Deep Learning-Based Classification of Alzheimer’s Disease from 2D Brain MRI

Early and precise identification of Alzheimer's remains a major concern with considerable consequences for patient care and treatment practices. Initial indications might involve trouble recalling recent happenings, issues with language, and challenges in planning and problem-solving. In this paper, a deep learning-based classification model is proposed which is in turn based on convolutional networks to classify different stages of AD based on 2D axial MRI scans of the brain. To train the model, a publicly available dataset consisting of images corresponding to Normal, Cognitive Impairment, and Alzheimer's disease classes is utilized. To address class imbalance and improve the generalization ability of the model, various data preprocessing techniques are employed. Data augmentation is also conducted including random horizontal flipping and image translation. The custom-optimized CNN architecture outperformed existing models, such as LeNet-5, VGG 16, and AlexNet, achieving 95.58% accuracy and a 95.71% F1 score on the test set. The exponential scheduling played a crucial part in stabilizing the training and reducing the volatility in validation and training losses. To assess model's performance, metrics such as precision, recall, and F1-score are used, which show the model's resilience across all classes and serve beneficial perspectives into its performance.