Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs

Ayis Pyrros; Stephen M. Borstelmann; Ramana Mantravadi; Zachary Zaiman; Kaesha Thomas; Brandon Price; Eugene Greenstein; Nasir Siddiqui; Melinda Willis; Ihar Shulhan; John Hines-Shah; Jeanne M. Horowitz; Paul Nikolaidis; Matthew P. Lungren; Jorge Mario Rodríguez-Fernández; Judy Wawira Gichoya; Sanmi Koyejo; Adam E. Flanders; Nishith Khandwala; Amit Gupta; John W. Garrett; Joseph Paul Cohen; Brian T. Layden; Perry J. Pickhardt; William Galanter

doi:10.1038/s41467-023-39631-x

Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs

Ayis Pyrros, Stephen M. Borstelmann, Ramana Mantravadi, Zachary Zaiman, Kaesha Thomas, Brandon Price, Eugene Greenstein, Nasir Siddiqui, Melinda Willis, Ihar Shulhan, John Hines-Shah, Jeanne M. Horowitz, Paul Nikolaidis, Matthew P. Lungren, Jorge Mario Rodríguez-Fernández, Judy Wawira Gichoya, Sanmi Koyejo, Adam E. Flanders, Nishith Khandwala, Amit GuptaJohn W. Garrett, Joseph Paul Cohen, Brian T. Layden, Perry J. Pickhardt, William Galanter

Neurology

Research output: Contribution to journal › Article › peer-review

Abstract

Deep learning (DL) models can harness electronic health records (EHRs) to predict diseases and extract radiologic findings for diagnosis. With ambulatory chest radiographs (CXRs) frequently ordered, we investigated detecting type 2 diabetes (T2D) by combining radiographic and EHR data using a DL model. Our model, developed from 271,065 CXRs and 160,244 patients, was tested on a prospective dataset of 9,943 CXRs. Here we show the model effectively detected T2D with a ROC AUC of 0.84 and a 16% prevalence. The algorithm flagged 1,381 cases (14%) as suspicious for T2D. External validation at a distinct institution yielded a ROC AUC of 0.77, with 5% of patients subsequently diagnosed with T2D. Explainable AI techniques revealed correlations between specific adiposity measures and high predictivity, suggesting CXRs’ potential for enhanced T2D screening.

Original language	English (US)
Article number	4039
Journal	Nature communications
Volume	14
Issue number	1
DOIs	https://doi.org/10.1038/s41467-023-39631-x
State	Published - Dec 2023

ASJC Scopus subject areas

Chemistry(all)
Biochemistry, Genetics and Molecular Biology(all)
Physics and Astronomy(all)

Access to Document

10.1038/s41467-023-39631-x

Cite this

Pyrros, A., Borstelmann, S. M., Mantravadi, R., Zaiman, Z., Thomas, K., Price, B., Greenstein, E., Siddiqui, N., Willis, M., Shulhan, I., Hines-Shah, J., Horowitz, J. M., Nikolaidis, P., Lungren, M. P., Rodríguez-Fernández, J. M., Gichoya, J. W., Koyejo, S., Flanders, A. E., Khandwala, N., ... Galanter, W. (2023). Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs. Nature communications, 14(1), Article 4039. https://doi.org/10.1038/s41467-023-39631-x

Pyrros, A, Borstelmann, SM, Mantravadi, R, Zaiman, Z, Thomas, K, Price, B, Greenstein, E, Siddiqui, N, Willis, M, Shulhan, I, Hines-Shah, J, Horowitz, JM, Nikolaidis, P, Lungren, MP, Rodríguez-Fernández, JM, Gichoya, JW, Koyejo, S, Flanders, AE, Khandwala, N, Gupta, A, Garrett, JW, Cohen, JP, Layden, BT, Pickhardt, PJ & Galanter, W 2023, 'Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs', Nature communications, vol. 14, no. 1, 4039. https://doi.org/10.1038/s41467-023-39631-x

@article{06f6f68fcced4f8385f4097e6ea829e7,

title = "Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs",

abstract = "Deep learning (DL) models can harness electronic health records (EHRs) to predict diseases and extract radiologic findings for diagnosis. With ambulatory chest radiographs (CXRs) frequently ordered, we investigated detecting type 2 diabetes (T2D) by combining radiographic and EHR data using a DL model. Our model, developed from 271,065 CXRs and 160,244 patients, was tested on a prospective dataset of 9,943 CXRs. Here we show the model effectively detected T2D with a ROC AUC of 0.84 and a 16% prevalence. The algorithm flagged 1,381 cases (14%) as suspicious for T2D. External validation at a distinct institution yielded a ROC AUC of 0.77, with 5% of patients subsequently diagnosed with T2D. Explainable AI techniques revealed correlations between specific adiposity measures and high predictivity, suggesting CXRs{\textquoteright} potential for enhanced T2D screening.",

author = "Ayis Pyrros and Borstelmann, {Stephen M.} and Ramana Mantravadi and Zachary Zaiman and Kaesha Thomas and Brandon Price and Eugene Greenstein and Nasir Siddiqui and Melinda Willis and Ihar Shulhan and John Hines-Shah and Horowitz, {Jeanne M.} and Paul Nikolaidis and Lungren, {Matthew P.} and Rodr{\'i}guez-Fern{\'a}ndez, {Jorge Mario} and Gichoya, {Judy Wawira} and Sanmi Koyejo and Flanders, {Adam E.} and Nishith Khandwala and Amit Gupta and Garrett, {John W.} and Cohen, {Joseph Paul} and Layden, {Brian T.} and Pickhardt, {Perry J.} and William Galanter",

note = "Funding Information: A.P., N.S., J.W.Gichoya and S.K. received funding from the U.S. Department of Health & Human Services | NIH | National Institute of Biomedical Imaging and Bioengineering (NIBIB) - 75N92020C00008 and U.S. Department of Health & Human Services | NIH | National Institute of Biomedical Imaging and Bioengineering (NIBIB) - 75N92020C00021. B.T.L received funding from the NIH (2R01 DK104927), Veterans Affairs (1I01BX00382-01A1) and Discovery Partnership Institute (DPI), through a Discovery Partners Institute (DPI) Science Team Seed Grant Program, where the DPI is part of the University of Illinois System. J.W.Garrett and P.P. received funding from the U.S. Department of Health & Human Services | NIH |U.S. National Library of Medicine (NLM) - R01LM013151. S.K. is funded by MIDRC, NSF III 2046795, IIS 1909577, CCF 1934986 and the Alfred P. Sloan Foundation. J.W.Gichoya is funded by US National Science Foundation (grant number 1928481) from the Division of Electrical, Communication & Cyber Systems and Emerging Issues, Health Disparities; and Debasing Image-Based Al Models for Population Health (EIHD2204). Michael J. Choe, MD, Monica Harrington and Samantha Baugus, Ph.D., for her valuable editing and feedback on the manuscript. The study was primarily funded by MIDRC. The funders of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report. Publisher Copyright: {\textcopyright} 2023, The Author(s).",

year = "2023",

month = dec,

doi = "10.1038/s41467-023-39631-x",

language = "English (US)",

volume = "14",

journal = "Nature communications",

issn = "2041-1723",

publisher = "Nature Publishing Group",

number = "1",

}

TY - JOUR

T1 - Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs

AU - Pyrros, Ayis

AU - Borstelmann, Stephen M.

AU - Mantravadi, Ramana

AU - Zaiman, Zachary

AU - Thomas, Kaesha

AU - Price, Brandon

AU - Greenstein, Eugene

AU - Siddiqui, Nasir

AU - Willis, Melinda

AU - Shulhan, Ihar

AU - Hines-Shah, John

AU - Horowitz, Jeanne M.

AU - Nikolaidis, Paul

AU - Lungren, Matthew P.

AU - Rodríguez-Fernández, Jorge Mario

AU - Gichoya, Judy Wawira

AU - Koyejo, Sanmi

AU - Flanders, Adam E.

AU - Khandwala, Nishith

AU - Gupta, Amit

AU - Garrett, John W.

AU - Cohen, Joseph Paul

AU - Layden, Brian T.

AU - Pickhardt, Perry J.

AU - Galanter, William

N1 - Funding Information: A.P., N.S., J.W.Gichoya and S.K. received funding from the U.S. Department of Health & Human Services | NIH | National Institute of Biomedical Imaging and Bioengineering (NIBIB) - 75N92020C00008 and U.S. Department of Health & Human Services | NIH | National Institute of Biomedical Imaging and Bioengineering (NIBIB) - 75N92020C00021. B.T.L received funding from the NIH (2R01 DK104927), Veterans Affairs (1I01BX00382-01A1) and Discovery Partnership Institute (DPI), through a Discovery Partners Institute (DPI) Science Team Seed Grant Program, where the DPI is part of the University of Illinois System. J.W.Garrett and P.P. received funding from the U.S. Department of Health & Human Services | NIH |U.S. National Library of Medicine (NLM) - R01LM013151. S.K. is funded by MIDRC, NSF III 2046795, IIS 1909577, CCF 1934986 and the Alfred P. Sloan Foundation. J.W.Gichoya is funded by US National Science Foundation (grant number 1928481) from the Division of Electrical, Communication & Cyber Systems and Emerging Issues, Health Disparities; and Debasing Image-Based Al Models for Population Health (EIHD2204). Michael J. Choe, MD, Monica Harrington and Samantha Baugus, Ph.D., for her valuable editing and feedback on the manuscript. The study was primarily funded by MIDRC. The funders of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report. Publisher Copyright: © 2023, The Author(s).

PY - 2023/12

Y1 - 2023/12

N2 - Deep learning (DL) models can harness electronic health records (EHRs) to predict diseases and extract radiologic findings for diagnosis. With ambulatory chest radiographs (CXRs) frequently ordered, we investigated detecting type 2 diabetes (T2D) by combining radiographic and EHR data using a DL model. Our model, developed from 271,065 CXRs and 160,244 patients, was tested on a prospective dataset of 9,943 CXRs. Here we show the model effectively detected T2D with a ROC AUC of 0.84 and a 16% prevalence. The algorithm flagged 1,381 cases (14%) as suspicious for T2D. External validation at a distinct institution yielded a ROC AUC of 0.77, with 5% of patients subsequently diagnosed with T2D. Explainable AI techniques revealed correlations between specific adiposity measures and high predictivity, suggesting CXRs’ potential for enhanced T2D screening.

AB - Deep learning (DL) models can harness electronic health records (EHRs) to predict diseases and extract radiologic findings for diagnosis. With ambulatory chest radiographs (CXRs) frequently ordered, we investigated detecting type 2 diabetes (T2D) by combining radiographic and EHR data using a DL model. Our model, developed from 271,065 CXRs and 160,244 patients, was tested on a prospective dataset of 9,943 CXRs. Here we show the model effectively detected T2D with a ROC AUC of 0.84 and a 16% prevalence. The algorithm flagged 1,381 cases (14%) as suspicious for T2D. External validation at a distinct institution yielded a ROC AUC of 0.77, with 5% of patients subsequently diagnosed with T2D. Explainable AI techniques revealed correlations between specific adiposity measures and high predictivity, suggesting CXRs’ potential for enhanced T2D screening.

UR - http://www.scopus.com/inward/record.url?scp=85164248540&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85164248540&partnerID=8YFLogxK

U2 - 10.1038/s41467-023-39631-x

DO - 10.1038/s41467-023-39631-x

M3 - Article

C2 - 37419921

AN - SCOPUS:85164248540

SN - 2041-1723

VL - 14

JO - Nature communications

JF - Nature communications

IS - 1

M1 - 4039

ER -

Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs

Abstract

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this