A deep learning algorithm that emphasizes bone suppression in the evaluation of chest X-rays for pulmonary nodules has demonstrated significantly increased sensitivity over radiologist evaluations and convolutional neural network algorithms using original chest radiographs, according to newly published research.
For the study published in JAMA Network Open, researchers compared the sensitivity rates and false-positive signs per image (FPPI) of a deep learning bone-suppressed (DLBS) model, a convolutional neural network (CNN) algorithm and radiologist evaluation of pulmonary nodules. On a chest X-ray. According to the study, the DLBS model was trained with data from 998 patients (average age 54.2) and the researchers evaluated the model on two external data sets consisting of 246 patients (average age 55.3 years) and 205 patients (average age 51.8 years). .
In external data sets, the DLBS model had a sensitivity rate of 91.5 percent and 92.4 percent compared to 79.8 percent and 80.4 percent for the CNN algorithm. The researchers also noted a slightly reduced FPPI for the DLBS model with the first external data set (.07 vs. .09 with the CNN model) and a 7 percent reduction in the second external data set (.09 vs. .16 for the CNN algorithm). .
“We hypothesized that our DLBS algorithm could generate lung parenchymal images while subtracting excessive bony structures from chest radiograph images and therefore effectively detect lung nodules from lung parenchymal images because the overlying bony structures had already been subtracted,” Jin Har, MD, Ph. . .D., who is affiliated with the Department of Radiology and Research Institute of Radiological Sciences and Clinical Image Data Science Center at Severance Hospital and Yonsei University College of Medicine in Seoul, Korea, and colleagues.
“The main finding was that our bone-suppressed model (DLBS model) can detect pulmonary nodules in chest radiographs more accurately than the original model (CNN algorithm). In addition, radiologists experienced improved nodule detection performance when assisted by the DLBS model. .
(Editor’s note: For related content, see “Deep learning model can predict lung cancer risk from a single CT scan” and “Deep learning model predicts 10-year heart disease risk from chest X-ray”.)
Using a second external data set, the researchers compared the DLBS model ratings to the ratings of three thoracic radiologists with more than five years of experience. Hur and colleagues noted a 14.6 percent higher sensitivity rate for the DLBS model (92.1 percent) compared to the average sensitivity rate of radiologists (77.5 percent). Combining the DLBS model with radiologist assessment resulted in individual increases of 12 percent, 15.3 percent, and 14.2 percent compared to sensitivity rates for individual radiologists. The study authors also pointed out that FPPI rates were lower when thoracic radiologists used the DLBS model (7.1 percent) compared to those who did not use the model (15.1 percent).
Regarding study limitations, the authors stated that selection bias is a possibility due to the validation of the deep learning model with retrospective data sets. They also noted that interstitial lung disease, pleural effusion and pneumonia were not considered in the study. Herr and colleagues maintain that a prospective multicenter study is needed to determine the feasibility of a deep learning model for use in clinical practice.