Building a Human Digital Twin (HDTwin) Using Large Language Models for Cognitive Diagnosis: Algorithm Development and...

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Digital Health

Background:

• Personalized cognitive health diagnosis faces challenges in integrating multifaceted data sources into a unified model.
• Human digital twins offer a promising approach to consolidate diverse health information for a holistic view.

Purpose of the Study:

• Introduce HDTwin, a novel method utilizing large language models to unify heterogeneous data for cognitive health.
• Develop a system capable of predicting cognitive diagnoses and providing interpretable explanations for its inferences.

Main Methods:

• HDTwin integrates diverse cognitive health data (demographic, behavioral, n-back, speech) by converting them into text prompts for large language models.
• The system combines patient data with external scientific literature to construct a predictive model.
• Model performance was validated against baseline machine learning classifiers using data from 124 participants across three studies.

Main Results:

• HDTwin achieved a peak accuracy of 0.81, significantly outperforming baseline classifiers (average accuracy 0.65).
• The system demonstrated superior predictive accuracy when fusing multiple information sources compared to single sources.
• HDTwin's chatbot interface facilitates interactive diagnosis interpretation and data exploration.

Conclusions:

• HDTwin effectively integrates diverse cognitive health data, enhancing diagnostic accuracy and explainability.
• This approach surpasses traditional models, offering a promising tool for early detection and intervention in cognitive health.
• The interactive interface aids clinicians in navigating complex patient information for improved decision-making.