Biomarker extraction-based Alzheimer's disease stage detection using optimized deep learning approach

Background Cognitive decline and memory loss in Alzheimer's disease (AD) progresses over time. Early diagnosis is crucial for initiating treatment that can slow progression and preserve daily functioning. However, challenges such as overfitting in prediction models, underutilized biomarker features, and noisy imaging data hinder the accuracy of current detection methods. Objective This study proposes a novel deep learning-based framework aimed at improving the identification of AD stages while addressing the limitations of existing diagnostic techniques. Methods Structural MRI scans are employed as the primary diagnostic tool. To enhance image quality, contrast-limited adaptive histogram equalization and wavelet soft thresholding are applied for noise reduction. Biomarker segmentation focuses on ventricular and hippocampal abnormalities, optimized using a firefly algorithm. Dimensionality reduction is performed via Linear Discriminant Analysis to minimize overfitting. Finally, a Deep Belief Network optimized using the Cuckoo Search algorithm is employed for classification and feature learning. Results The proposed framework demonstrates improved performance over existing methods, achieving a 0.66% increase in accuracy and a 0.0345% decrease in error rate for AD stage detection. Conclusions This deep learning strategy shows promise as an effective tool for early and accurate AD stage identification. Enhanced segmentation, dimensionality reduction, and classification contribute to its improved performance, offering a meaningful advancement in AD diagnostics.