Predicting Dementia Progression: An Evaluation of Tree-based Models with Clinical and Neuroimaging Data

Dementia is a complex neurological condition characterized by a decline in cognitive function that impairs daily life activities. Early detection and accurate prediction of dementia progression are crucial for effective intervention and management. In this study, we evaluate the performance of tree-based machine learning models using clinical and neuroimaging data to predict dementia progression. Our dataset comprises longitudinal data from 150 subjects aged 60 to 96, including multiple visits and MRI scans. We employ tree-based models, specifically Decision Tree, Random Forest, and Gradient Boosting Machine (GBM), to analyze the relationship between various predictors such as Mini-Mental State Examination (MMSE) scores, Clinical Dementia Rating (CDR), age, educational level, socioeconomic status (SES), and neuroimaging features. The results demonstrate that the GBM model outperforms the other treebased models, achieving an AUC of 0.90 and an accuracy of 0.92 in predicting dementia progression. The Random Forest model follows with an AUC of 0.87 and an accuracy of 0.88, while the Decision Tree model achieves an AUC of 0.819 and an accuracy of 0.82. We also observe a strong correlation between MMSE scores and CDR, highlighting the significance of cognitive function in dementia assessment.