of 96.23%, sensitivity of 95.89%, precision of 95.671%, and specificity of 97.13%. The experimental results of this study show that the MDBN-IBSSA algorithm proposed for AD diagnosis is effective, superior, and applicable.