A Comparative Analysis on Metaheuristic Algorithms Based Vision Transformer Model for Early Detection of Alzheimer's Disease

TL;DR

This work tackles early Alzheimer's disease detection from brain MRI by integrating a Vision Transformer (ViT) with metaheuristic hyper-parameter optimization. Four optimization strategies—Differential Evolution, Genetic Algorithm, Particle Swarm Optimization, and Ant Colony Optimization—tune ViT hyper-parameters to maximize Sparse Categorical Accuracy across AD, MCI, and HC. On the ADNI dataset (600 scans), the DE-augmented ViT achieves about $96.8\%$ classification accuracy, outperforming other metaheuristic variants and several SOTA methods while maintaining strong precision, recall, and F1-score. The approach demonstrates robustness and scalability for large MRI datasets and suggests extensions to EMCI/LMCI and other neurological conditions.

Abstract

A number of life threatening neuro-degenerative disorders had degraded the quality of life for the older generation in particular. Dementia is one such symptom which may lead to a severe condition called Alzheimer's disease if not detected at an early stage. It has been reported that the progression of such disease from a normal stage is due to the change in several parameters inside the human brain. In this paper, an innovative metaheuristic algorithms based ViT model has been proposed for the identification of dementia at different stage. A sizeable number of test data have been utilized for the validation of the proposed scheme. It has also been demonstrated that our model exhibits superior performance in terms of accuracy, precision, recall as well as F1-score.

Figures (2)

• Figure 1: Design flow of proposed model
• Figure 2: Confusion matrix for (a) ViT model, (b) DE based ViT model, (c) GA based ViT model, (d) PSO based ViT model, (e) ACO based ViT model