Attentive deep learning with Randomized Vector Energy Least Square Twin Support Vector Machine for Alzheimer's Disease diagnosis

Alzheimer's Disease (AD), the most common form of dementia, progressively deteriorates cognitive functions, emphasizing the importance of early and accurate diagnosis for effective treatment and management. This study proposes an advanced framework combining neuroimaging and machine learning to enhance the diagnostic precision of AD. Leveraging T1-weighted structural Magnetic Resonance Imaging (MRI) scans, the model employs a 10-layer Residual Network (ResNet) integrated with a multi-head attention mechanism to extract high-resolution features from sagittal slices, focusing on critical regions such as the hippocampus and amygdala. These features are classified using the Randomized Vector Energy Least Square Twin Support Vector Machine (RV-ELSTSVM), a novel classifier designed to improve generalization by employing randomized feature transformations and energy-based regularization. Tested on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, the proposed framework demonstrates superior performance, achieving classification accuracies of 94.38% for CN vs AD, 88.88% for CN vs MCI, and 92.88% for MCI vs AD. By surpassing existing state-of-the-art methods, this approach highlights the efficacy of combining advanced feature extraction with robust classification techniques for early AD diagnosis. These findings pave the way for impactful clinical applications, offering healthcare professionals a powerful tool for timely intervention and management of AD. The source code of the proposed model is available at https://github.com/rsharma2612/Randomised-SVM.