Alzheimer's Disease Detection Using an Ensemble of Transfer Learning Models

Alzheimer's Disease (AD) is the most common and quickly developing neurodegenerative condition in the elderly and a primary cause of dementia. AD causes pain, impairs everyday chores, and kills individuals. Although there is no cure, detecting it early can help with treatments that can slow down its progression. TL greatly improves the detection of AD by analysing brain imaging data to identify early biomarkers, enhancing diagnostic accuracy and providing more precise predictions of disease progression compared to traditional methods. This article presents an ensemble methodology for TL models to detect Alzheimer's from brain MRIs, which can be helpful for medical practitioners. The framework starts by preprocessing MRI images and augmenting data. These images are used to train four different TL models. Each model is responsible for classifying the images into one of four stages of AD: ND, VMD, MD, and MoD. By combining these predictions through ensembling, we are able to achieve a final prediction that is highly accurate. This approach also helps to greatly reduce false positive rates when compared to using individual models alone. The ensemble model demonstrates exceptional performance with an overall accuracy of 98.91%, a specificity of 0.9965, and an AUC of 0.9830, highlighting its effectiveness in AD detection. In the future, we will also be exploring different ensembling techniques to improve accuracy and reliability.