Meta-Heuristic Optimized Dual-Branch CNN- Transformer with Attention-Guided Multimodal Fusion for Alzheimer’s Diagnosis

Alzheimer's disease (AD) is a progressive neurodegenerative disorder, characterised by progressive cognitive impairment, memory and behaviour defect. Early and accurate diagnosis is still a major challenge due to the overlapping symptoms between Mild Cognitive Impairment (MCI) and normal aging and the complex and heterogeneous nature of the disease in the different imaging possibilities. To surmount this issue, in this paper, Meta-Heuristic Optimization-Driven Dual-Branch CNN- Transformer Framework with Attention-Guided Cross-Modal Fusion for automatic AD diagnosis using multi-modal neuroimaging data (MRI and FDG-PET) is proposed. The proposed framework consists of using Convolutional Neural Networks (CNN) to extract the local features of the structure and transformer encoders to extract the global dependencies in the context. An attention-guided cross modal fusion module is used to dynamically highlight complementary disease relevant features from both modalities while a hybrid Ant Lion Optimization-Whale Optimization Algorithm (ALO-WOA) is used to optimize hyperparameters and attention weights with a view to provide better generalization and efficiency. Experiments run on benchmark data sets such as ADNI show an improved performance of 95.0% accuracy, 93.0% sensitivity and 94.0% specificity compared to state-of-the-art multi-modal AD classification models. The results confirm that fusion based on meta-heuristic optimization is able to greatly improve the discriminative feature learning and diagnostic reliability. In conclusion, the proposed framework allows an interpretable and robust approach for early Alzheimer's diagnosis with an enormous potential for use in clinical decision support and neuroimaging analytics in the real world.