Multimodal Ordinal Modeling of Alzheimer's Disease Severity Using Structural MRI and Clinical Data

A novel attention-enhanced multimodal machine learning framework for assessing Alzheimer's disease (AD) severity has been introduced, integrating T1-weighted MRI with demographic and genetic data. The model employs ordinal regression and demonstrates superior performance, achieving an adjacent-stage accuracy of 0.970 and a quadratic weighted kappa (QWK) of 0.549, indicating better alignment with clinical staging compared to unimodal approaches. This framework provides a scalable and interpretable solution for automated AD severity staging, enhancing clinical decision support through improved prediction accuracy and transparency via explainability techniques like Grad CAM++ and SHAP.