Uncertainty-Aware Structured Data Extraction from Full CMR Reports via Distilled LLMs

2026-05-08

AI Summary

This work addresses the challenge of structuring unstructured free-text cardiac magnetic resonance (CMR) reports by proposing CMR-EXTR, the first framework in the CMR domain to integrate confidence estimation into structured information extraction. Combining large language model distillation, uncertainty quantification, and structured extraction algorithms, the approach leverages teacher–student distillation for efficient offline inference. It generates field-level confidence scores guided by three principles: distributional plausibility, sampling stability, and cross-field consistency, thereby prioritizing human review efforts. The system achieves 99.65% accuracy at the variable level, substantially reducing manual annotation burden and offering robust support for clinical cohort construction, longitudinal data integration, and automated report validation.

Abstract

Converting free-text cardiac magnetic resonance (CMR) reports into auditable structured data remains a bottleneck for cohort assembly, longitudinal curation, and clinical decision support. We present CMR-EXTR, a lightweight framework that converts free-text CMR reports into structured data and assigns per-field confidence for quality control. A teacher-student distillation pipeline enables fully offline inference while limiting manual annotation. Uncertainty integrates three complementary principles -- distribution plausibility, sampling stability, and cross-field consistency -- to triage human review. Experiments show that CMR-EXTR achieves 99.65% variable-level accuracy, demonstrating both reliable extraction and informative confidence scores. To our knowledge, this is the first CMR-specific extraction system with integrated confidence estimation. The code is available at https://github.com/yuyi1005/CMR-EXTR.

Problem

• structured data extraction
• cardiac magnetic resonance
• uncertainty estimation
• clinical reports
• data curation

Innovation

• distilled LLMs
• uncertainty estimation
• structured data extraction
• CMR reports
• teacher-student distillation