Alzheimer Disease Classification using Deep CNN Methods Based on Transfer Learning and Data Augmentation
Abstract
Alzheimer’s disease (AD) primarily impacts a distinct demographic, specifically individuals who are 60 years old and older, predominantly affecting the elderly. AD is an emerging neurological disorder characterized by profound disruption of memory and the onset of behavioral abnormalities, significantly complicating an individual’s life. Recent progress in research has allowed precise diagnosis by leveraging intelligent technologies, including the use of deep learning and convolutional neural algorithms for tasks such as image quality improvement and magnetic resonance imaging. This research utilized convolutional neural network (CNN) models to automatically extract pertinent features associated with Alzheimer’s disease and classify brain MRI images. In contrast to conventional approaches, CNN models exhibit enhanced proficiency in discerning among the four phases of Alzheimer’s disease, which encompass mild dementia, very mild dementia, non-dementia, and moderate dementia. This entails utilizing a pre-trained model that has already captured valuable features from one dataset and adjusting it to perform efficiently on a related yet different task. This study delves into the potential of transfer learning to enhance AD classification. Specifically, we employed three state-of-the-art architectures, ResNet-152, VGG16, and Inception-V3, to discern intricate patterns from brain images, after that we applied data augmentation technique that enlarges the size of a training dataset by implementing various modeling and analytical methods. Transfer learning allows us to adapt and refine these pre-trained networks for our Alzheimer’s disease classification objective, even in situations where there is a scarcity of labeled data.