Longitudinal Analysis for Disease Progression via Simultaneous Multi-Relational Temporal-Fused Learning.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Longitudinal prediction of Alzheimer's disease (AD) progression and mild cognitive impairment (MCI) to AD conversion is vital for early diagnosis and intervention.
• Current prediction methods often use separate models for different clinical scores or time points, limiting coordinated learning.
• This fragmentation hinders the development of comprehensive predictive models for AD patient management.

Purpose of the Study:

• To develop a novel joint learning method for predicting multiple clinical scores in Alzheimer's disease (AD) patients.
• To enable coordinated learning across multiple longitudinal prediction models for various future time points.
• To improve the accuracy of predicting AD progression and MCI to AD conversion.

Main Methods:

• A joint learning framework was proposed to predict multiple clinical scores simultaneously for different future time points.
• The method explores relationships among training samples, features, and clinical scores.
• A common feature set was utilized to capture interdependencies between longitudinal prediction models at different time points.

Main Results:

• The proposed joint learning method demonstrated significant improvements over existing methods in predicting multiple clinical scores.
• Experimental results were validated using data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
• The approach effectively captures complex relationships for enhanced predictive accuracy.

Conclusions:

• The developed joint learning method offers a more effective approach for longitudinal prediction of Alzheimer's disease (AD) progression.
• This method facilitates coordinated prediction of multiple clinical scores, outperforming traditional separate modeling techniques.
• The findings have implications for improving early diagnosis and patient management strategies in AD.