Comparison of Imaging-Derived Features and Multimodal Models for Prognosis Prediction in Motor Neuron Disease

Motivation: Motor neuron disease (MND) prognosis is critical for patient care planning. Current methods rely on clinical metrics, but combining structural MRI with clinical data may enhance predictive performance. Goal(s): This study aimed to evaluate whether integrating structural MRI features with clinical data improves survival predictions in MND and to identify the most useful MRI features. Approach: Six multimodal data fusion models were trained on clinical and imaging-derived features from 220 MND patients. Results: Combining clinical data with large tissue volumes produced the greatest improvement over clinical-only predictions, highlighting the potential of this underused data type in MND prognosis. Impact: This study demonstrates that integrating routinely collected MRI, typically not used for prognosis, with clinical data can enhance prognostic predictions in motor neuron disease without additional patient data collection, as found through a comprehensive evaluation of models and imaging features.