Uncertainty-Aware Structured Data Extraction from Full CMR Reports via Distilled LLMs
Yi Yu, Parker Martin, Zhenyu Bu, Yixuan Liu, Yi-Yu Zheng + 3 more
TLDR
CMR-EXTR extracts structured data from free-text cardiac MRI reports with high accuracy and provides per-field confidence scores using distilled LLMs.
Key contributions
- Converts free-text CMR reports into structured data.
- Assigns per-field confidence scores for robust quality control.
- Utilizes a distilled LLM for efficient, fully offline inference.
- Achieves 99.65% variable-level extraction accuracy.
Why it matters
This paper tackles the bottleneck of converting free-text CMR reports into structured data, crucial for medical research and decision-making. CMR-EXTR offers a highly accurate, uncertainty-aware solution, significantly improving data quality and efficiency. This advances clinical decision support and cohort analysis.
Original 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.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.