A new computer-aided screening tool created by KAUST researchers may be able to help eliminate some of the difficulties associated with assessing lung health after a viral illness.
COVID-19, like other respiratory infections, can cause long-term lung damage, but clinicians have struggled to envision this. Traditional chest scans are ineffective at detecting evidence of lung scarring and other pulmonary issues, making it difficult to monitor the health and recovery of those who have prolonged respiratory issues or other post-COVID difficulties.
Deep-Lung Parenchyma-Enhancing (DLPE), a new approach discovered by KAUST, integrates artificial intelligence algorithms on top of regular chest imaging data to discover otherwise undetectable visual characteristics suggestive of lung disease.
With DLPE intensification, “radiologists can discover and analyze novel sub-visual lung lesions,” says computer scientist and computational biologist Xin Gao. “Analysis of these lesions could then help explain patients’ respiratory symptoms,” enhancing disease supervision and treatment, he adds.
The tool was developed by Gao and members of his Structural and Functional Bioinformatics Group, as well as artificial intelligence scientist and current KAUST Provost Lawrence Carin and clinical associates from Harbin Medical University in China.
The technique starts by removing any anatomical characteristics that are not related to the lung parenchyma—the tissues engaged in gas exchange are the primary sites of COVID-19 induced damage. This entails eliminating airways and blood arteries, then boosting the images of what is left to reveal lesions that would otherwise go undetected without the computer’s assistance.
The scientists used CT chest scans from hundreds of individuals who have been hospitalized in China with COVID-19 to test and evaluate their algorithms. They improved the approach with the help of specialist radiologists, and then used it in a prospective study on dozens of COVID-19 victims with lung difficulties, all of whom had serious illnesses that required urgent care.
Gao and his coworkers showed that the method could detect symptoms of pulmonary fibrosis in COVID long-haulers, attempting to explain shortness of breath, coughing, and other lung issues. He claims that traditional CT image analytics would be unable to make such a diagnosis.
With DLPE, for the first time, we proved that long-term CT lesions can explain such symptoms. Thus, treatments for fibrosis may be very effective at addressing the long-term respiratory complications of COVID-19.
Xin Gao, Computer Scientist and Computational Biologist, King Abdullah University of Science And Technology
Although the KAUST researchers designed DLPE with post-COVID recovery in mind, they also evaluated it on chest scans from individuals with pneumonia, tuberculosis, and lung cancer. The authors indicated how their tool might be used as a general diagnostic aid for all lung disorders, allowing radiologists to “see the unseen,” as Gao describes it.
Zhou, L., et al. (2022) An interpretable deep learning workflow for discovering subvisual abnormalities in CT scans of COVID-19 inpatients and survivors. Nature Machine Intelligence. doi.org/10.1038/s42256-022-00483-7.