Abstract
AI-assisted chest X-ray (CXR) report generation has shown strong retrospective results, but real-world clinical evidence remains scarce, particularly for lines and tubes (L&T), which are clinically critical and disproportionately error-prone. We report a large-scale analysis of an AI CXR reporting assistant deployed in routine inpatient, emergency, and intensive-care practice at a single academic medical center, covering 10,514 deployed studies with both the model draft and the radiologist’s signed final report. We quantify draft quality using RadFact, an LLM-based factuality metric, and RAD-LT-EVAL, an L&T-specific structured-report suite. Excluding patients seen during training, we analyze two cohorts separately: 519 cases where the draft was loaded and possibly edited, and 9,995 cases where the draft was generated but not loaded (i.e. reported independently). On independently written reports the draft reaches a RadFact F1 of 56.8 and an L&T type F1 of 85.4 (compared to 77.6 and 91.8 on loaded drafts), with the largest residual gap on incorrectly placed L&T. A survey of reading radiologists is mixed on usefulness and safety, and seven of eight (87.5%) oppose default use, with free-text concerns related to L&Ts, other findings, and report style. This study contributes rare real-world evidence on factual and L&T-reporting quality compared with perceived clinical readiness of a tool.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ELAMI_014.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: Not Available
BibTex
@InProceedings{ReyMax_ARealWorld_MICCAISAT2026,
author = { Reynolds, Maxwell AND Khandelwal, Ashish AND Sharma, Harshita AND Pérez-García, Fernando AND Hansen, Michael AND Colak, Ceylan AND Mugu, Vamshi K. AND Ouellette, Heather A. AND Blasi, Marc E. AND Nadkarni, Nishant AND Swanstrom, Dana K. AND Florers, Jennifer X. AND Roering, Chris J. AND Báez Suárez, Abraham AND Cook, Cole J. AND Kline, Timothy L. AND Blezek, Daniel AND Myers, Charles R. AND Henckel, John D. AND Mensing-Diggs, Amanda AND Edwards, Matthew T. AND Korfiatis, Panagiotis},
title = { { A Real-World Analysis of an AI Chest X-ray Reporting Assistant Focused on Factuality and Lines & Tubes } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17262},
month = {pending},
page = {pending}
}