Chest Radiograph Interpretation with Deep Learning Models: Assessment with Radiologist-adjudicated Reference Standards and Population-adjusted Evaluation
The DOI matches its Crossref record; 30 of the 30 references it cites were found in a registry or an index.
Abstract
Background Deep learning has the potential to augment the use of chest radiography in clinical radiology, but challenges include poor generalizability, spectrum bias, and difficulty comparing across studies. Purpose To develop and evaluate deep learning models for chest radiograph interpretation by using radiologist-adjudicated reference standards. Materials and Methods Deep learning models were developed to detect four findings (pneumothorax, opacity, nodule or mass, and fracture) on frontal chest radiographs. This retrospective study used two data sets. Data set 1 (DS1) consisted of 759 611 images from a multicity hospital network and ChestX-ray14 is a publicly available data set with 112 120 images. Natural language processing and expert review of a subset of images provided labels for 657 954 training images. Test sets consisted of 1818 and 1962 images from DS1 and ChestX-ray14, respectively. Reference standards were defined by radiologist-adjudicated image review. Performance was evaluated by area under the receiver operating characteristic curve analysis, sensitivity, specificity, and positive predictive value. Four radiologists reviewed test set images for performance comparison. Inverse probability weighting was applied to DS1 to account for positive radiograph enrichment and estimate population-level performance. Results In DS1, population-adjusted areas under the receiver operating characteristic curve for pneumothorax, nodule or mass, airspace opacity, and fracture were, respectively, 0.95 (95% confidence interval [CI]: 0.91, 0.99), 0.72 (95% CI: 0.66, 0.77), 0.91 (95% CI: 0.88, 0.93), and 0.86 (95% CI: 0.79, 0.92). With ChestX-ray14, areas under the receiver operating characteristic curve were 0.94 (95% CI: 0.93, 0.96), 0.91 (95% CI: 0.89, 0.93), 0.94 (95% CI: 0.93, 0.95), and 0.81 (95% CI: 0.75, 0.86), respectively. Conclusion Expert-level models for detecting clinically relevant chest radiograph findings were developed for this study by using adjudicated reference standards and with population-level performance estimation. Radiologist-adjudicated labels for 2412 ChestX-ray14 validation set images and 1962 test set images are provided. © RSNA, 2019 Online supplemental material is available for this article. See also the editorial by Chang in this issue.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Only eight studies in ophthalmology imaging14,21,32,33,43,55,108,109, ten studies in respiratory imaging64,66,70,72,75,79,82,87,89,110 and six studies in breast imaging62,91,97,104,106,111 mentioned adherence to the STARD-2015 guidelines or had a STARD flow diagram in the manuscript.”
From Diagnostic accuracy of deep learning in medical imaging: a systematic review and meta-analysis (2021) · cites this paper as [82]
Citations
Cites
Fleischner Society: Glossary of Terms for Thoracic Imaging
Hansell, D. M., Bankier, A. A., MacMahon, H., McLoud, T. C., Müller, N. L., & Remy, J. (2008) · Radiology · doi:10.1148/radiol.2462070712 · Cited by 4,198 · cited 1×
“Clinical definitions for these categories were on the basis of the Fleischner Society Glos-sary of Terms for Thoracic Imaging, except for osseous fracture, which was defined as visible rib, clavicle, humeral, or vertebral body fractures (20).”
MethodArmato, S. G., McLennan, G., Bidaut, L., McNitt‐Gray, M. F., Meyer, C. R., Reeves, A. P., Zhao, B., Aberle, D. R., Henschke, C. I., Hoffman, E. A., et al. (2011) · Medical Physics · doi:10.1118/1.3528204 · Cited by 2,887
Revisiting Unreasonable Effectiveness of Data in Deep Learning Era
Sun, C., Shrivastava, A., Singh, S., & Gupta, A. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.97 · Cited by 2,484 · cited 1×
“We used Xception (22) as the convolutional neural network architecture that was pretrained on 300 000 000 natural images (23).”
MethodRajpurkar, P., Irvin, J., Ball, R. L., Zhu, K., Yang, B., Mehta, H., Duan, T., Ding, D., Bagul, A., Langlotz, C. P., et al. (2018) · PLOS Medicine · doi:10.1371/journal.pmed.1002686 · Cited by 1,449 · cited 2×
“Recent advances in deep learning offer promise for improving chest radiograph interpretation (1–4) and there are several recent reports of machine learning models achieving radiologist-level performance for different chest radiograph findings (5–7).”
BackgroundVery Deep Convolutional Networks for Text Classification
Conneau, A., Schwenk, H., Barrault, L., & Lecun, Y. (2017) · Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers · doi:10.18653/v1/e17-1104 · Cited by 1,046
Park, S. H., & Han, K. (2018) · Radiology · doi:10.1148/radiol.2017171920 · Cited by 894 · cited 1×
“However, because such data do not necessarily reflect real-world prevalence or diversity (33,34), such enrichment can also prevent meaningful clinical interpretation of diagnostic performance.”
BackgroundKrause, J., Gulshan, V., Rahimy, E., Karth, P., Widner, K., Corrado, G. S., Peng, L., & Webster, D. R. (2018) · Ophthalmology · doi:10.1016/j.ophtha.2018.01.034 · Cited by 564 · cited 1×
“…adjudicated labels, extensive work has been performed on the value of multiple interpretations in mammography (30) and other clinical settings (31,32). the most obviously positive and negative for pneumothorax, opacity, nodule or mass, or fracture for training, which can fail to address the…”
BackgroundSelf-citationMazurowski, M. A., Buda, M., Saha, A., & Bashir, M. R. (2019) · Journal of Magnetic Resonance Imaging · doi:10.1002/jmri.26534 · Cited by 554 · cited 1×
“Recent advances in deep learning offer promise for improving chest radiograph interpretation (1–4) and there are several recent reports of machine learning models achieving radiologist-level performance for different chest radiograph findings (5–7).”
BackgroundNam, J. G., Park, S., Hwang, E. J., Lee, J. H., Jin, K.-N., Lim, K. Y., Vu, T. H., Sohn, J. H., Hwang, S., Goo, J. M., et al. (2019) · Radiology · doi:10.1148/radiol.2018180237 · Cited by 539 · cited 1×
“Recent advances in deep learning offer promise for improving chest radiograph interpretation (1–4) and there are several recent reports of machine learning models achieving radiologist-level performance for different chest radiograph findings (5–7).”
BackgroundMcBee, M. P., Awan, O. A., Colucci, A. T., Ghobadi, C. W., Kadom, N., Kansagra, A. P., Tridandapani, S., & Auffermann, W. F. (2018) · Academic Radiology · doi:10.1016/j.acra.2018.02.018 · Cited by 496
Hwang, E. J., Park, S., Jin, K.-N., Kim, J. I., Choi, S. Y., Lee, J. H., Goo, J. M., Aum, J., Yim, J.-J., Cohen, J. G., et al. (2019) · JAMA Network Open · doi:10.1001/jamanetworkopen.2019.1095 · Cited by 447
Mansournia, M. A., & Altman, D. G. (2016) · BMJ · doi:10.1136/bmj.i189 · Cited by 433 · cited 1×
“For DS1, images were selected on the basis of radiology reports to enrich with radiographs positive for pneumothorax, opacity, nodule or mass, or fracture while maintaining radiograph diversity and also allowing for population adjustment at analysis by inverse probability weighting (19).”
MethodWang, X., Peng, Y., Lu, L., Lu, Z., Bagheri, M., & Summers, R. M. (2019) · Advances in Computer Vision and Pattern Recognition · doi:10.1007/978-3-030-13969-8_18 · Cited by 371
Current perspectives in medical image perception
Krupinski, E. A. (2010) · Attention, Perception & Psychophysics · doi:10.3758/app.72.5.1205 · Cited by 343 · cited 1×
“Inter-reader variability in establishing reference standard image labels can significantly impact performance evaluation (8–14).”
BackgroundSupervised learning from multiple experts: whom to trust when everyone lies a bit
Raykar, V. C., Yu, S., Zhao, L. H., Jerebko, A., Florin, C., Valadez, G. H., Bogoni, L., & Moy, L. (2009) · Proceedings of the 26th Annual International Conference on Machine Learning · doi:10.1145/1553374.1553488 · Cited by 333
Automated Triaging of Adult Chest Radiographs with Deep Artificial Neural Networks
Annarumma, M., Withey, S. J., Bakewell, R. J., Pesce, E., Goh, V., & Montana, G. (2019) · Radiology · doi:10.1148/radiol.2018180921 · Cited by 303
Kerlikowske, K., Grady, D., Barclay, J., Ernster, V., Frankel, S. D., Ominsky, S. H., & Sickles, E. A. (1998) · JNCI Journal of the National Cancer Institute · doi:10.1093/jnci/90.23.1801 · Cited by 298
Obuchowski, N. A., & Rockette, H. E. (1995) · Communications in Statistics - Simulation and Computation · doi:10.1080/03610919508813243 · Cited by 234 · cited 1×
“Model performance was compared against radiologists by using the Obuchowski-Rockette-Hillis procedure (24,25).”
MethodTaylor, A. G., Mielke, C., & Mongan, J. (2018) · PLOS Medicine · doi:10.1371/journal.pmed.1002697 · Cited by 212
Common patterns in 558 diagnostic radiology errors
Donald, J. J., & Barnard, S. A. (2012) · Journal of Medical Imaging and Radiation Oncology · doi:10.1111/j.1754-9485.2012.02348.x · Cited by 157
Barnett, M. L., Boddupalli, D., Nundy, S., & Bates, D. W. (2019) · JAMA Network Open · doi:10.1001/jamanetworkopen.2019.0096 · Cited by 154 · cited 1×
“…of adjudicated labels, extensive work has been performed on the value of multiple interpretations in mammography (30) and other clinical settings (31,32). the most obviously positive and negative for pneumothorax, opacity, nodule or mass, or fracture for training, which can fail to address the…”
BackgroundA comparison of denominator degrees of freedom methods for multiple observer ROC analysis
Hillis, S. L. (2007) · Statistics in Medicine · doi:10.1002/sim.2532 · Cited by 151 · cited 1×
“Model performance was compared against radiologists by using the Obuchowski-Rockette-Hillis procedure (24,25).”
MethodMeasuring Performance in Chest Radiography
Potchen, E. J., Cooper, T. G., Sierra, A. E., Aben, G. R., Potchen, M. J., Potter, M. G., & Siebert, J. E. (2000) · Radiology · doi:10.1148/radiology.217.2.r00nv14456 · Cited by 111
Learning From Errors in Radiology: A Comprehensive Review
Pinto, A., Caranci, F., Romano, L., Carrafiello, G., Fonio, P., & Brunese, L. (2012) · Seminars in Ultrasound CT and MRI · doi:10.1053/j.sult.2012.01.015 · Cited by 58 · cited 1×
“Inter-reader variability in establishing reference standard image labels can significantly impact performance evaluation (8–14).”
BackgroundPark, S. H. (2018) · Radiology · doi:10.1148/radiol.2018182294 · Cited by 58 · cited 1×
“However, because such data do not necessarily reflect real-world prevalence or diversity (33,34), such enrichment can also prevent meaningful clinical interpretation of diagnostic performance.”
BackgroundRosenkrantz, A. B., Duszak, R., Babb, J. S., Glover, M., & Kang, S. K. (2018) · Journal of the American College of Radiology · doi:10.1016/j.jacr.2018.05.037 · Cited by 54
Hypothesis Testing in Noninferiority and Equivalence MRMC ROC Studies
Chen, W., Petrick, N., & Sahiner, B. (2012) · Academic Radiology · doi:10.1016/j.acra.2012.04.011 · Cited by 43 · cited 1×
“%) into the numerator of the test statistic (28).”
BackgroundDouble Reading in Breast Cancer Screening: Cohort Evaluation in the CO-OPS Trial
Taylor‐Phillips, S., Jenkinson, D., Stinton, C., Wallis, M., Dunn, J., & Clarke, A. (2018) · Radiology · doi:10.1148/radiol.2018171010 · Cited by 42 · cited 1×
“Regarding the value of adjudicated labels, extensive work has been performed on the value of multiple interpretations in mammography (30) and other clinical settings (31,32). the most obviously positive and negative for pneumothorax, opacity, nodule or mass, or fracture for training, which can fail…”
BackgroundObserver Performance Methods for Diagnostic Imaging
Chakraborty, D. P. (2017) · CRC Press eBooks · doi:10.1201/9781351228190 · Cited by 38
Chen, W.-J., Wunderlich, A., Petrick, N., & Gallas, B. D. (2014) · Journal of medical imaging · doi:10.1117/1.jmi.1.3.031011 · Cited by 17 · cited 1×
“First, the presence of a chest tube is certainly highly correlated with the manifestation of pneumothorax and may influence established by using the operating point corresponding to the average radiologist sensitivity (when comparing specificity) and average radiologist specificity (when comparing sensitivity), and bi…”
Method
Research works printed but not linked: 3
- Auffermann WF, 2019, AJR Am J Roentgenol, 1
- Summers RM, NIH Chest X-ray Dataset of 14 Common Thorax Disease Categories
- Chollet F. Xception: Deep Learning with Depthwise Separable Convolutions. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, July 21–26, 2017. Piscataway, NJ: IEEE, 2017; 1800–1807.
Cited by
Diagnostic accuracy of deep learning in medical imaging: a systematic review and meta-analysis
Aggarwal, R., Sounderajah, V., Martin, G., Ting, D. S. W., Karthikesalingam, A., King, D., Ashrafian, H., & Darzi, A. (2021) · npj Digital Medicine · doi:10.1038/s41746-021-00438-z · Cited by 946
Huang, S.-C., Pareek, A., Seyyedi, S., Banerjee, I., & Lungren, M. P. (2020) · npj Digital Medicine · doi:10.1038/s41746-020-00341-z · Cited by 934
“As an example consider the “ simple ” task in radiology of identifying pneumonia on a chest radiograph, something that has been achieved by many investigators training deep learning models for automated detection and classi fi cation of pathologies on chest X-rays 19,20 .”
COVID-19 Image Data Collection: Prospective Predictions are the Future
Cohen, J. P., Morrison, P., Dao, L., Roth, K., Duong, T., & Ghassem, M. (2020) · Machine Learning for Biomedical Imaging · doi:10.59275/j.melba.2020-48g7 · Cited by 927
“, 2019]), neural networks have even reported being able to achieve performance near radiologist levels [Rajpurkar et al., 2018, 2017; Irvin et al., 2019; Putha et al., 2018a; Majkowska et al., 2019; Putha et al., 2018b].”
BackgroundDeep learning for chest X-ray analysis: A survey
Çallı, E., Sogancioglu, E., van Ginneken, B., van Leeuwen, K. G., & Murphy, K. (2021) · Medical Image Analysis · doi:10.1016/j.media.2021.102125 · Cited by 506
VinDr-CXR: An open dataset of chest X-rays with radiologist’s annotations
Nguyen, H. Q., Lam, K., Le, L. T., Pham, H. H., Tran, D. Q., Nguyen, D. B., Le, D. D., Pham, C. M., Tong, H. T. T., Dinh, D. H., et al. (2022) · Scientific Data · doi:10.1038/s41597-022-01498-w · Cited by 435
Metadata
- Authors
- Majkowska, A., Mittal, S., Steiner, D. F., Reicher, J. J., McKinney, S. M., Duggan, G. E., Eswaran, K., Cameron Chen, P.-H., Liu, Y., Kalidindi, S. R., Ding, A., Corrado, G. S., Tse, D., & Shetty, S.
- Affiliations
- Stanford Health Care (US)
- From Google Health, 1600 Amphitheatre Pkwy, Mountain View, CA 94043 (A.M., S.M., D.F.S., S.M.M., G.E.D., K.E., P.H.C.C., Y.L., G.S.C., D.T., S.S.); Stanford Healthcare and Palo Alto Veterans Affairs, Palo Alto, Calif (J.J.R.); Apollo Radiology International, Hyderabad, India (S.R.K.); and California Advanced Imaging, Novato, Calif (A.D.).
- ORCID
- Venue
- Radiology
- Published
- 2020-02
- Volume
- 294
- Issue
- 2
- Pages
- 421-431
- ISSN
- 0033-8419 (ISSN-L)
- 1527-1315
- OpenAlex
- W2991676991
- Cited by
- 277
- References
- 30
- Citations by year
- 2026: 21 · 2025: 37 · 2024: 36 · 2023: 48 · 2022: 48 · 2021: 52 · 2020: 34
- FWCI
- 22.28
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- Radiology (publishedVersion)
- PubMed (publishedVersion)
- Topics
- Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › COVID-19 diagnosis using AI
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Radiomics and Machine Learning in Medical Imaging
- Keywords
- chest X-ray interpretation, deep learning, pneumothorax detection, lung nodule detection, lung opacity, rib fracture detection, inverse probability weighting, spectrum bias, generalizability, ChestX-ray14, positive predictive value, sensitivity and specificity
- MeSH terms
- Radiologists, Deep Learning, Adolescent, Adult, Aged, Aged, 80 and over, Child, Child, Preschool, Female, Humans, Infant, Male, Middle Aged, Pneumothorax, Radiographic Image Interpretation, Computer-Assisted, Reference Standards, Respiratory Tract Diseases, Retrospective Studies, Sensitivity and Specificity, Thoracic Injuries, Radiography, Thoracic, Reproducibility of Results, Young Adult
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.