Deep-Learning Software Tool
Built on CAD4TB, an existing deep-learning software tool for detecting tuberculosis in chest X-rays, CAD4COVID was developed and validated by Thirona using 24,678 chest X-ray images. It was then tested on X-rays of 454 patients suspected to have COVID-19 pneumonia. Of these patients, 223 were COVID-19 confirmed cases and 231 tested negative for the virus. The test set was also independently and blindly evaluated by six radiologists with five to 24 years of experience. The research conducted by the Dutch hospital Radboud University Medical Center in collaboration with Jeroen Bosch Hospital and Bernhoven Hospital showed that the AI system performed comparably to the radiologists.
Support in Triaging COVID-19 Cases
Guido Geerts, CEO of Delft Imaging and Thirona, said about the promising test results: “CAD4COVID-XRay is meant to provide support for radiologists and clinicians in chest x-ray assessment as part of the COVID-19 triage process. The publication of this research in a journal like Radiology is a great milestone for our initiative. Results achieved by the AI-system compared to radiologist readings are noteworthy, because the presentation of COVID-19 pneumonia on chest X-ray can be highly variable, making differentiation from other diseases challenging. We expect that as we collect more data, the performance of the algorithm will improve even further.”
Since it’s launch on March 31, Delft Imaging and Thirona have made CAD4COVID-XRay available to more than 20 healthcare facilities across 15 countries and also launched software for COVID-19 detection on CT scans. Dr. Henk Smits, radiologist at Bernhoven Hospital commented: “Bernhoven Hospital has seen some of the highest numbers of COVID-19 cases in the Netherlands during the pandemic. At our hospital, a chest X-ray is done on all incoming patients. When the chest X-ray is normal or otherwise not suspect for COVID-19, a chest CT is performed. In the case of an abnormal chest X-ray, an RT-PCR test is conducted without the need for a CT scan, alleviating the burden on the CT. The CAD4COVID tool offers a second opinion for our radiologists: in the case of higher abnormality scores or abnormal areas on the heatmap, we can pay more attention to these cases and relevant lung areas.”