A computational framework for Alzheimer's disease detection using SwinRes Transformer

Alzheimer's disease is a progressive neurodegenerative disorder that severely impacts cognitive functions. Early and accurate detection of Alzheimer's disease is significant to improve patient outcomes. However, the conventional model is difficult to capture the structural abnormalities and long-range inter-regional dependencies from the medical images, while maintaining computational efficiency. To address these challenges, this research implements a novel SwinRes Transformer model to detect Alzheimer's disease accurately. The Dilated InceptionV3 framework is employed to extract the features from the image that utilized a Fused MBConv and Group Convolution to mitigate the training time and parameters. The SwinRes Transformer is designed to detect Alzheimer's disease, which is designed by integrating a Modified Residual Network, Convolutional Shifted-Window Spatial-Channel Swin Transformer, and Feature Fusion Module. The Convolutional Shifted-Window Spatial-Channel Swin Transformer is employed to capture the global information and enhance the model's computational efficiency. The Modified Residual Network is employed to extract the local features and contextual information, solve the degradation trouble, and reduce the training time. Moreover, the Feature Fusion Module is designed to refine the features from the Modified Residual Network and Convolutional Shifted-Window Spatial-Channel Swin Transformer, thereby improving the relevant information and suppressing redundancy. Experiments are conducted on four Magnetic Resonance Imaging-based datasets, and the proposed SwinRes Transformer model achieves a superior accuracy of 98.93% and a lower execution time of 1.1 s when compared to existing Alzheimer's disease detection methodologies, providing a computationally efficient solution for accurate and scalable Alzheimer's disease detection.