A Novel MRI Image Enhancement Pipeline for Alzheimer's Staging via Grad-Cam-Integrated CNN and KNN Based Patch-Level Attention

The use of various image processing techniques and machine learning algorithms to assist in the detection and treatment of Alzheimer's has led to marginally significant improvements in the accuracy of diagnosis and treatment of the disease. This paper proposes a novel pipeline that attempts to utilize the accuracy of Convolutional Neural Networks (CNNs) with the efficiency of K-Nearest Neighbors (KNN) models, in an attempt to develop a machine-learning-guided image processing technique tailored for MRI scans of Alzheimer's disease. Achieving an accuracy of 93.5% for the CNN model and an 89% testing accuracy for the KNN model, the resulting images highlight the more important areas of the brain most associated with the disease, potentially leading to improvements in the detection and treatment of Alzheimer's disease.