Cross-Scale ResNet-Vision Transformer for Alzheimer’s Severity Detection using MRI Images
Main Article Content
Abstract
Alzheimer’s disease is a progressive neurodegenerative disorder that poses a significant challenge in neuroimaging. It is primarily characterized by progressive memory impairment and most commonly affects individuals over the age of sixty-five. Currently, no definitive cure exists, and available medications merely alleviate symptoms or slow disease progression. This paper proposed an approach for anatomical feature extraction from a brain important region responsible for memorization and map these anatomical features on global context as patches based on its distribution. The proposed approach structured into two distinct phases for prior non-invasive detection of disease. In the first phase, pre-processing techniques are applied to improve the quality, consistency of MRI class samples, and minimize inter-sample variability, thereby enabling more reliable probability estimation across the four class samples. In addition to this, second phase perform model hybridization as ResNet-Vision Transformer model emerges as a notable approach for improving Alzheimer’s severity detection. It combines the local spatial features extracted by ResNet with the global contextual embedding captured by the Vision Transformer. This fusion enables the model to jointly analyze fine-grained structural details and long-range dependencies within brain MRI image samples. The impact of this hybrid model is significant, as it enables a more detailed representation of neurodegenerative patterns, which could revolutionize the detection of Alzheimer’s disease. As a result, the model learns robust features that improve both performance and generalization across class samples. The experimental results are not only robust but obtained having precise accuracy of 98.10%.