Deep learning helps radiologists detect malignant lung nodules
Commercially available deep convolutional neural network–based software helped 12 radiologists in an international multi-center study improve their detection of malignant lung nodules in chest X-rays.
Results of the retrospective study, funded by Samsung Electronics and published in the journal Radiology, showed that the average sensitivity of the radiologists was improved by 5.2 percent when they re-reviewed X-rays with the deep learning software.
“Radiologists had better performance with deep convolutional network software for the detection of malignant pulmonary nodules on chest radiographs than without,” conclude the study’s authors. “The DCNN software not only helped the readers find missed nodules but also helped dismiss false-positives.”
An editorial accompanying the study in Radiology points out that missed lung cancer is a source of concern among radiologists.
“In 90 percent of cases, errors in diagnosis occur on chest radiographs,” writes Francine Jacobson, MD, radiologist at Brigham and Women's Hospital and assistant professor of radiology at Harvard Medical School.
However, with the help of the Samsung Auto Lung Nodule Detection software, which is commercially available in Europe and South Korea, the number of false positives--incorrectly reporting that cancer is present--per X-ray declined from 0.2 for radiologists alone to 0.18 with the help of the software.
“Computer-aided detection software to detect lung nodules has not been widely accepted and utilized because of high false positive rates, even though it provides relatively high sensitivity,” says Byoung Wook Choi, MD, professor at Yonsei University College of Medicine, and cardiothoracic radiologist in the Department of Radiology in the Yonsei University Health System in Seoul, Korea.. “DCNN may be a solution to reduce the number of false positives.”