Predicting Alzheimer’s Disease with Deep Learning Techniques

Abstract

Alzheimer's disease (AD) represents the most common type of dementia, advancing from mild to severe stages and ultimately necessitating full-scale care. The increasing incidence of the disease is attributed to aging and late-stage diagnosis. Conventional diagnostic techniques, such as medical history assessments, cognitive assessments, and MRI scans, are marred by inconsistencies and limitations. This research investigates the utilization of a convolutional neural network (CNN) to identify MRI abnormalities related to AD, taking into account four stages of dementia using visually interpretable models in addition to addressing class imbalance. The DenseNet264 model is employed to classify MRI scans, ranging from "not demented" to "moderately insane." MRI data from Kaggle reveals a significant class imbalance, thereby emphasizing the need for a good classification model. The model's ability to distinguish between dementia stages is assessed by predictions derived from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.