Artificial Intelligence Model Accurately Detects Pneumothorax on Chest Radiographs

Use of an artificial intelligence (AI) model can assist radiologists in the accurate detection of simple pneumothorax and tension pneumothorax, according to study results published in JAMA Open Network.

The researchers sought to assess the accuracy of an AI model in detecting the presence of pneumothorax and tension pneumothorax in chest radiography by comparing the performance of AI with the consensus interpretations of thoracic radiologists. The investigators conducted a diagnostic, retrospective, standalone performance assessment using a data set of 1000 chest radiographs captured between June 1, 2015, and May 31, 2021. All radiographs had been obtained from adult inpatients and outpatients at 4 hospitals in Mass General Brigham, a hospital network in the Northeastern US.

Investigators followed 2 strategies in choosing chest radiographs; the first approach identified consecutive radiographs with pneumothorax via a manual review of radiology reports, whereas the second approach identified consecutive radiographs with tension pneumothorax with the use of natural language processing. With both approaches, negative radiographs were selected by taking the next negative radiograph obtained from the same radiography machine as each positive radiograph. The final data set comprised a combination of these processes.

Each of the radiographs was interpreted independently by up to 3 radiologists, in an effort to establish consensus ground-truth interpretations. Next, each radiograph was interpreted by the AI model for the presence of pneumothorax and tension pneumothorax. The primary study endpoint was the areas under receiver operating characteristic curves (AUCs) for the detection of pneumothorax and tension pneumothorax. Secondary endpoints included the sensitivities and specificities for the detection of pneumothorax and tension pneumothorax.

Radiographs from a total of 985 patients were utilized for analysis. Overall, 44.3% of the patients were female. The mean (SD) participant age was 60.8 (19.0) years. Among the radiographs used in the analysis, 44.2% (435 of 985) were deemed positive for pneumothorax (307 revealing nontension pneumothorax and 128 revealing tension pneumothorax), and 55.8% (550 of 985) were considered negative for pneumothorax. A total of 128 radiographs were positive for tension pneumothorax (29.4% of radiographs with pneumothorax; 13.0% of all radiographs).

Results of the study demonstrated that the AI model identified pneumothorax (incorporating nontension and tension pneumothorax) with an AUC of 0.979 (95% CI, 0.970 to 0.987), a sensitivity of 94.3% (95% CI, 92.0% to 96.3%), and a specificity of 92.0% (95% CI, 89.6% to 94.2%). Further, the AI model identified tension pneumothorax with an AUC of 0.987 (95% CI, 0.980 to 0.992), a sensitivity of 94.5% (95% CI, 90.6% to 97.7%), and a specificity of 95.3% (95% CI, 93.9% to 96.6%).

A key limitation of the current study is the fact that it is a retrospective analysis that was conducted outside of the clinical workflow. Thus, although the results demonstrate accuracy of the AI model in interpretation of the imaging across many demographic and technical subgroups, it does not do so within the broader clinical environment.

“This diagnostic study assessed an AI model that accurately detected pneumothorax and tension pneumothorax. Its use in the clinical environment may lead to improved care for patients with pneumothorax,” concluded study authors.

Disclosure: Some of the study authors have declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.