DALL-M: Context-aware clinical data augmentation with large language models.

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Clinical Data Augmentation

Background:

• Medical diagnostics heavily rely on X-ray images, but often lack sufficient clinical context for accurate disease identification.
• Integrating structured clinical features with radiology reports is crucial for enhancing diagnostic capabilities.
• Existing clinical datasets may be limited in scope and predictive power for AI-driven healthcare solutions.

Purpose of the Study:

• To introduce DALL-M, a novel framework for generating contextual synthetic clinical data to augment existing datasets.
• To enhance the integration of structured patient data, radiology reports, and domain-specific knowledge for clinical consistency.
• To improve the performance of machine learning models in medical diagnostics through data augmentation.

Main Methods:

• DALL-M employs a three-phase process: clinical context storage, expert query generation, and context-aware feature augmentation.
• Large language models (LLMs) are utilized to generate synthetic values for existing features and create new, clinically relevant features.
• The framework integrates structured patient data (vitals, demographics, radiology findings) with knowledge from reports and resources like Radiopaedia and Wikipedia.

Main Results:

• DALL-M successfully expanded the clinical features from 9 to 91 for 799 cases in the MIMIC-IV dataset.
• Empirical validation using various machine learning models showed a 16.5% improvement in F1 score.
• Significant increases in Precision (25%) and Recall (25%) were observed, demonstrating enhanced predictive modeling capabilities.

Conclusions:

• DALL-M effectively bridges the gap in clinical data augmentation by generating reliable synthetic data.
• The framework preserves data integrity while significantly enhancing the performance of AI-driven predictive models in healthcare.
• DALL-M offers a scalable and practical approach for advancing AI-driven medical diagnostics and improving patient outcomes.