Developing a Classification Model for Early Stage Alzheimer's Detection Using MRI Images with Deep Learning Methods

Alzheimer's disease, a neurodegenerative disorder caused by the deterioration of brain neurons, results in a decline in daily functional abilities and remains incurable. However, medication can slow its progression, making early detection critical for prevention. This study aims to develop a classification model for early-stage Alzheimer's detection. The dataset utilized is the “Modified Alzheimer's Dataset,” comprising 6,400 axial view MRI brain images categorized into four classes: Non-Demented (3,200 images), Very Mild Demented (2,240 images), Mild Demented (896 images), and Moderate Demented (64 images). The training process employed pretrained models, SqueezeNet and EfficientNet. To address class imbalance, preprocessing was performed using the SMOTE-ENN technique. Post-preprocessing, various hyperparameter configurations were tested on both models. The best configurations for each archi-tecture were then analyzed to compare their performance. The highest accuracies achieved by the optimal models of SqueezeNet and EfficientNet were 98.75% and 99.25%, respectively.