Can AI Extract Echo Report Data as Accurately as Expert Annotation? - Report - MDSpire

Can AI Extract Echo Report Data as Accurately as Expert Annotation?

In a study of 50 echocardiography reports, GPT-5 mini extracted 55 cardiovascular fields from free-text echocardiography reports with 92.5% exact-match agreement with expert annotation.

  • By

  • Kerri Miller

  • June 26, 2026

  • 5 min

Share

Clinical Report: AI's Accuracy in Extracting Echo Report Data

Overview

A large language model achieved 92.5% exact-match agreement with expert annotation in extracting structured cardiovascular data from echocardiography reports.

Background

Echocardiography reports contain critical clinical information, often presented in unstructured text, which can lead to inconsistent or ambiguous findings. Accurate extraction of this data is essential for effective clinical decision-making and quality improvement.

Data Highlights

MetricValue
Exact-match agreement92.5%
Precision (severity fields)96%
Recall (severity fields)95%
Additional clinical values identified120

Key Findings

  • The large language model achieved 92.5% exact-match agreement with expert annotation.
  • Precision ranged from 96% to 98% across different categories of data.
  • Recall ranged from 85% to 95%, indicating variability in extraction accuracy.
  • The model identified 120 additional clinical values not documented by human annotators.
  • Performance varied across exam types, with stress echocardiograms showing lower performance.
  • Further validation is necessary before clinical integration of AI in echocardiography workflows.

Clinical Implications

Further validation is essential before implementing these technologies in clinical settings.

Conclusion

Further validation is necessary to ensure clinical reliability.

Related Resources & Content

  1. Barrett RB, Johns Hopkins University, 2026 -- Can AI Extract Echo Report Data as Accurately as Expert Annotation?
  2. npj Digital Medicine — EchoGraph system for automated quality assessment of echocardiography reports
  3. conexiant — Accuracy of AI Laryngeal Disorder Detection
  4. npj Digital Medicine — Exploring the Untested Hazards of AI Scribes in Healthcare Settings
  5. BMJ Health & Care Informatics — AI-generated clinical summaries: errors and susceptibility to speech and speaker variability
  6. Guidelines for the Standardization of Adult Echocardiography Reporting
  7. 2025 ESC/EACTS Valvular Heart Disease Guidelines
  8. Detecting Bicuspid Aortic Valve From Echocardiographic Reports Using Natural Language Processing: A Veterans Affairs Study | JACC: Advances

Original Source(s)

Can AI Extract Echo Report Data as Accurately as Expert Annotation?

  • by Kerri Miller

  • June 26, 2026

Related Content

Tenecteplase Fails to Boost EVT Outcomes

Phase 3 stroke trial leaves open the possibility of clinically meaningful benefit or harm in selected late-window patients.

Does Mitral Regurgitation Mechanism Matter After M-TEER?

Left ventricular diameter changes were similar across subtypes at 1 month, while pulmonary artery systolic pressure responses differed.

Postoperative Step Counts Tied to Surgical Recovery

Wearable-measured mobility—but not heart rate variability or wellness scores—was associated with length of stay, complications, and readmissions