Predicting multiple sclerosis severity with multimodal deep neural networks.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Informatics

Background:

• Multiple Sclerosis (MS) is a chronic neurological disease affecting the brain and spinal cord, leading to nerve damage.
• Accurate MS severity classification is vital for timely therapeutic interventions to prevent disease progression.
• Existing single-modal machine learning approaches have limitations in predicting MS severity due to data constraints.

Purpose of the Study:

• To develop a multimodal deep learning framework for predicting future Multiple Sclerosis (MS) disease severity.
• To integrate diverse patient data, including structured EHR, neuroimaging, and clinical notes, for enhanced prediction.
• To assess the predictive value of individual data modalities in MS progression.

Main Methods:

• Proposed a novel multimodal deep learning framework integrating structured EHR, neuroimaging, and clinical notes.
• Utilized longitudinal patient data for predicting future MS disease severity.
• Evaluated model performance using the Area Under the Receiver Operating Characteristic curve (AUROC).

Main Results:

• The multimodal deep learning framework achieved up to a 19% increase in AUROC compared to single-modal models.
• Demonstrated the effectiveness of integrating diverse data sources for improved MS severity prediction.
• Identified the utility of different data modalities for predicting MS disease progression.

Conclusions:

• Multimodal data integration with deep learning offers a significant advancement in predicting MS disease severity.
• This approach enhances prediction accuracy, potentially enabling earlier and more effective treatment strategies.
• Insights gained can optimize future data collection for MS research and clinical practice.