Alzheimer's Disease Prediction using Machine Learning and Transfer Learning Models

Abstract

The brain, a marvelous three-pound organ, governs many physical activities, including memory, creativity, emotion, and intellect. It also analyses input from the outside world. The brain shows structural and functional changes because of Alzheimer's Disease (AD). AD often starts off slowly and gets worse over time. It is the underlying factor in 60-70% of dementia cases. Memory loss, paranoia, and delusions in the short term thoughts are symptoms of AD, a degenerative neurological disorder that is often misdiagnosed as stress or ageing-related symptoms. Because AD is chronic, it might last for a huge time or for the rest of life. Early AD therapy is more efficient and results in minor harm in comparison with medication administered later in the disease's progression. Alzheimer's diagnosis can be done using different scans such as PET, MRI or CT. Traditional medical testing takes a long time, and cannot predict early warning signs. Machine learning models can be used for early detection and to achieve efficient classification accuracy. In this paper, for efficient detection of Alzheimer's disease, the most effective parameters are considered using 2 different datasets: Longitudinal dataset which contains text values, and dataset OASIS which has MRI images. OASIS Longitudinal dataset runs on 14 Machine Learning algorithms including Random Forest Algorithm giving a maximum accuracy of 92.1385% and KNN Algorithm with a baseline accuracy of 47.1910%. MRI images are run on different Transfer learning models with the best accuracy got from the InceptionV3 model and ADAM as Optimizer.