Sage Journals: Discover world-class research

Abstract

French

Background:

Pneumothorax is a common acute presentation in healthcare settings. A chest radiograph (CXR) is often necessary to make the diagnosis, and minimizing the time between presentation and diagnosis is critical to deliver optimal treatment. Deep learning (DL) algorithms have been developed to rapidly identify pathologic findings on various imaging modalities.

Purpose:

The purpose of this systematic review and meta-analysis was to evaluate the overall performance of studies utilizing DL algorithms to detect pneumothorax on CXR.

Methods:

A study protocol was created and registered a priori (PROSPERO CRD42023391375). The search strategy included studies published up until January 10, 2023. Inclusion criteria were studies that used adult patients, utilized computer-aided detection of pneumothorax on CXR, dataset was evaluated by a qualified physician, and sufficient data was present to create a 2 × 2 contingency table. Risk of bias was assessed using the QUADAS-2 tool. Bivariate random effects meta-analyses and meta-regression modeling were performed.

Results:

Twenty-three studies were selected, including 34 011 patients and 34 075 CXRs. The pooled sensitivity and specificity were 87% (95% confidence interval, 81%, 92%) and 95% (95% confidence interval, 92%, 97%), respectively. The study design, use of an institutional/public data set and risk of bias had no significant effect on the sensitivity and specificity of pneumothorax detection.

Conclusions:

The relatively high sensitivity and specificity of pneumothorax detection by deep-learning showcases the vast potential for implementation in clinical settings to both augment the workflow of radiologists and assist in more rapid diagnoses and subsequent patient treatment.

Visual Abstract

This is a visual representation of the abstract.

Keywords

artificial intelligence pneumothorax deep learning imaging chest X-ray

Introduction

Chest radiographs (CXRs) are one of the most common imaging examinations requested in acute inpatient and outpatient settings. Pneumothorax is a potentially life-threatening condition requiring urgent clinical attention.¹ A CXR is a common first-line imaging study used when a pneumothorax is suspected. As imaging examination volumes continue to rise, artificial intelligence (AI) can serve an important role in identifying high acuity conditions, such as pneumothorax, and alerting the radiologist and referring clinician.^2,3 Quicker identification may result in faster clinical intervention for the patient. Past studies have suggested that AI assisted interpretation can improve performance and efficiency for identifying pneumothorax on CXR.⁴

Computer-aided image segmentation typically utilizes convolutional neural networks, a subfield of AI. This model is comprised of multiple layers of connected weighted nodes, inspired by neurons in the brain.⁵ During the training process, the neural network will initially randomly assign weights to the nodes, comparing its outputs to the real “ground truth” label. Based on output results, the weights are adjusted until a satisfactory performance has been achieved.⁵ In computer-aided pneumothorax detection, a training set of both normal and abnormal chest radiographs are labeled by an expert (typically a radiologist or physician in a different specialty). They are entered into a convolutional neural network and will train the algorithm. This deep learning (DL) process attempts to learn the image features.⁶ After analysis and adjustments, the network is tested against a new set of chest radiographs for assessment of diagnostic accuracy. The results of the test phase are often compared to the true interpretation of the image, as decided by an experienced radiologist or another physician.

There are multiple studies on computer-aided detection of pneumothorax on chest radiograph, however, their results are varied.^2,7 Understanding the value of its addition to clinical practice can serve as a valuable step in the implementation of this technology. As a result, we performed a diagnostic test accuracy systematic review and meta-analysis to assess the performance of computer-aided pneumothorax detection, as trained by DL, on CXR.

Methods

A study protocol was created and registered a priori (PROSPERO CRD42023391375). Research Ethics Board approval was not sought as all data are available in the public domain. Contemporary guidelines and reporting criteria were followed based on the Cochrane handbook and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Diagnostic Test Accuracy (PRISMA DTA) statement.^8,9 Literature resources were utilized for strategies.¹⁰

Search Strategy

A comprehensive literature search was performed to identify and select studies that assess the diagnostic performance of computer-aided detection of pneumothorax on adult CXRs. MEDLINE, EMBASE, Cochrane Central Register of Controlled Trials (CENTRAL), and Scopus databases were searched for potentially relevant studies. The search was conducted by an experienced hospital librarian (RC) and included studies published up until January 10, 2023. The search strategy is provided in the Supplemental Materials.

The following inclusion criteria were used: (1) the patients in the study were adults (>18 years of age); (2) study discussed computer-aided detection of pneumothorax on CXR; (3) Dataset was evaluated by a qualified physician (ie, Radiologist, Emergency Department physician); (4) sufficient data to construct 2 × 2 contingency table provided. Studies that did not meet these criteria were excluded.

Study Selection

Literature results were exported to an Excel spreadsheet to facilitate the selection. Title and abstract review was conducted by a radiology fellow with 5 years of experience performing systematic reviews (N.Z.) to ensure papers met the inclusion criteria. Full-text screening was conducted by 2 medical students with 2 years of experience conducting systematic reviews (B.K., N.I.), and N.Z. An initial pilot phase with 3 studies was performed to ensure consistency. Results were discussed and discrepancies were adjudicated by a radiologist with 25 years of experience (M.P.).

Data Extraction

Data extraction was done by 3 authors (BK, NI, NZ). An initial pilot phase with 5 studies was conducted and discussed to ensure consistency, with discrepancies adjudicated by MP. The following metrics were extracted: first author surname, journal, year of publication, study funding, study design, patient age, patient sex, total number of studies, source of studies (eg, publicly available database or independent dataset), number of patients with pneumothorax, how the reference standard was determined, cross-validation, performance scores (ie, sensitivity, specificity, area under curve [AUC]). A true positive was a CXR that AI and an expert opinion concur that a pneumothorax is present. A false negative was a CXR that had a pneumothorax, as identified by an expert, but was missed by AI. A true negative was a CXR that AI and the expert agreed that there was no pneumothorax. A false positive was a CXR that AI deemed to be positive for pneumothorax but an expert disagreed.

Quality Assessment and Risk of Bias

Risk of Bias in individual studies was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool modified for the specific study question.¹¹ The following criteria were assessed for each included study: patient selection, index test, reference standard, flow, and timing. This was done by 3 authors (BK, NI, NZ); with discrepancies resolved via discussion with a fourth author (MP). Overall study risk of bias was considered “high” if the risk of bias in any category was considered “high” or if the risk of bias was “unclear” in 2 or more categories. For patient selection, a study was considered low risk of bias if a random selection of patients was used for inclusion. For the index test, utilizing a separate image set for training/validation of the algorithm versus testing was classified as a low risk of bias. The reference standard was deemed low risk if the training/validation set were labeled by a qualified physician (eg, radiologist). A study was given an “unclear” or “high” rating if the data set source was not stated, or if not confirmed by a qualified physician. For flow and timing, the study was deemed low risk if all CXRs were included in the test set analysis.

Statistical Analysis

A bivariate random-effects model meta-analysis was performed to determine estimates of the mean for the sensitivity, specificity, positive and negative likelihood ratios of computer-aided detection of pneumothorax on chest radiographs with 95% confidence intervals.¹² Coupled forest plots and hierarchical summary receiver operating characteristic (hsROC) curves were created using the estimated model parameters.

Sources for variability in accuracy were explored through meta-regression. A multivariate meta-regression model was utilized, and the following variables were assessed for their impact on sensitivity and specificity: study design factors (single vs multicentre), use of an institutional training/validation data set versus commercially available software, and risk of bias. As per contemporary guidance for diagnostic accuracy systematic reviews, publication bias was not assessed.^13,14 Analysis was performed using the “midas,” “metandi,” and “metaprop” packages in STATA version 11.2 (Texas, United States), as well as the “mada” package in R version 3.5.1 (Auckland, New Zealand).^12,15

Results

Search Results and Selection Characteristics

The literature search yielded 835 unique studies, and ultimately 23 met the inclusion criteria, with a total of 34 011 patients and 34 075 CXRs.^16-38 A study flow diagram is shown in Figure 1. The studies were published between 2016 and 2023 from various institutions around the world. All studies were retrospective in design; 18 were multi-centre studies, while 5 were single-centre studies. The amount of CXRs used to evaluate the algorithms ranged from 425 to 4574. Table 1 summarizes the study characteristics.

Figure 1.

Diagram of search and selection process.

Table 1.

Study Characteristics.

Study	Year	Country of patient sample	Study design	Centres	Training/validation set	Number of images in testing set	TP	FN	TN	FP
Agrawal and Choudhary¹⁶	2023	India	Retrospective	Multi-centre	Public	1376	210	155	79	932
Ahn et al¹⁷	2022	USA	Retrospective	Multi-centre	Institutional & Public	497	79	6	1	411
Park et al¹⁸	2019	South Korea	Retrospective	Multi-centre	Public	2972	251	160	2	2559
Cicero et al¹⁹	2016	Canada	Retrospective	Single centre	Institutional	4574	167	857	47	3037
Gipson et al²⁰	2022	Australia	Retrospective	Single centre	Institutional	1400	67	2	104	1227
Hallinan et al²¹	2021	Singapore	Retrospective	Single centre	Institutional	525	106	56	19	344
Hillis et al²²	2022	USA	Retrospective	Multi-centre	Institutional	985	435	48	26	550
Hwang et al²³	2019	South Korea	Retrospective	Multi-centre	Institutional	1015	113	3	6	407
Iqbal et al²⁴	2022	Pakistan	Retrospective	Multi-centre	Public	2036	297	365	48	1326
Kim et al²⁵	2022	South Korea	Retrospective	Multi-centre	Institutional	4000	828	68	172	2932
Kitamura and Deible²⁶	2020	USA	Retrospective	Multi-centre	Institutional	1335	48	97	14	1287
Lai and Ma²⁷	2023	Hong Kong	Retrospective	Multi-centre	Institutional	175	53	11	62	49
Lee et al²⁸	2022	South Korea	Retrospective	Multi-centre	Institutional	1050	250	2	12	500
Mosquera et al²⁹	2021	Argentina	Retrospective	Multi-centre	Public	999	97	14	22	866
Nam et al³⁰	2021	South Korea	Retrospective	Multi-centre	Private & Public	863	23	3	0	164
Rudolph et al³¹	2022	Germany	Retrospective	Multi-centre	Institutional	563	50	25	8	480
Park et al³²	2019	South Korea	Retrospective	Single centre	Institutional	1379	151	79	96	1053
Thian et al³³	2022	Singapore	Retrospective	Single centre	Institutional	525	107	12	18	388
Tian et al³⁴	2021	China	Retrospective	Multi-centre	Public	1684	784	50	58	792
Wang et al³⁵	2020	China	Retrospective	Single centre	Institutional	2213	959	21	124	1109
Wang et al³⁶	2020	China	Retrospective	Multi-centre	Public	1372	226	238	64	844
Wang et al³⁷	2022	China	Retrospective	Multi-centre	Public	1372	266	80	23	1003
Yi et al³⁸	2020	USA	Retrospective	Multi-centre	Public	602	150	141	26	285

Note. TP: True Positive; FN: False Negative; TN: True Negative; FP: False Positive.

Quality Assessment

The risk of bias was assessed for each study using the QUADAS-2 criteria. A summary of the assessment is provided in Table 2. 20 of the 23 studies were deemed to be at low risk of bias, while 3 studies were assessed to have a high risk of bias. The most frequent source of bias was “Reference Standard,” where the source data was unclear or not directly stated. In addition, if the patient selection method from a large database was not specified, the study would receive an “unclear” bias rating for “Patient Selection.”

Table 2.

Risk of Bias Based on QUADAS-2 Tool for Each Included Study.

Study	Patient selection	Index test	Reference standard	Flow and timing	Overall assessment

Note. All low are coloured green, unclear are yellow and high are red. Colour is used to distinguish them from one another.

Meta Analysis and Regression

Pooled estimates of the mean sensitivity and specificity forest plots and the hsROC curve are displayed in Figures 2 and 3 respectively. The pooled sensitivity was 87% (95% confidence interval [CI], 81%, 92%) and pooled specificity was 95% (95% CI, 92%, 97%). The area under the curve (AUC) of the hsROC is 97% (95% CI, 95%, 98%).

Figure 2.

Forest plot of selected studies.

Figure 3.

Hierarchical summary receiver operating characteristic curve for AI pneumothorax detection.

A comparative meta-regression to examine the impact of multiple covariates on the sensitivity and specificity was performed and is summarized in Table 3. There was no significant effect of study design, use of an institutional training/validation set versus commercially available software, or risk of bias on the sensitivity or specificity (P = .07-.86).

Table 3.

Meta Regression Model Evaluating Impact of Several Covariates on Sensitivity and Specificity.

Covariates	Beta coefficient (95% CI)	Standard error	P value
Sensitivity
Study design: Single—reference multi-centre design	−0.98 (−2.04, 0.08)	0.54	.07
Use of training/validation data set—reference commercially available software	−0.43 (−1.56, 0.70)	0.58	.46
QUADAS Risk of bias: low—reference high	−0.14 (−1.61, 1.34)	0.75	.86
Specificity
Study design: Single—reference multi-centre design	−0.42 (−1.73, 0.89)	0.67	.53
Use of training/validation data set—reference commercially available software	0.58 (−0.80, 1.96)	0.70	.41
QUADAS Risk of bias: low—reference high	0.99 (−0.84, 2.83)	0.94	.29

Discussion

This systematic review and meta-analysis assessed the performance of pneumothorax detection on CXRs by DL algorithms, with a total of 23 studies, 34 011 patients, and 34 075 CXRs included in the analysis. The pooled sensitivity and specificity were 87% (95% CI, 81%, 92%) and 95% (95% CI, 92%, 97%), respectively. The meta-regression models found that study design, use of an institutional training/validation set versus commercially available software, and risk of bias had no significant effect on the sensitivity and specificity of pneumothorax detection.

The ability to detect pneumothorax rapidly and reliably can have critical impacts on patient care. As imaging volumes increase, it is necessary to diagnose critical findings efficiently.³⁹ Our findings demonstrate the remarkable capabilities of DL algorithms for pneumothorax detection on CXR. One similar review study found a pooled sensitivity of 84% and specificity of 96% for the performance of DL for pneumothorax detection across all imaging modalities, in close agreement with our results.⁴⁰ Our study focused specifically on pneumothorax detection in CXRs (which are the first line imaging technique for pneumothorax) and used meta-regression to further evaluate several factors that may have an impact on performance. DL algorithms have also been developed to examine a variety of other common CXR pathologies, such as pneumonia. In a recent systematic review and meta analysis, DL detected pneumonia with a 98% sensitivity and 94% specificity.⁴¹

The performance of DL algorithms is often compared to the capabilities of radiology trainees (ie, residents, fellows) and radiologists. In one study, junior residents generally performed better than DL algorithms detecting pneumothorax, though the algorithm could interpret images 1000 times faster.³⁸ In a similar study, DL generally performed at a comparable level to mid-level radiology residents at producing preliminary reports of CXR reads.⁴² Limitations of DL technology generally include a higher false positive rate and factors such as pneumothorax size and presence of tubes/lines affecting performance.^43,44

The consensus amongst a majority of the literature is that DL, and AI in general, serve to greatly augment the clinical duties of a radiologist. Several studies suggest that DL provides an additive benefit to interpreting radiologists, resulting in improved performance and faster interpretation times for those who utilized DL.^45-48 Beyond the role of DL in conjunction with radiologists, there is a benefit to smaller community centres. For instance, the role of AI may be of substantial clinical value in emergency departments that do not have 24/7 radiology coverage.³¹ Furthermore, it has been deployed to assist with image triaging, allowing for a system to prioritize images that contain more urgent findings and reduce reporting delay.^49,50 As pneumothorax is a common presentation in the emergency department, deployment of algorithms to prioritize and detect pneumothorax will likely expedite care for patients. Our study supports the performance of DL in pneumothorax detection and demonstrates the exciting potential of these algorithms.

Our study has several limitations. The literature search was performed through several major databases, therefore gray literature was not included. Studies that were published in languages other than English were excluded as well. The quality of CXR included in each study could not be evaluated. Meta-regression was used to evaluate the influence of several variables, though other potentially important covariates such as experience level of interpreting radiologist were not considered. Specific subgroup analysis, such as size or cause (eg, trauma) of pneumothorax was not performed. The influence of a larger training set was not assessed. Finally, studies including pediatric imaging were excluded.

In conclusion, this systematic review and meta-analysis/regression evaluated the performance of pneumothorax detection on CXR by DL algorithms. As AI remains a field of ongoing development, the next generation of algorithms will hopefully continue to improve in performance. The promising results of this study provide an exciting foundation toward integrating DL technology into the emergency radiology workflow.

Supplemental Material

sj-docx-1-caj-10.1177_08465371231220885 – Supplemental material for Deep Learning for Pneumothorax Detection on Chest Radiograph: A Diagnostic Test Accuracy Systematic Review and Meta Analysis

Supplemental material, sj-docx-1-caj-10.1177_08465371231220885 for Deep Learning for Pneumothorax Detection on Chest Radiograph: A Diagnostic Test Accuracy Systematic Review and Meta Analysis by Benjamin D. Katzman, Mostafa Alabousi, Nabil Islam, Nanxi Zha and Michael N. Patlas in Canadian Association of Radiologists Journal

Footnotes

Abbreviations

AI Artificial Intelligence

CXR Chest X-ray

hsROC hierarchical summary receiver operating characteristic

QUADAS Quality Assessment of Diagnostic Accuracy Studies

Author Contributions

All authors attest that they meet the current International Committee of Medical Journal Editors (ICMJE) criteria for Authorship.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Funding

The author(s) received no financial support for the research,authorship,and/or publication of this article.

Informed Consent and Patient Details

The authors declare that this report does not contain any personal information that could lead to the identification of the patients.

ORCID iDs

Benjamin D. Katzman

Mostafa Alabousi

Michael N. Patlas

Supplemental Material

Supplemental material for this article is available online.

References

Yarmus

Feller-Kopman

Pneumothorax in the critically ill patient. Chest. 2012;141:1098-1105. doi:10.1378/chest.11-1691

Kao

C-Y

Lin

C-Y

Chao

C-C

, et al. Automated Radiology Alert System for Pneumothorax Detection on Chest Radiographs Improves Efficiency and Diagnostic Performance. Diagnostics. 2021;11:1182. doi:10.3390/diagnostics11071182

Katzman

van der Pol

Soyer

Patlas

MN.

Artificial intelligence in emergency radiology: a review of applications and possibilities. Diagn Interv Imaging. 2023;104:6-10. doi:10.1016/j.diii.2022.07.005

Deng

Lim

X-Y

Luo

T-Y

Lee

M-H

Lin

T-C.

Application of deep learning techniques for detection of pneumothorax in chest radiographs. Sensors (Basel). 2023;23:7369. doi:10.3390/s23177369

Tang

Tam

Cadrin-Chênevert

, et al. Canadian association of radiologists white paper on artificial intelligence in radiology. Can Assoc Radiol J. 2018;69:120-135. doi:10.1016/j.carj.2018.02.002

Cheng

Montagnon

Yamashita

, et al. Deep learning: an update for radiologists. RadioGraphics. 2021;41:1427-1445. doi:10.1148/rg.2021200210

Taylor

Mielke

Mongan

Automated detection of moderate and large pneumothorax on frontal chest X-rays using deep convolutional neural networks: a retrospective study. PLoS Med. 2018;15:e1002697. doi:10.1371/journal.pmed.1002697

Salameh

J-P

Bossuyt

McGrath

, et al. Preferred reporting items for systematic review and meta-analysis of diagnostic test accuracy studies (PRISMA-DTA): explanation, elaboration, and checklist. BMJ. 2020;370:m2632. doi:10.1136/bmj.m2632

Cohen

Deeks

Hooft

, et al. Preferred reporting items for journal and conference abstracts of systematic reviews and meta-analyses of diagnostic test accuracy studies (PRISMA-DTA for Abstracts): checklist, explanation, and elaboration. BMJ. 2021;372:n265. doi:10.1136/bmj.n265

10.

Alabousi

Soyer

Patlas

MN.

Writing a successful systematic review manuscript for a radiology journal. Can Assoc Radiol J. 2023;74:471-473. doi:10.1177/08465371221120594

11.

Whiting

Rutjes

Westwood

, et al. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. 2011;155:529-536. doi:10.7326/0003-4819-155-8-201110180-00009

12.

Reitsma

Glas

Rutjes

AWS

, et al. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol. 2005;58:982-990. doi:10.1016/j.jclinepi.2005.02.022

13.

McInnes

MDF

Moher

Thombs

, et al. Preferred reporting items for a systematic review and meta-analysis of diagnostic test accuracy studies: the PRISMA-DTA statement. JAMA. 2018;319:388-396. doi:10.1001/jama.2017.19163

14.

McGrath

Alabousi

Skidmore

, et al. Recommendations for reporting of systematic reviews and meta-analyses of diagnostic test accuracy: a systematic review. Syst Rev. 2017;6:194. doi:10.1186/s13643-017-0590-8

15.

Vogelgesang

Schlattmann

Dewey

The evaluation of bivariate mixed models in meta-analyses of diagnostic accuracy studies with SAS, Stata and R. Methods Inf Med. 2018;57:111-119. doi:10.3414/ME17-01-0021

16.

Agrawal

Choudhary

ALCNN: Attention based lightweight convolutional neural network for pneumothorax detection in chest X-rays. Biomed Signal Proccess Control. 2023;79:104126. doi:10.1016/j.bspc.2022.104126

17.

Ahn

Ebrahimian

McDermott

, et al. Association of artificial intelligence–aided chest radiograph interpretation with reader performance and efficiency. JAMA Netw Open. 2022;5:e2229289. doi:10.1001/jamanetworkopen.2022.29289

18.

Park

Cho

Lee

, et al. A curriculum learning strategy to enhance the accuracy of classification of various lesions in chest-PA X-ray screening for pulmonary abnormalities. Sci Rep. 2019;9:15352. doi:10.1038/s41598-019-51832-3

19.

Cicero

Bilbily

Colak

, et al. Training and validating a deep convolutional neural network for computer-aided detection and classification of abnormalities on frontal chest radiographs. Invest Radiol. 2017;52:281-287. doi:10.1097/rli.0000000000000341

20.

Gipson

Tang

Seah

, et al. Diagnostic accuracy of a commercially available deep-learning algorithm in supine chest radiographs following trauma. Br J Radiol. 2022;95:20210979. doi:10.1259/bjr.20210979

21.

Hallinan

JTPD

Feng

, et al. Detection of pneumothorax with deep learning models: learning from radiologist labels vs natural language processing model generated labels. Acad Radiol. 2022;29:1350-1358. doi:10.1016/j.acra.2021.09.013

22.

Hillis

Bizzo

Mercaldo

, et al. Evaluation of an artificial intelligence model for detection of pneumothorax and tension pneumothorax in chest radiographs. JAMA Netw Open. 2022;5:e2247172. doi:10.1001/jamanetworkopen.2022.47172

23.

Hwang

Park

Jin

K-N

, et al. Development and validation of a deep learning–based automated detection algorithm for major thoracic diseases on chest radiographs. JAMA Netw Open. 2019;2:e191095. doi:10.1001/jamanetworkopen.2019.1095

24.

Iqbal

Shaukat

Akram

Muzaffar

Mustansar

Byun

YC.

A hybrid VDV model for automatic diagnosis of pneumothorax using class-imbalanced chest X-rays dataset. IEEE Access. 2022;10:27670-27683. doi:10.1109/ACCESS.2022.3157316

25.

Kim

Rim

Choi

Lee

Min

Hong

Deep learning in multi-class lung disease’s classification on chest X-ray images. Diagnostics (Basel). 2022;12:915. doi:10.3390/diagnostics12040915

26.

Kitamura

Deible

Retraining an open-source pneumothorax detecting machine learning algorithm for improved performance to medical images. Clin Imaging. 2020;61:15-19. doi:10.1016/j.clinimag.2020.01.008

27.

Lai

KHA

. Sensitivity and specificity of artificial intelligence with Microsoft Azure in detecting pneumothorax in emergency department: a pilot study. Hong Kong J Emerg Med. 2023;30:8-15. doi:10.1177/1024907920948994

28.

Lee

Jeon

, et al. Localization-adjusted diagnostic performance and assistance effect of a computer-aided detection system for pneumothorax and consolidation. NPJ Digit Med. 2022;5:107. doi:10.1038/s41746-022-00658-x

29.

Mosquera

Diaz

Binder

, et al. Chest x-ray automated triage: a semiologic approach designed for clinical implementation, exploiting different types of labels through a combination of four Deep Learning architectures. Comput Methods Programs Biomed. 2021;206:106130. doi:10.1016/j.cmpb.2021.106130

30.

Nam

Kim

Park

, et al. Development and validation of a deep learning algorithm detecting 10 common abnormalities on chest radiographs. Eur Respir J. 2021;57:2003061. doi:10.1183/13993003.03061-2020

31.

Rudolph

Huemmer

Ghesu

F-C

, et al. Artificial intelligence in chest radiography reporting accuracy: added clinical value in the emergency unit setting without 24/7 radiology coverage. Invest Radiol. 2022;57:90-98. doi:10.1097/rli.0000000000000813

32.

Park

Lee

Kim

, et al. Application of deep learning–based computer-aided detection system: detecting pneumothorax on chest radiograph after biopsy. Eur Radiol. 2019;29:5341-5348. doi:10.1007/s00330-019-06130-x

33.

Thian

Hallinan

JTPD

, et al. Deep learning systems for pneumothorax detection on chest radiographs: a multicenter external validation study. Radiol Artif Intell. 2021;3:e200190. doi:10.1148/ryai.2021200190

34.

Tian

Wang

Yang

Wang

Qian

Deep multi-instance transfer learning for pneumothorax classification in chest X-ray images. J Med Phys. 2022;49:231-243. doi:10.1002/mp.15328

35.

Wang

Liu

Luo

, et al. Automated segmentation and diagnosis of pneumothorax on chest X-rays with fully convolutional multi-scale ScSE-DenseNet: a retrospective study. BMC Med Inform Decis Mak. 2020;20:317. doi:10.1186/s12911-020-01325-5

36.

Wang

Qin

Wang

CheXLocNet: automatic localization of pneumothorax in chest radiographs using deep convolutional neural networks. PLoS One. 2020;15:e0242013. doi:10.1371/journal.pone.0242013

37.

Wang

Yang

Lan

, et al. Automatic segmentation of pneumothorax in chest radiographs based on a two-stage deep learning method. IEEE Trans Cogn Develop Syst. 2022;14:205-218. doi:10.1109/TCDS.2020.3035572

38.

Kim

Bennett

Eng

Lin

CT.

Can AI outperform a junior resident? Comparison of deep neural network to first-year radiology residents for identification of pneumothorax. Emerg Radiol. 2020;27:367-375. doi:10.1007/s10140-020-01767-4

39.

Zha

Patlas

Duszak

Jr.

Radiologist burnout is not just isolated to the United States: perspectives from Canada. J Am Coll Radiol. 2019;16:121-123. doi:10.1016/j.jacr.2018.07.010

40.

Sugibayashi

Walston

Matsumoto

Mitsuyama

Miki

Ueda

Deep learning for pneumothorax diagnosis: a systematic review and meta-analysis. Eur Respir Rev. 2023;32:220259. doi:10.1183/16000617.0259-2022

41.

Zhang

Dai

Dong

Badrigilan

Accuracy of deep learning for automated detection of pneumonia using chest X-Ray images: a systematic review and meta-analysis. Comput Biol Med. 2020;123:103898. doi:10.1016/j.compbiomed.2020.103898

42.

Wong

KCL

Gur

, et al. Comparison of chest radiograph interpretations by artificial intelligence algorithm vs radiology residents. JAMA Netw Open. 2020;3:e2022779. doi:10.1001/jamanetworkopen.2020.22779

43.

Rueckel

Trappmann

Schachtner

, et al. Impact of confounding thoracic tubes and pleural dehiscence extent on artificial intelligence pneumothorax detection in chest radiographs. Invest Radiol. 2020;55:792-798. doi:10.1097/rli.0000000000000707

44.

Lind Plesner

Müller

Brejnebøl

, et al. Commercially available chest radiograph AI tools for detecting airspace disease, pneumothorax, and pleural effusion. Radiology. 2023;308:e231236. doi:10.1148/radiol.231236

45.

Pehrson

Lauridsen

, et al. The added effect of artificial intelligence on physicians’ performance in detecting thoracic pathologies on CT and chest X-ray: a systematic review. Diagnostics (Basel). 2021;11:2206. https://www.mdpi.com/2075-4418/11/12/2206

46.

Spear

Nikpanah

, et al. Augmented radiologist workflow improves report value and saves time: a potential model for implementation of artificial intelligence. Acad Radiol. 2020;27:96-105. doi:10.1016/j.acra.2019.09.014

47.

Boeken

Feydy

Lecler

, et al. Artificial intelligence in diagnostic and interventional radiology: where are we now? Diagn Interv Imaging. 2023;104:1-5. doi:10.1016/j.diii.2022.11.004

48.

Soyer

Fishman

Rowe

Patlas

Chassagnon

Does artificial intelligence surpass the radiologist?

Diagn Interv Imaging. 2022;103:445-447. doi:10.1016/j.diii.2022.08.001

49.

Weisberg

Chu

Fishman

EK.

The first use of artificial intelligence (AI) in the ER: triage not diagnosis. Emerg Radiol. 2020;27:361-366. doi:10.1007/s10140-020-01773-6

50.

Annarumma

Withey

Bakewell

Pesce

Goh

Montana

Automated triaging of adult chest radiographs with deep artificial neural networks. Radiology. 2019;291:196-202. doi:10.1148/radiol.2018180921

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB