Classification of Alzheimer's disease with EfficientNet B3

Alzheimer's disease (AD) is a condition that manifests as a loss of consciousness and cognitive dysfunction, eventually leaving the individual incapable of performing basic functions. The process culminates in death. The brain anomalies caused by the disease can be monitored using magnetic resonance imaging (MRI). This study aims to facilitate the clinical diagnosis of AD and proposes a hybrid model to classify the stages of the disease. The magnetic resonance images used in the study were obtained from the Kaggle database and include the classes non-demented, very mild dementia, mild dementia, and moderate dementia. Background removal was applied to the images, which were then segmented using the k-means clustering method. By combining EfficientNet B3 and the Gray Level Co-Occurrence Matrix (GLCM) feature extractor, this hybrid model was trained to perform the classification task. The model was trained five times, and experimental results were recorded. In training, the batch size was set to 18, the number of epochs was 20, and the learning rate was set to 0.0001. Experimental results showed an average training accuracy of 99.99% and a testing accuracy of 99.67%. Additional performance metrics, such as precision, recall, and F1-score, are also reported.