Applying AI to chest x-rays improves care for congenital heart disease
Deep learning-based analysis of chest x-rays can be used to predict the pulmonary to systemic flow ratio of patients with congenital heart disease, according to a new study published in JAMA Cardiology.
While echocardiography and MRI can be used to assess a patient’s pulmonary to systemic flow ratio, both modalities have certain limitations. X-rays are another option for healthcare providers, the authors explained, but assessing such findings is often “subjective and qualitative.” Could deep learning be the answer?
“Previous studies have shown that a deep learning–based approach can be used to recognize diseases or findings objectively in various imaging modalities, and one of the studies showed the possibility that deep learning–based analysis may outperform clinicians,” wrote lead author Shuhei Toba, MD, Mie University Graduate School of Medicine in Japan, and colleagues. “Given the potential capability of deep learning shown in the previous studies, we hypothesized that deep learning–based analysis can predict the pulmonary to systemic flow ratio from chest radiographs quantitatively in patients with congenital heart disease.”
The researchers explored data from more than 657 patients treated from Jan. 1, 2005, to April 30, 2019 at a single facility. Seventy-eight patients—and a total of 100 cardiac catheterizations—were then chosen at random for this study.
Toba et al. used transfer learning to develop their AI models, drawing inspiration from previous studies into the effectiveness of deep learning-based analysis. The authors finalized 10 models in all, comparing their results with the assessments of three pediatric cardiologists.
Overall, deep learning achieved a “significantly higher” diagnostic concordance rate than the three cardiologists. When it came to identifying a high pulmonary to systemic flow ratio, the highest sensitivity of any deep learning model was 0.47, the highest specificity was 0.95 and the area under the receiver operating curve (AUC) was 0.88. For the cardiologists, sensitivity was 0.11, specificity was 0.94 and AUC was 0.67.
These findings, the authors noted, showed that using deep learning “may confer an objective and quantitative evaluation of chest radiographs.”
“Because of our model’s capability to quantitatively predict the pulmonary to systemic flow ratio from chest radiographs and to outperform clinicians, the present proof-of-concept study suggests that there may be hidden information in routine imaging tests that deep learning can identify, adding clinical value,” they concluded.
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.