Multi-class Alzheimer’s disease (AD) classification using Swin Transformer wavelet and Gray Wolf Optimization (GWO)

Alzheimer’s disease (AD) is a slow-growing neurological disorder that destroys human thought and consciousness. This disease directly affects the development of mental ability and neurocognitive function. The number of Alzheimer’s patients is increasing day by day, especially in the elderly over 60 years of age, and it gradually becomes a cause of their death. Machine learning (ML) and deep learning (DL) approaches have been developed in the literature to improve the diagnosis and classification of AD. Machine learning approaches have cumbersome feature selection. Deep learning has been used in recent research because it automatically selects features. This research aims to present a Swin Transformer wavelet for Alzheimer’s classification based on structural MRI images in two-class, three-class and four-class modes. The proposed approach uses wavelet fusion in the Swin Transformer network to extract features. The outputs of the modified capsule are fed into a wavelet as feature vectors. The wavelet is a relevant feature selector in the proposed model. The Gray Wolf Optimization (GWO) method was used to find the model’s hyperparameters. The proposed approach achieved an accuracy of 0.9812 in 4-class classification, 0.9980 in 3-class classification, and 1.0 in 2-class classification. In the studies conducted in this research, the Swin Transformer wavelet+GWO model is the heaviest model in terms of the evaluation criteria Parameters(10e6), GFlops, and Memory (GB). This is while the EfficientNet model is the lightest in these criteria. • A hybrid deep learning framework combining Swin Transformer, wavelet transform, and Gray Wolf Optimization is proposed for multi-class Alzheimers disease MRI classification. • Wavelet decomposition enhances multi-scale frequency feature extraction, improving discrimination between dementia stages compared with baseline CNN models. The proposed model achieves high performance across 2-class, 3-class, and 4-class Alzheimers classification tasks, reaching up to 0.98120.9875 accuracy in four-class experiments. • Gray Wolf Optimization effectively tunes model hyperparameters and contributes to consistent performance improvements over transfer-learning baselines. • Grad-CAM visualization is incorporated to enhance interpretability and provide clinically meaningful insight into model decision regions.