A study at the University of Waterloo in Canada has demonstrated that a system powered by an artificial intelligence (AI) technology could accurately identify collapsed lungs or pneumothorax using chest X-rays.

The AI technology is designed to analyse a database comprising more than 550,000 X-ray images with known diagnoses, as well as compare with chest images of patients with an unknown condition.

If the technology finds that the known diagnosis in most of the similar X-rays is a collapsed lung, it recommends the same diagnosis.

During the study, the system showed higher accuracy in diagnosing collapsed lungs compared to radiologists.

On average, medical specialists detect less than 50% of collapsed lungs cases from chest X-rays, while the AI technology was able to pinpoint 75% of cases.

According to the university’s systems design engineering professor Hamid Tizhoosh, most severe cases of a collapsed lung are easy to identify than minor cases, leading to missed diagnoses in up to 50% of patients.

Tizhoosh said: “We spend a lot of time, energy and resources needlessly investigating other possible causes of the same symptoms and people suffer in the meantime.”

Waterloo researchers have partnered with the University Health Network (UHN) on an AI project supported by the non-profit Vector Institute.

The team aims to enhance the accuracy of AI technology to more than 90%. They also plan to integrate the technology next year into the Coral Review software system created and used at UHN-affiliated hospitals.

If successful, the technology will be expanded to additional hospitals using the Coral system, which can be used by doctors to offer a second opinion after reviewing the diagnosis of medical images made by their peers.

The system should help specialists prioritise cases that need review, decreasing treatment delays.

The AI search system could also be expanded to include additional conditions identified from X-ray images such as pneumonia and chronic obstructive pulmonary disease (COPD).