Does Artificial Intelligence Improve the Detection of Pneumothorax on Supine Chest X-rays?

A rapid detection of pneumothorax, an accumulation of air between the chest wall and the lung, is crucial for critically ill patients. This condition can worsen if not treated promptly, especially in those on mechanical ventilation. Chest X-rays taken in the supine position are often used in emergencies, but spotting a pneumothorax on these images remains challenging even for experienced physicians.

A recent study evaluated the ability of artificial intelligence-based software to identify pneumothorax on supine chest X-rays. The results show that this software correctly detects large pneumothoraces in nearly 97% of cases, particularly those requiring drainage with a chest tube. However, its sensitivity drops to 44% for small pneumothoraces, those measuring less than 35 millimeters on a CT scan. Artificial intelligence also proves more effective at detecting air accumulation in the upper lung region than in the lower region.

The study also revealed that this software helps physicians in training to better detect pneumothoraces, increasing their detection rate from 47% to 57%. For experts, the contribution of artificial intelligence does not significantly change their ability to identify cases, but it reduces diagnostic errors by confirming the absence of pneumothorax in 96% of cases compared to 91% without assistance. This suggests that the tool acts as a safety net for less experienced practitioners and reinforces the confidence of specialists.

However, detection in the lower lung region remains a challenge, even for artificial intelligence. False positives sometimes occur due to artifacts such as skin folds or external medical devices. Physicians must therefore remain vigilant, especially in patients on mechanical ventilation or with a history of thoracic surgery.

This advancement could help avoid additional tests such as CT scans, which are often difficult to perform in unstable patients. Artificial intelligence does not replace clinical judgment, but it provides valuable support for faster and safer decisions. Further improvements are still needed to refine its accuracy, particularly in the lower lung areas where signs are more subtle.

Sources and Credits

Source Study

DOI: https://doi.org/10.1007/s10140-026-02448-4

Title: Performance of an artificial intelligence–based software in detecting pneumothorax on supine chest radiographs: a retrospective study

Journal: Emergency Radiology

Publisher: Springer Science and Business Media LLC

Authors: Hitomi Nakamura; Tomoki Wada; Ryota Inokuchi; Shouhei Hanaoka; Naoya Sakamoto; Kent Doi