Deep learning to detect acute respiratory distress syndrome on chest radiographs: a retrospective study with external validation

Sjoding, Michael W.; Taylor, Daniel; Motyka, Jonathan; Lee, Elizabeth; Co, Ivan; Claar, Dru; McSparron, Jakob I.; Ansari, Sardar; Kerlin, Meeta Prasad; Reilly, John P.; Shashaty, Michael G. S.; Anderson, Brian J.; Jones, T.K.; Drebin, Harrison; Ittner, C.A.G.; Meyer, Nuala J.; Iwashyna, Theodore J.; Ward, Kevin R.; Gillies, Christopher E.

doi:10.1016/s2589-7500(21)00056-x

Cited by 49 publications

(50 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…AI diagnostic models using biomedical imaging are increasingly being investigated to improve diagnostic accuracy and minimize the workload of radiologists. They are used to facilitate imaging diagnosis for simple tasks and have been successfully used for several disease processes, including COVID-19 [ 50 ], acute respiratory distress syndrome [ 51 ], and pneumothorax [ 52 ] detection. We envision using a similar model for rib fracture detection.…”

Section: Discussionmentioning

confidence: 99%

Barriers to and Facilitators for Acceptance of Comprehensive Clinical Decision Support System–Driven Care Maps for Patients With Thoracic Trauma: Interview Study Among Health Care Providers and Nurses

Jones¹,

Banks²,

Melton³

et al. 2022

JMIR Hum Factors

View full text Add to dashboard Cite

Background Comprehensive clinical decision support (CDS) care maps can improve the delivery of care and clinical outcomes. However, they are frequently plagued by usability problems and poor user acceptance. Objective This study aims to characterize factors influencing successful design and use of comprehensive CDS care maps and identify themes associated with end-user acceptance of a thoracic trauma CDS care map earlier in the process than has traditionally been done. This was a planned adaptive redesign stage of a User Acceptance and System Adaptation Design development and implementation strategy for a CDS care map. This stage was based on a previously developed prototype CDS care map guided by the Unified Theory of Acceptance and Use of Technology. Methods A total of 22 multidisciplinary end users (physicians, advanced practice providers, and nurses) were identified and recruited using snowball sampling. Qualitative interviews were conducted, audio-recorded, and transcribed verbatim. Generation of prespecified codes and the interview guide was informed by the Unified Theory of Acceptance and Use of Technology constructs and investigative team experience. Interviews were blinded and double-coded. Thematic analysis of interview scripts was conducted and yielded descriptive themes about factors influencing the construction and potential use of an acceptable CDS care map. Results A total of eight dominant themes were identified: alert fatigue (theme 1), automation (theme 2), redundancy (theme 3), minimalistic design (theme 4), evidence based (theme 5), prevent errors (theme 6), comprehensive across the spectrum of disease (theme 7), and malleability (theme 8). Themes 1 to 4 addressed factors directly affecting end users, and themes 5 to 8 addressed factors affecting patient outcomes. More experienced providers prioritized a system that is easy to use. Nurses prioritized a system that incorporated evidence into decision support. Clinicians across specialties, roles, and ages agreed that the amount of extra work generated should be minimal and that the system should help them administer optimal care efficiently. Conclusions End user feedback reinforces attention toward factors that improve the acceptance and use of a CDS care map for patients with thoracic trauma. Common themes focused on system complexity, the ability of the system to fit different populations and settings, and optimal care provision. Identifying these factors early in the development and implementation process may facilitate user-centered design and improve adoption.

show abstract

Section: Discussionmentioning

confidence: 99%

Barriers to and Facilitators for Acceptance of Comprehensive Clinical Decision Support System–Driven Care Maps for Patients With Thoracic Trauma: Interview Study Among Health Care Providers and Nurses

Jones¹,

Banks²,

Melton³

et al. 2022

JMIR Hum Factors

View full text Add to dashboard Cite

show abstract

“…Compared with previous studies using a similar model framework, they developed the ARDS classification model using 8073 CXRs as a training dataset and performed an excellent AUC (0.92, 95% CI 0.89–0.94). 14 Moreover, the results of Venn diagram indicated that there are some differences between the pattern of the clinical data classifier and the CNN model. Importantly, the combined machine learning model aims to provide related clinical and CXR information to identify ARDS for clinicians, which could result in shorter time to diagnosis and treatment decision-making.…”

Section: Discussionmentioning

confidence: 99%

“…In the fine-tuning step, we only retrained the parameters in the last convolutional block and subsequent layers to detect ARDS, while all the others were kept fixed after the pre-training. 14 Binary cross-entropy loss was used when adjusting model weights during training. The adaptive moment estimation (ADAM) optimizer was used to optimize the parameters of model training.…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Artificial intelligence–aided diagnosis model for acute respiratory distress syndrome combining clinical data and chest radiographs

et al. 2022

View full text Add to dashboard Cite

Objective The aim of this study was to develop an artificial intelligence–based model to detect the presence of acute respiratory distress syndrome (ARDS) using clinical data and chest X-ray (CXR) data. Method The transfer learning method was used to train a convolutional neural network (CNN) model with an external image dataset to extract the image features. Then, the last layer of the model was fine-tuned to determine the probability of ARDS. The clinical data were trained using three machine learning algorithms—eXtreme Gradient Boosting (XGB), random forest (RF), and logistic regression (LR)—to estimate the probability of ARDS. Finally, ensemble-weighted methods were proposed that combined the image model and the clinical data model to estimate the probability of ARDS. An analysis of the importance of clinical features was performed to explore the most important features in detecting ARDS. A gradient-weighted class activation mapping (Grad-CAM) model was used to explain what our CNN sees and understands when making a decision. Results The proposed ensemble-weighted methods improved the performances of the ARDS classifiers (XGB + CNN, area under the curve [AUC] = 0.916; RF + CNN, AUC = 0.920; LR + CNN, AUC = 0.920; XGB + RF + LR + CNN, AUC = 0.925). In addition, the ML model using clinical data to present the top 15 important features to identify the risk factors of ARDS. Conclusion This study developed combined machine learning models with clinical data and CXR images to detect ARDS. According to the results of the Shapley Additive exPlanations values and the Grad-CAM techniques, an explicable ARDS diagnosis model is suitable for a real-life scenario.

show abstract

“…image recognition, tumor identification, respiratory syndrome differentiation, etc. [ 4 – 6 ] Smartphone apps and other mHealth technologies have also been introduced as screening and monitoring tools for a wide range of ailments, including eye health, mental health, and a range of chronic diseases [ 7 – 9 ]. As such, digital health technology is playing an increasingly important role in clinical practice.…”

Section: Introductionmentioning

confidence: 99%

Digital health competencies in medical school education: a scoping review and Delphi method study

et al. 2022

View full text Add to dashboard Cite

Introduction In order to fulfill the enormous potential of digital health in the healthcare sector, digital health must become an integrated part of medical education. We aimed to investigate which knowledge, skills and attitudes should be included in a digital health curriculum for medical students through a scoping review and Delphi method study. Methods We conducted a scoping review of the literature on digital health relevant for medical education. Key topics were split into three sub-categories: knowledge (facts, concepts, and information), skills (ability to carry out tasks) and attitudes (ways of thinking or feeling). Thereafter, we used a modified Delphi method where experts rated digital health topics over two rounds based on whether topics should be included in the curriculum for medical students on a scale from 1 (strongly disagree) to 5 (strongly agree). A predefined cut-off of ≥4 was used to identify topics that were critical to include in a digital health curriculum for medical students. Results The scoping review resulted in a total of 113 included articles, with 65 relevant topics extracted and included in the questionnaire. The topics were rated by 18 experts, all of which completed both questionnaire rounds. A total of 40 (62%) topics across all three sub-categories met the predefined rating cut-off value of ≥4. Conclusion An expert panel identified 40 important digital health topics within knowledge, skills, and attitudes for medical students to be taught. These can help guide medical educators in the development of future digital health curricula.

show abstract

Deep learning to detect acute respiratory distress syndrome on chest radiographs: a retrospective study with external validation

Cited by 49 publications

References 25 publications

Barriers to and Facilitators for Acceptance of Comprehensive Clinical Decision Support System–Driven Care Maps for Patients With Thoracic Trauma: Interview Study Among Health Care Providers and Nurses

Barriers to and Facilitators for Acceptance of Comprehensive Clinical Decision Support System–Driven Care Maps for Patients With Thoracic Trauma: Interview Study Among Health Care Providers and Nurses

Artificial intelligence–aided diagnosis model for acute respiratory distress syndrome combining clinical data and chest radiographs

Digital health competencies in medical school education: a scoping review and Delphi method study

Contact Info

Product

Resources

About