Early Detection of Alzheimer’s Disease Using Deep Learning

Given that Alzheimer's is among the most prevalent neurodegenerative illnesses, early diagnosis is crucial for efficient treatment planning and patient care. The goal was to create an automated system that uses deep learning techniques which uses MRI brain images to detect Alzheimer's illness brain scans. The suggested method uses a Vision Transformer-based model for more accurately diagnose various illness stages.    The choice of ViT as the core architecture draws from Dosovitskiy et al. [6], who showed that transformers applied directly to image patches can match or outperform CNNs when pretrained on sufficient data. This is especially valuable in medical imaging, where understanding global brain structure matters more than local texture patterns. As Mauricio et al. [8] further support, ViT's multi-head self-attention mechanism contributes to stronger classification accuracy, making it well-suited for the spatial complexity of MRI-based Alzheimer's staging.    The proposed system incorporates typical preprocessing steps, enhances data through augmentation, and involves fine-tuning of a pretrained ViT model, and an early stopping mechanism to retain the best-performing weights. A Flask-based web application was also built on top of the trained model, allowing MRI images to be uploaded and classified with confidence scores — making it practically usable in a clinical environment.   Despite promising results, the system has a few notable limitations. First, the model was trained and evaluated on a Kaggle-sourced dataset, which may not fully represent the variability seen in real-world clinical MRI data. Differences in scanner hardware, acquisition protocols, and patient demographics could influence overall generalization effectiveness. Second, the current system processes only 2D MRI slices and does not exploit the volumetric information present in full 3D brain scans, which could carry additional diagnostic value. Third, ViT models generally demand more data to train effectively compared to CNNs, as noted by Dosovitskiy et al. [6], and performance on smaller datasets may be more dependent on the quality of pretraining. Finally, the system currently lacks explainability features — clinicians are not shown which regions of the MRI influenced the prediction, which is an important requirement for real-world medical deployment.  

[1]  A. H. J. Albaidhani, G. Serbes and H. O. Ilhan, "Early Detection of Alzheimer’s Disease Using Vision Transformers on CWT-Transformed EEG Signals in Response to Olfactory Stimuli," in IEEE Access, vol. 14,   pp.   6968-6985,   2026,   doi: 10.1109/ACCESS.2026.3652249. [2]   M. M. S. Fareed et al., "ADD-Net: An Effective Deep Learning Model for Early Detection of Alzheimer Disease in MRI Scans," in IEEE Access, vol. 10, pp. 96930-96951,    2022,    doi: 10.1109/ACCESS.2022.3204395. [3]   G. N. Alwakid, S. Tahir, M. Humayun and W. Gouda, "Improving Alzheimer’s Detection With Deep Learning and Image Processing Techniques," in IEEE Access, vol. 12, pp. 153445-153456, 2024, doi: 10.1109/ACCESS.2024.3481238. [4]  H. V. Bansal, P. Gupta and V. Juneja, "A Multimodal Deep Learning Framework Using ResNet-101 and Firefly-Based Feature Selection for Early Diagnosis of Dementia and Alzheimer’s Disease," in IEEE Access, vol. 13, pp. 184709-184721, 2025, doi: 10.1109/ACCESS.2025.3621157. [5]   A. Baktash, Y. Sarbaz, S. Meshgini and R. Afrouzian, "Advancing Early Diagnosis: Predicting Mild Cognitive Impairment Progression in Normal Individuals Using Deep Learning on MRI Features," in IEEE Access, vol. 13, pp. 122591-122602, 2025, doi: 10.1109/ACCESS.2025.3588801. [6]  A. Dosovitskiy et al., "An Image is Worth 16×16 Words: Transformers for Image Recognition at Scale," in International Conference on Learning Representations (ICLR),   2021,   doi: 10.48550/arXiv.2010.11929. [7]   A. Baktash, Y. Sarbaz, S. Meshgini and R. Afrouzian, "Advancing Early Diagnosis: Predicting Mild Cognitive Impairment Progression in Normal Individuals Using Deep Learning on MRI Features," in IEEE Access, vol. 13, pp. 122591-122602, 2025, doi: 10.1109/ACCESS.2025.3588801. [8]      J. Maurício, I. Domingues, and J. Bernardino,   "Comparing    Vision Transformers and Convolutional Neural Networks for Image Classification: A Literature Review,"Applied Sciences, vol. 13,  no.  9,  p.  5521,  2023,  doi: 10.3390/app13095521. [9]  A. Varma et al., "Vision Transformers in Image Classification: A Comprehensive Review of Architectures and Applications," arXiv preprint arXiv:2302.09462, 2023, doi: 10.48550/arXiv.2302.09462. [10]  J. Chen, Y. Lu, Q. Yu, X. Luo, E. Adeli, Y. Wang, L. Lu, A. L. Yuille, and Y. Zhou, "TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation," arXiv preprint arXiv:2102.04306, 2021.