Transfer Learning approach for Alzheimer’s Disease Stage Classification by integrating 3D CNN with Multi-modal inputs

This research concentrates on approaches to analysing the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. It utilises transfer learning across multiple 3D convolutional neural networks, drawing on pre-trained weights from ImageNet. The primary focus lies in integrating 3D neuroimaging data alongside subject-specific data such as gender, age, and visit month, all available in the ADNI dataset. The outcomes of these models are compared with single modal models and complex models, such as vision transformer model trained on 2D images generated by selecting the middle slice of the 3D images. Our multimodal 3D models performed remarkably well, with ResNet-18 being the most effective of the four models with an accuracy of 98.75%. SResNeXt-101 showed greater accuracy with multimodal (95%) over single-modal (67.26%). Our suggested multimodal architecture significantly improved the detection results compared to previous research. This study enhances early AD detection by using deep learning and multimodal data for diagnosing complex neurological disorders.