A medical worker talks to a patient, infected with the new coronavirus, in front of a CT scanner room at Catholic University Medical Center in the southeastern city of Daegu on March 13, 2020, in this photo released by the hospital.
DAEGU, Oct. 1 (Korea Bizwire) — A new deep learning model has been developed that can automatically classify bacterial pneumonia and COVID-19 patients through the analysis of computed tomography (CT) images.
The Daegu Gyeongbuk Institute of Science and Technology said Thursday that along with eungnam University Medical Center, it had developed a new method to diagnose pneumonia using artificial intelligence (AI) technology.
The difference between COVID-19 pneumonia and bacterial pneumonia is subtle. It is difficult to check and distinguish every lesion in three-dimensional CT images.
The research team developed a new model in which AI automatically classifies major lesions in CT images without the need for doctors to check them one by one.
Using multiple instance learning, which makes a final decision on the basis of comprehensive considerations of various instances, the research team developed a new deep learning model.
To improve the performance of the model, the research team applied an attention module that can check the location of pneumonia lesions in CT images, while using contrastive learning to maximize the accuracy for each individual patient’s characteristics.
The research team noted that the new deep-learning model achieved an accuracy of 98.6 percent in diagnosis of COVID-19, greatly exceeding the performance of other previously proposed deep-learning approaches.
M. H. Lee (mhlee@koreabizwire.com)
