QAMT: An LLM-Based Framework for Quality-Assured Medical Time-Series Data Generation

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Data Science

Background:

• Real-world medical time-series data are crucial for research and clinical decisions but face challenges like limited volume, poor quality, and privacy concerns.
• Existing data generation methods, including Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs), struggle with the complexity of medical data, particularly static event data, and often lack interpretability.
• Large language models (LLMs) show promise but face difficulties in generating temporal data and domain-specific nuances.

Purpose of the Study:

• To propose QAMT, the first LLM-based framework for modularly generating quality-assured and interpretable medical time-series data.
• To overcome the limitations of existing methods in generating high-quality static and temporal medical data.
• To enhance the utility of synthetic medical data for downstream tasks like medical research and clinical decision-making.

Main Methods:

• Developed QAMT, a modular framework leveraging LLMs for medical time-series data generation.
• Constructed a health knowledge graph to imbue LLMs with medical expertise.
• Designed dual modules for simultaneous generation of static event and temporal data, incorporating a quality assurance module.

Main Results:

• QAMT successfully generates medical time-series data with improved quality compared to existing methods.
• The framework ensures the interpretability of the data generation process, a key advantage over traditional approaches.
• Experimental results validate the effectiveness of QAMT in producing reliable synthetic medical data.

Conclusions:

• QAMT represents a significant advancement in generating high-quality, interpretable medical time-series data using LLMs.
• The modular design and integration of a knowledge graph and quality assurance module address critical challenges in synthetic medical data generation.
• QAMT offers a promising solution for augmenting real-world medical data, thereby supporting advancements in medical research and clinical practice.