FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3138244435> ?p ?o ?g. }

• W3138244435 endingPage "e210004" @default.
• W3138244435 startingPage "e210004" @default.
• W3138244435 abstract "HomeRadiology: Artificial IntelligenceVol. 3, No. 2 PreviousNext CommentaryFree AccessQuantifying Pulmonary Edema on Chest RadiographsWilliam F. Auffermann William F. Auffermann Author AffiliationsFrom the Department of Radiology and Imaging Sciences, University of Utah School of Medicine, 30 North 1900 East, Room 1A71, Salt Lake City, UT 84132.Address correspondence to the author (e-mail: [email protected]).William F. Auffermann Published Online:Mar 24 2021https://doi.org/10.1148/ryai.2021210004MoreSectionsPDF ToolsImage ViewerAdd to favoritesCiteTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinked InEmail See article by Horng et al in this issue.William F. Auffermann, MD, PhD, is an associate professor of radiology and imaging sciences at the University of Utah School of Medicine. Dr Auffermann is a cardiothoracic radiologist and is ABPM board certified in clinical informatics. His research interests include imaging informatics, clinical informatics, applications of AI in radiology, medical image perception, and perceptual training. Recent research projects include image annotation for AI using eye tracking, human factors engineering, and developing simulation-based perceptual training methods to facilitate radiology education.Download as PowerPointOpen in Image Viewer The role of artificial intelligence (AI) has been evolving rapidly in medicine. A decade or two ago, computer evaluation of a full medical image such as a chest radiograph was infeasible. AI algorithms are now able to perform complicated image evaluation tasks, which only a few years ago could have only been performed by humans. The article “Deep Learning to Quantify Pulmonary Edema in Chest Radiographs” by Horng et al (1), in this issue of Radiology: Artificial Intelligence, explores using AI techniques to quantify pulmonary edema on chest radiographs.Recent advances in AI hardware and software have significantly improved the abilities and usefulness of AI algorithms performing complex tasks, such as medical image evaluation. Specifically, techniques such as deep learning have revolutionized the application of AI to medical imaging. Previously relegated to more specialized applications in radiology, AI has been applied to an increasing array of clinical care issues. The interested reader is referred to the article by Chartrand et al (2) for an introduction to the use of AI and deep learning in radiology.Pulmonary edema can be a life-threatening condition and results in more than 1 million emergency department visits per year in the United States (3). Pulmonary edema has an estimated prevalence of 75%–83% in patients with congestive heart failure (CHF) and reduced ejection fraction (4). Imaging is useful in the evaluation of pulmonary edema partially due to its ability to help diagnose disorders with overlapping symptoms, including pneumothoraces and pneumonia. The early diagnosis and treatment of pulmonary edema has been shown to decrease patient mortality (5). Chest radiographs also have been used to monitor patient treatment. To date, there has been limited work examining the ability of AI algorithms to quantify pulmonary edema.As described in their article, Horng et al developed two AI algorithms. The first algorithm was pretrained on generic images (such as images of cats and dogs). The second algorithm used semisupervised learning, starting with unlabeled chest radiographs from the MIMIC-CXR dataset. Next, both algorithms were trained using a supervised learning approach on labeled chest radiographs from the MIMIC-CXR dataset for patients with CHF and varying degrees of pulmonary edema. The chest radiographs used for training were graded and labeled by a natural language processing algorithm. After training and validation, the AI algorithms were applied to grading an independent chest radiograph dataset for model testing. The AI algorithm labeling was compared with ground truth for each test image, as defined by a panel of three senior radiology residents and an attending radiologist.The resultant models showed an ability to distinguish between various grades of pulmonary edema and were better able to differentiate between more extreme differences in the grade of edema. For example, the algorithms were better able to distinguish edema grade 0 from grade 3, compared with differentiating edema grade 1 from grade 2. On the basis of receiver operating characteristic analysis, the semisupervised model generally performed better than the pretrained model. The reason proposed for its superior performance is that the semisupervised model was able to learn the basic aspects of chest radiographic features on the unsupervised portion of its training.Prior work has been done to detect pulmonary abnormalities on chest radiographs. However, limited work has been done concerning the quantification of pulmonary edema. When deciding on the efficacy of treatment, chest radiography may be used to assess for changes. An AI algorithm such as the one presented by Horng et al may be useful to automate quantification of pulmonary edema and to monitor treatment. Although imaging may offer limited advantage in an outpatient setting where patients can report their symptoms, using chest radiography to diagnose and quantify edema may be of greater utility for patients who are unable to communicate effectively with their health care team, such as intubated patients. While their article did not focus on examining changes in patient volume status using chest radiography as a function of time, it would be an interesting topic of further study.This pilot study shows the feasibility of using AI to grade pulmonary edema. The symptoms of pulmonary edema overlap with those of other disorders of the chest, but Horng et al only consider grading chest radiographs for patients with documented CHF. Such an AI algorithm running in isolation may fail to diagnose other disease states. To translate to clinical application, the presented algorithm would need to be trained to recognize other diseases, used in conjunction with one or more additional AI algorithms focused on diagnosis of pulmonary edema and differentiation from other disease entities, or used with the understanding that the algorithm is limited to the quantification of pulmonary edema in patients with CHF.One factor that is often an issue when training AI algorithms is the quality of the “reference standard” used to label images. As the authors point out, measurement of pulmonary capillary wedge pressure is considered the reference standard to diagnose and quantify pulmonary edema. However, such a reference standard may be impractical to implement on a large scale due to its invasive nature, and chest radiography may be the best method available for approximating the ground truth.As Horng et al point out, the AI algorithms presented in their article would on occasion make rather significant errors. When implementing an AI algorithm for clinical use, it is important for a health care professional to be able to trust the algorithm’s results, understand the situations where the algorithm may be more likely to fail, and have some metric of explainability to help determine if the algorithm made the correct decision. The authors used gradient-weighted class activation mapping (Grad-CAM) to provide heatmaps that demonstrated areas of an image containing information regarding the assigned labels. Improved metrics of explainability would be helpful and likely would facilitate the adoption of such a technology into clinical practice.One limitation of this study was the exclusive use of the chest radiographs from the MIMIC-CXR dataset. The MIMIC-CXR dataset derives all its images from one source, the Beth Israel Deaconess Medical Center (6). Generally speaking, the fewer the sources of images in a dataset, the less likely an algorithm is to generalize well to images outside that dataset. The exact reason for suboptimal generalization is not fully understood, but probably is related to factors including different imaging protocols, different image acquisition hardware, and differences in patient populations across imaging centers and institutions. Generalizability is a challenge for much of the research involving the application of AI to medical images, as large multi-institutional image sets with all the relevant data elements can be difficult to procure. To demonstrate generalizable clinical utility, an AI algorithm would need to be tested on image data from several institutions.In summary, Horng et al present two AI algorithms that successfully quantify pulmonary edema on chest radiographs. This work shows the potential for new clinical applications of AI in radiology and thoracic imaging. Their article answered several questions and revealed gaps in our knowledge that would benefit from further study.Disclosures of Conflicts of Interest: W.F.A. disclosed no relevant relationships. References1. Horng S, Liao R, Wang X, Dalal S, Golland P, Berkowitz SJ; Deep learning to quantify pulmonary edema in chest radiographs. Radiol Artif Intell 2021;3(2):e190228. Link, Google Scholar2. Chartrand G, Cheng PM, Vorontsov E, et al. Deep learning: a primer for radiologists. RadioGraphics 2017;37(7):2113–2131. Link, Google Scholar3. Blecker S, Ladapo JA, Doran KM, Goldfeld KS, Katz S. Emergency department visits for heart failure and subsequent hospitalization or observation unit admission. Am Heart J 2014;168(6):901–8.e1. Crossref, Medline, Google Scholar4. Platz E, Jhund PS, Campbell RT, McMurray JJ. Assessment and prevalence of pulmonary oedema in contemporary acute heart failure trials: a systematic review. Eur J Heart Fail 2015;17(9):906–916. Crossref, Medline, Google Scholar5. Yancy CW, Jessup M, Bozkurt B, et al. 2013 ACCF/AHA guideline for the management of heart failure: a report of the American College of Cardiology Foundation/American Heart Association Task Force on practice guidelines. Circulation 2013;128(16):e240–e327. Medline, Google Scholar6. Johnson AEW, Pollard TJ, Berkowitz SJ, et al. MIMIC-CXR, a de-identified publicly available database of chest radiographs with free-text reports. Sci Data 2019;6(1):317. Crossref, Medline, Google ScholarArticle HistoryReceived: Jan 4 2021Revision requested: Jan 7 2021Revision received: Jan 12 2021Accepted: Jan 13 2021Published online: Mar 24 2021 FiguresReferencesRelatedDetailsAccompanying This ArticleDeep Learning to Quantify Pulmonary Edema in Chest Radiographs06 Jan 2021Radiology: Artificial IntelligenceRecommended Articles Deep Learning to Quantify Pulmonary Edema in Chest RadiographsRadiology: Artificial Intelligence2021Volume: 3Issue: 2The Dean Effect: An Aortic Arch Flow Artifact Mimicking DissectionRadiology: Cardiothoracic Imaging2022Volume: 4Issue: 1Review of the Chest CT Differential Diagnosis of Ground-Glass Opacities in the COVID EraRadiology2020Volume: 297Issue: 3pp. E289-E302Disease Severity Scoring for COVID-19: A Welcome Semiquantitative Role for Chest RadiographyRadiology2021Volume: 302Issue: 2pp. 470-472Hypertrophic Cardiomyopathy from A to Z: Genetics, Pathophysiology, Imaging, and ManagementRadioGraphics2016Volume: 36Issue: 2pp. 335-354See More RSNA Education Exhibits Quality Assurance for Crowdsource Annotation of the Chest X-ray 14 Dataset for the RSNA-STR Machine Learning Challenge: How We Did ItDigital Posters2018Cardiopulmonary Devices the Unknown, the Overlooked, and the MisplacedDigital Posters2018Cardiopulmonary Physiology in the ICU: Understanding Commonly Monitored Parameters with Illustrative Cases and Radiographic FindingsDigital Posters2019 RSNA Case Collection Swimming Induced Pulmonary EdemaRSNA Case Collection2021Hibernating MyocardiumRSNA Case Collection2020Re-entry High Altitude Pulmonary Edema RSNA Case Collection2021 Vol. 3, No. 2 Metrics Downloaded 582 times Altmetric Score" @default.
• W3138244435 created "2021-03-29" @default.
• W3138244435 creator A5064037990 @default.
• W3138244435 date "2021-03-01" @default.
• W3138244435 modified "2023-10-16" @default.
• W3138244435 title "Quantifying Pulmonary Edema on Chest Radiographs" @default.
• W3138244435 cites W1593615185 @default.
• W3138244435 cites W2042569545 @default.
• W3138244435 cites W2119340816 @default.
• W3138244435 cites W2767236661 @default.
• W3138244435 cites W2995225687 @default.
• W3138244435 cites W3118612102 @default.
• W3138244435 cites W4211082352 @default.
• W3138244435 doi "https://doi.org/10.1148/ryai.2021210004" @default.
• W3138244435 hasPubMedCentralId "https://www.ncbi.nlm.nih.gov/pmc/articles/8043360" @default.
• W3138244435 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/33939783" @default.
• W3138244435 hasPublicationYear "2021" @default.
• W3138244435 type Work @default.
• W3138244435 sameAs 3138244435 @default.
• W3138244435 citedByCount "2" @default.
• W3138244435 countsByYear W31382444352022 @default.
• W3138244435 countsByYear W31382444352023 @default.
• W3138244435 crossrefType "journal-article" @default.
• W3138244435 hasAuthorship W3138244435A5064037990 @default.
• W3138244435 hasBestOaLocation W31382444351 @default.
• W3138244435 hasConcept C126322002 @default.
• W3138244435 hasConcept C126838900 @default.
• W3138244435 hasConcept C141071460 @default.
• W3138244435 hasConcept C2776468924 @default.
• W3138244435 hasConcept C2777714996 @default.
• W3138244435 hasConcept C2780886150 @default.
• W3138244435 hasConcept C2989005 @default.
• W3138244435 hasConcept C36454342 @default.
• W3138244435 hasConcept C71924100 @default.
• W3138244435 hasConceptScore W3138244435C126322002 @default.
• W3138244435 hasConceptScore W3138244435C126838900 @default.
• W3138244435 hasConceptScore W3138244435C141071460 @default.
• W3138244435 hasConceptScore W3138244435C2776468924 @default.
• W3138244435 hasConceptScore W3138244435C2777714996 @default.
• W3138244435 hasConceptScore W3138244435C2780886150 @default.
• W3138244435 hasConceptScore W3138244435C2989005 @default.
• W3138244435 hasConceptScore W3138244435C36454342 @default.
• W3138244435 hasConceptScore W3138244435C71924100 @default.
• W3138244435 hasIssue "2" @default.
• W3138244435 hasLocation W31382444351 @default.
• W3138244435 hasLocation W31382444352 @default.
• W3138244435 hasLocation W31382444353 @default.
• W3138244435 hasOpenAccess W3138244435 @default.
• W3138244435 hasPrimaryLocation W31382444351 @default.
• W3138244435 hasRelatedWork W2068771676 @default.
• W3138244435 hasRelatedWork W2129363868 @default.
• W3138244435 hasRelatedWork W2136995523 @default.
• W3138244435 hasRelatedWork W21715879 @default.
• W3138244435 hasRelatedWork W2191963770 @default.
• W3138244435 hasRelatedWork W2374732650 @default.
• W3138244435 hasRelatedWork W2410701312 @default.
• W3138244435 hasRelatedWork W2411915371 @default.
• W3138244435 hasRelatedWork W2416374722 @default.
• W3138244435 hasRelatedWork W2991746746 @default.
• W3138244435 hasVolume "3" @default.
• W3138244435 isParatext "false" @default.
• W3138244435 isRetracted "false" @default.
• W3138244435 magId "3138244435" @default.
• W3138244435 workType "article" @default.