Chest Radiograph AI Diagnostic Tool for COVID-19 (M Health Fairview)
model2026-01-24https://doi.org/10.1148/atlas.1769275256266
141

Overview

Schema Version

https://atlas.rsna.org/schemas/2025-11/model.json

Name

Chest Radiograph AI Diagnostic Tool for COVID-19 (M Health Fairview)

Link

https://dx.doi.org/10.1148/ryai.210217

Indexing

Keywords: COVID-19, chest radiograph, clinical decision support, deep learning, prospective validation, external validation, model drift, equity analysis, sensitivity, specificity
Content: CH, ER, IN
RadLex: RID34696, RID5350, RID43261
SNOMED: 75570004, 840539006

Author(s)

Ju Sun
Le Peng
Taihui Li
Dyah Adila
Zach Zaiman
Genevieve B. Melton-Meaux
Nicholas E. Ingraham
Eric Murray
Daniel Boley
Sean Switzer
John L. Burns
Kun Huang
Tadashi Allen
Scott D. Steenburg
Judy Wawira Gichoya
Erich Kummerfeld
Christopher J. Tignanelli

Organization(s)

University of Minnesota
Emory University
M Health Fairview Informatics
Indiana University
North Memorial Health Hospital

Version

1.0

License

Text: © 2022 Radiological Society of North America, Inc. Article available via the PMC Open Access Subset for unrestricted re-use during the COVID-19 pandemic with acknowledgment of the original source; thereafter PMC has a perpetual license to host the article.

Contact

Corresponding author: Christopher J. Tignanelli, MD, MS (email: ude.nmu@enangitc)

Funding

Microsoft AI for Health COVID-19 grant (GPU support); AHRQ/PCORI K12HS026379; NIH NCATS KL2TR002492 and UL1TR002494; NIH NHLBI T32HL07741; NIBIB 75N92020D00018/75N92020F00001; NIBIB MIDRC 75N92020C00008 and 75N92020C00021; NSF #1928481; University of Minnesota OVPR COVID-19 Impact Grant.

Ethical review

Approved by the University of Minnesota IRB with waiver of consent (STUDY 00011158). External validation: Indiana University IRB exempt (STUDY 2010169012); Emory University IRB approved (STUDY 00000506).

Date

Published: 2022-06-01

References

[1] Sun J, Peng L, Li T, Adila D, Zaiman Z, Melton-Meaux GB, Ingraham NE, Murray E, Boley D, Switzer S, Burns JL, Huang K, Allen T, Steenburg SD, Gichoya JW, Kummerfeld E, Tignanelli CJ. "Performance of a Chest Radiograph AI Diagnostic Tool for COVID-19: A Prospective Observational Study". Radiology: Artificial Intelligence. 2022 Jul;4(4):e210217.. . doi:10.1148/ryai.210217. PMID: 35923381. PMCID: PMC9344211.

Model

Architecture

Deep learning pipeline comprising lung segmentation, outlier detection (GAN-based), and convolutional neural network feature extraction/classification; interpretable heatmaps generated from gradients.

Availability

Model not publicly available; can be made available upon request for investigative purposes.

Clinical benefit

Assists clinicians by providing an image-based COVID-19 likelihood score from chest radiographs to support diagnostic decision-making.

Clinical workflow phase

Clinical decision support systems in the emergency department and early inpatient care.

Decision threshold

Institution-specific thresholds selected to maximize the Youden index during validation.

Degree of automation

Decision support; provides scores and heatmaps to assist, not replace, clinician diagnosis.

Indications for use

Adult patients (≥18 years) in ED or inpatient settings undergoing chest radiography with unknown or negative COVID-19 status to assess likelihood of COVID-19 infection.

Input

Frontal chest radiographs (DICOM) from adult ED and inpatient encounters.

Instructions

Integrated into the Epic EHR; evaluates adult ED/inpatient chest radiographs in real time; provides a 0–1 COVID-19 diagnostic score and heatmap; institution-specific thresholding recommended.

Limitations

Underperforms compared with board-certified radiologists; reduced performance for mild disease; validation using public datasets yielded unrealistically high AUC; performance varies by site requiring institution-specific thresholds; segmentation may miss regions obscured by heart/diaphragm; controls not limited to suspected COVID-19 population; model evaluated only for unknown/negative COVID-19 status; models were trained/validated on fixed data and may need updating.

Output

CDEs: RDE339, RDE193
Description: COVID-19 diagnostic score (0–1 likelihood) with interpretable heatmap highlighting image regions contributing to the prediction.

Recommendation

Use as an adjunct to, not a replacement for, clinical decision-making and confirmatory testing.

Reproducibility

Performance reported across multiple sites with institution-specific thresholds; model not publicly released; real-time performance stable over 19 weeks without drift.

Use

Intended: Detection and diagnosis
Out-of-scope: Diagnosis
Excluded: Diagnosis

User

Intended: Physician, Radiologist, Referring provider
Out-of-scope: Patient
Excluded: Layperson