FGI-CogViT: Fuzzy Granule-based Interpretable Cognitive Vision Transformer for Early Detection of Alzheimer’s Disease using MRI Scan Images

Early detection of Alzheimer’s disease (AD) is crucial for timely intervention and management of this debilitating neurodegenerative disorder. However, it demands further serious attention. State-of-the-art vision transformers for multi-class AD detection techniques cannot handle the uncertainty issue arising between various stages of AD. Moreover, AD identification based on magnetic resonance imaging (MRI) scans is likewise computationally expensive. Further, vision transformers used in AD detection often suffer from a lack of interpretability of results. To address these issues, a new vision transformer, namely Fuzzy Granule-based Interpretable Cognitive Vision Transformer (FGI-CogViT) is developed. It has three parts, namely feature extraction, fuzzy logic-based granulation, and I-CogViT-based classification. Various vision and statistical features are computed over the MRI scan image(s). The statistical features are used to obtain the disease-prone regions in terms of fuzzy granules. In these regions, uncertainty may arise among the different stages of AD. Fuzzy logic-based rules are defined to obtain the crisp granules. Instead of considering the entire image, statistical features corresponding to the crisp granules are added with vision features for classification tasks through the I-CogViT that consists of three modules, namely residual network, traditional vision transformer, and classification network. These characteristics improve the speed and accuracy of FGI-CogViT. It synergizes the robust feature extraction capabilities of vision transformers with cognitive computing principles, aiming to augment the model’s interpretability. The efficacy of the FGI-CogViT has been demonstrated over 6,460 MRI scan images. Results reveal that FGI-CogViT outperforms some state-of-the-art. Furthermore, robustness checking and statistical significance testing support the findings.