Predicting Alzheimer's Disease Progression from Sparse Multimodal Data by NeuralODE Models.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Data Science

Background:

• Alzheimer's disease (AD) progression varies significantly among patients, complicating diagnosis and care.
• Current data-driven models often require extensive, specific datasets not readily available in clinical settings.
• Accurate prediction of individual disease trajectories is crucial for effective management of neurodegenerative disorders.

Purpose of the Study:

• To develop a novel modeling framework for predicting individual Alzheimer's disease trajectories.
• To utilize sparse, irregularly sampled, multi-modal clinical data for disease progression modeling.
• To enhance early diagnosis and monitoring of neurodegenerative diseases.

Main Methods:

• Implementation of (recurrent) Neural Ordinary Differential Equations (NODEs).
• Forecasting patient disease progression and biomarker evolution over time.
• Utilizing sparse, multi-modal clinical data for model training and validation.

Main Results:

• The developed model accurately detected early signs of Alzheimer's disease.
• It effectively tracked changes in biomarker trajectories, aligning with clinical knowledge.
• Demonstrated superior performance compared to common data-driven alternatives on the ADNI cohort.

Conclusions:

• The proposed modeling framework offers a versatile tool for personalized Alzheimer's disease diagnosis and monitoring.
• This approach addresses the limitations of existing models in handling sparse, real-world clinical data.
• The findings support the potential of AI in advancing the management of neurodegenerative diseases.