Predicting Alzheimer's Disease Progression Using Multimodal Longitudinal Analysis: A Machine Learning Approach

Abstract

Patients with Mild Cognitive Impairment (MCI) face an increased risk of developing Alzheimer's disease (AD), highlighting the importance of early diagnosis for effective interventions and management of the disease. In our study, we investigated the progression of AD in patients initially diagnosed with MCI using multimodal longitudinal data. A classification based framework was proposed for MCI prediction with baseline data of 569 stable MCI (sMCI) and 268 progressive MCI (pMCI) patients. Employing three supervised machine learning (ML) algorithms—support vector machine (SVM), logistic regression (LR), Random Forest (RF) and incorporating features such as cognitive function assessments, MRI, PET scans, CSF biomarkers, and genetic APOE status, the classification accuracies of 83.4%, 80.2%, and 80% were achieved respectively. Significant differences were observed in the performance of the models, with the SVM notably outperforming both LR and RF (p < 0.05). Impaired memory function and lower clinical tests scores were found as primary indicators of MCI patients progressing towards AD. Although the fusion of all modalities yielded accurate results for predicting MCI progression to AD, our analysis revealed less significant differences in evaluation metrics when only cognitive test results were used as features. This suggests that cognitive assessments alone are nearly as effective in predicting MCI progression, which can lead to more cost-effective strategies in clinical settings. This study underscores the need for further research aimed at developing new tools to assist clinicians in prognostic decision making.