Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Report-Angel: AI-Powered Automation in Endoscopy Reporting
Meet Report-Angel: an AI system that watches endoscopy videos and writes clinical reports—maintaining accuracy while easing endoscopists’ workload.
Background
Writing endoscopy reports is exhausting, repetitive work. Endoscopists spend countless hours documenting findings from upper GI procedures, and it’s easy to miss details or introduce inconsistencies. Report-Angel tackles this by training a multimodal large language model on 20,617 image-text pairs to automatically generate detailed draft reports.
Key Findings
Prospective validation delivered solid clinical results:
- 79.3% clinically acceptable reports in the internal cohort, 83.3% in external validation
- 88.51% case-level completeness and 78.93% accuracy
- 1.5 seconds to process each lesion
- Lesion-level accuracy: 91.92% (retrospective images), 89.07% (single-center prospective videos), 83.94% (multi-center prospective videos)
Why It Matters
Beyond speed, Report-Angel could standardize endoscopy reports across clinics and free endoscopists from documentation grind, creating more consistent foundation reports for clinical review.
Limitations
Performance dipped when the system encountered multi-center video data, suggesting generalization challenges across clinical settings. That gap needs addressing before wider deployment.
Original paper: Domain specific multimodal large language model for automated endoscopy reporting with multicenter prospective validation. — NPJ digital medicine. 10.1038/s41746-025-00897-5
