AI outperforms clinicians for predicting pulmonary to systemic flow ratio in congenital heart disease
Click Here to Manage Email Alerts
Artificial intelligence-based analysis of chest X-rays was able to successfully predict the pulmonary to systemic flow ratio in patients with congenital heart disease with greater efficacy than trained pediatric cardiologists, according to findings published in JAMA Cardiology.
Researchers demonstrated that, using a deep learning model that predicted the pulmonary to systemic flow ratio from chest X-rays developed using the method of transfer learning, the intraclass correlation coefficient for Fick-derived and AI-derived flow ratios was 0.68, the log-transformed bias was 0.02 and the log-transformed precision was 0.12.
“Deep learning-based analysis was able to predict a hemodynamic parameter, pulmonary to systemic flow ratio, quantitatively from chest radiographs,” Shuhei Toba, MD, of the department of thoracic and cardiovascular surgery, and Yoshihide Mitani, MD, PhD, of the department of pediatrics at the Mie University Graduate School of Medicine, Japan, told Healio. “This will confer an opportunity for quantitative and objective analysis of pulmonary blood flow status in chest radiographs in the daily practice.”
Moreover, diagnostic concordance rate of the AI model was higher compared with the experts (correctly classified 64 of 100 vs. 49 of 100 chest radiographs; P = .02 by McNemar test), according to the study.
System performance
“The analysis is reasonably fast in the daily practice,” Toba and Mitani said in an interview. “In our study, in fact, the time for completing the analysis of one radiograph was 3.6 seconds by deep learning, while about 8.9 seconds by clinicians. The speed will surely become faster by using any higher-performance clinical-level computers.”
In other findings, for detection of high pulmonary to systemic flow ratio, the sensitivity of the AI model was 0.47, the specificity was 0.95 and the area under the receiver operating curve was 0.88.
“In addition to the superb performance of the present system, deep learning may give us a novel medical insight into the chest radiograph findings predictive of hemodynamic status,” Toba and Mitani told Healio. “Specifically, the deep learning-based analysis may change our medical knowledge by uncovering previously unrecognized findings predictive of the pulmonary blood flow status.”
In this retrospective observational study, researchers included chest radiographs of 1,031 catheterizations among 657 patients (51% boys; median age, 3 years; mean Fick-derived pulmonary to systemic flow ratio, 1.43) to assess the use of AI compared with expert predictive value for pulmonary to systemic flow ratio. Researchers compared AI diagnostic concordance with three certified pediatric cardiologists.
“We need to optimize the performance of our model by teaching more cases for clinical application,” Toba and Mitani said in an interview. “In addition, we will surely make a more user-friendly clinical-quality application in collaboration with Health-Tech companies. We would start such collaboration with companies that are interested in our study.”
How clinicians can adopt AI technology
“To perform deep learning-based analysis of radiographs, clinicians need to input an x-ray image data (DICOM or other image formats) to a deep learning model which has been developed using open-source softwares (TensorFlow, Keras, etc.),” Toba and Mitani told Healio. “This process can be easily achieved if a software which integrates the deep learning model with an existing image viewer is developed.” – by Scott Buzby
Disclosures: The authors report no relevant financial disclosures.
Click Here to Manage Email Alerts