Integrating Traditional Machine and Deep Learning Methods for Enhanced Alzheimer's Detection from MRI Images

Alzheimer's Disease (AD) is a prominent progressive neurodegenerative disorder that causes impairments in cognition and physically affects the brain. As there is no cure for AD, early detection is pertinent to prevention and slowed progression of the disease. Current diagnostic methods involve the manual evaluation of MRI images by trained professionals. This is useful yet has limitations that could be overcome by utilization of machine learning classification models for early detection of AD, which is the aim of this research. To do this, an OASIS-3 dataset containing images of brain MRIs labeled as mild cognitive impairment and cognitively normal was split into testing, training, and validation data. The data was augmented, featurized, and trained on various algorithms. Featurization was done using the deep learning convolutional neural network ConvNextXLarge, which was pre-trained on ImageNet. The algorithms were tested on validation data and the best model was selected. MLP, KNN, and RF were models that had an accuracy of 0.979 and XGBoost had an accuracy of 0.959. MLP was selected as the final model and performed with a final accuracy of 0.953 on the testing data with a recall value of 1. The results of this study demonstrate that machine learning models can be used to aid in diagnosis of Alzheimer's disease, allowing for improved health conditions and treatment of AD.