Multi Feature-Based Alzheimer's Disease Detection from MRI Images

Abstract

The early and correct diagnosis of Alzheimer disease (AD) is a major challenge since brain changes occurring at the early stages of AD are mild and overlap with one another. MRI is a non-invasive way to pick up the anatomical distribution of brain degeneration and structural MRI allows us to pick up anatomical patterns of brain degeneration, but what tends to be missing is the complexity of what is happening in AD with the single feature set based models. This paper suggests a multi feature-based deep learning framework combining deep features learned with a proprietary CNN and pretrained ResNet50 and handcrafted descriptors such as GLCM (Shape, Texture, Intensity). MRI images consisted of four categories of AD, namely, MildDemented, ModerateDemented, VeryMildDemented, and NonDemented. The hybrid model outperformed the rest by displaying 99.70% accuracy, 98.21% precision, 95.14% recall, 96.59% F1-score, 95.14% sensitivity, and 98.84% specificity. The outcomes demonstrate the optionality of joining several forms of features to increase the precision and stability of the diagnosis. The method serves an encouraging means of detecting and diagnosing AD as early as possible and helps clinicians make decisions based on AI-enhanced neuroimaging.