A Novel Machine Learning Model for Alzheimer’s Disease (AD) Medical Image Analysis

This study introduces a machine learning model to classify individuals as Alzheimer’s Disease (AD) patients, Mild Cognitive Impairment (MCI) sufferers, or Cognitively Normal (CN) volunteers, using axial T2*-weighted MRI scans from the ADNI database. Our approach encompasses two primary phases: feature extraction and classification prediction. For feature extraction, we employ the Residual Network, converting MRI images into numerical vectors that represent key observable features. In the classification phase, to manage the high dimensionality of the data, we utilize a neural network incorporating a dimensionality reduction layer and a three-layer neural network for enhanced learning of class characteristics. The process concludes with a softmax layer for distinguishing between AD, MCI, and CN states. Our model demonstrates significant promise, achieving an average accuracy of $\mathbf{8 3. 2 4 \%}$ in cross-validation tests. This result notably shows advantages compared with other models, especially in larger datasets. We creatively used MCI as part of the label. Thus, our three-label classification system offers more nuanced insights compared to traditional binary models This is because, in clinical scenarios, doctors care more about MCI patients since they are easier to treat compared with AD patients who already get serious Alzheimer’s symptoms. In all, Our findings suggest that this model could be an effective tool for early detection and staging of Alzheimer’s Disease, offering a comprehensive and nuanced understanding of the disease progression.