Causal Explanation from Mild Cognitive Impairment Progression using Graph Neural Networks

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Mild Cognitive Impairment (MCI) represents a critical transitional phase between normal cognitive aging and dementia.
• Understanding the factors driving MCI progression is crucial for early intervention and treatment strategies.
• Existing research on MCI progression lacks comprehensive analysis of longitudinal data integrating genotypes, biomarkers, and chronic diseases using explainable AI.

Purpose of the Study:

• To develop and validate a novel explainable artificial intelligence approach for predicting MCI transitions using longitudinal data.
• To identify key clinical and genetic factors influencing the progression or reversion of MCI.
• To enhance the interpretability and reliability of AI models in cognitive health research.

Main Methods:

• Construction of a comprehensive graph representation for each individual using longitudinal temporal data.
• Application of a temporal graph convolutional network for predicting MCI transitions.
• Utilizing a causal explanation method to identify significant predictive variables and assess explanation quality.

Main Results:

• The temporal graph convolutional network achieved 72.4% accuracy in predicting MCI transitions.
• The causal explanation method demonstrated superior stability, accuracy, and faithfulness compared to existing techniques.
• Identification of a causal subgraph including hypertension, arrhythmia, congestive heart failure, coronary artery disease, stroke, lipid-related issues, and sex as informative variables.

Conclusions:

• Explainable graph neural networks offer a powerful approach for understanding MCI progression dynamics.
• The identified causal subgraph highlights the significant role of cardiovascular and lipid-related factors in MCI transitions.
• This study provides a foundation for developing more accurate and interpretable predictive models for cognitive decline.