Robust Multimodal MRI Biomarkers for Early Cognitive Decline Under Missing Data Conditions: A Pilot Machine Learning Study

This study presents a pilot framework for integrating structural (T1-weighted) and susceptibility-sensitive (T2*) MRI with behavioural features for early detection of cognitive decline. A multimodal machine learning pipeline was developed using ensemble models (Random Forest and Gradient Boosting) to evaluate classification performance and robustness under missing-data conditions.

Results demonstrate improved separability using multimodal features, with T2*-derived biomarkers contributing significantly to classification performance. Although limited by a small sample size (n = 4), this work establishes a reproducible pipeline and highlights the importance of multimodal integration and robustness for clinically relevant AI systems.

The study provides a foundation for future large-scale validation using datasets such as ADNI and UK Biobank.