Artificial intelligence can improve reporting of urgent cases from chest x-rays

A summary of research published in the European Respiratory Journal

A growing field of medicine is using machines to help to detect and diagnose health problems. This approach, known as ‘machine learning’, involves using technology that can be trained to find patterns in large amounts of data. This technology can then help healthcare professionals make decisions and predictions.

A number of studies are being published that investigate specific sets of rules that machines can use. These rules, known as algorithms, are tested to look at how successful they can be when used to support the work of healthcare professionals. This new study looked at how a specific algorithm could be used to find 10 common lung abnormalities and whether it helped to reduce the workload and amount of time to reach a diagnosis.

What did the study look at?

Researchers developed a new algorithm using 140,000 chest x-rays, taken from 108,053 people. The machine was trained to find 10 common issues including:

  • a collapsed lung,
  • scarring of the lungs, and
  • a mass in the lungs.

The algorithm was then tested using existing collections of chest radiographs. The results were compared to the opinions of three radiologists. Six radiologists also participated in sessions to test the technology. They analysed the results with and without the algorithm and compared the results.  

What do the results show?

The results found that the algorithm could correctly find more critical abnormalities than the results from the radiologists. The radiologists were able to detect more critical and urgent cases when working with the technology. It was also able to shorten the average time taken to report crucial and urgent chest x-rays and reduced the time the radiologists spent interpreting the results.

Why is this important?

The authors believe the algorithm they developed could be useful for helping to detect health conditions with the lungs. The algorithm was successful at detecting 10 common problems with lungs and helping to improve radiologists’ performance. It could also improve the reporting time for critical and urgent cases, making it a useful tool for healthcare settings.

Read the original research paper

Title: Development and validation of a deep learning algorithm detecting 10 common abnormalities on chest radiographs