Alzheimer's Disease MRI Classification using EfficientNet: A Deep Learning Model

The most common form of dementia, Alzheimer's disease (AD) ranks sixth in terms of fatality for people 65 years of age and older. Additionally, according to official statistics, the number of deaths caused by AD has increased substantially. Thus, early AD diagnosis can improve the prognosis for patients. Magnetic resonance images (MRI) have been often used for the diagnosis of AD. This research aims to enhance the accuracy of AD recognition through the development of an innovative system. Initially, the brain images were acquired from the AD Neuroimaging Initiative (ADNI) and the Open Access Series of Imaging Studies (OASIS), both of which are virtual datasets. The processed images are subsequently passed through the region-growing method, which also reduces the complexity of the system and extracts a connected region of an image based on predefined criteria. The segmentation procedure utilizes U-Net to partition the brain tissues. Finally, The EfficientNet-B0 deep learning model was utilized for classification, feature extraction and selection. The training set comprises 75% of the dataset, while the testing set comprises 25%. Specification: 98.120 % accuracy: 98.12 % sensitivity: 97.48%, precision: 98.40 % and f1-score: 97.89 % were the operational metrics of the U-Net + EfficientNet-B0 (UNet+ EffNet B0) model.