Hybrid Deep Learning Architecture with Adaptive Feature Fusion for Multi-Stage Alzheimer’s Disease Classification

Abstract

This work proposes a hybrid deep learning framework combining convolutional neural networks and transformer-based architectures for image classification. This method uses the feature extraction functionalities of ResNet and Vision Transformer (ViT) models for better classification performance. The features are fused using a feature fusion mechanism using extracts of the two models, used to train a classification model on the merged space. The method allows to leverage the strengths of CNNs for spatial features and transformers for global dependencies, thus resulting in improved prediction. The proposed system consists of preprocessing, feature extraction, feature fusion and model training and prediction modules. In the experiments, we showed that the hybrid model was more generalize and robust than the individual models. The system is constructed to facilitate real-time prediction and it is designed to be easy to use, which makes it applicable in medical diagnosis, image analysis, and decision support system.