Artificial intelligence-based software approved for use in interpreting adult chest X-rays recently displayed promising results for use in a pediatric setting.
In a sampling of 2,273 chest radiographs of kids aged 2 to 18, the AI-based software achieved diagnostic accuracies ranging from 86% to 96.9% for detecting a myriad of pathologies.
“Chest radiographs still play a critical role in pediatric radiology with even more importance than in adults, because advanced imaging studies cannot be freely performed in children,” corresponding author Eun-Kyung Kim of the Department of Radiology at Yonsei University in Seoul, South Korea, and colleagues shared. “This means the clinical necessity of the AI-based approach is no less for children than it is for adults.”
The software used for the study assessed the pediatric radiographs for nodules, consolidation, fibrosis, atelectasis, cardiomegaly, pleural effusion, pneumothorax and pneumoperitoneum. A pediatric radiologist’s assessment of the radiographs was used as the reference standard to determine the software’s detection accuracy.
The researchers found age to be a significant factor in any incorrect results. Patients two-years-old and younger accounted for more than 80% of missed or misdiagnoses.
When excluding cardiomegaly and radiographs of children aged two years and younger, the sensitivity, specificity, PPV, NPV and accuracy of the software significantly increased to 86.4%, 97.9%, 79.7%, 98.7% and 96.9%—up from 67.2%, 91.1%, 57.7%, 93.9% and 87.5%. The highest accuracy was recorded for pneumothorax.
Though the results of their work are promising, the researchers stopped short of recommending the software’s deployment in a pediatric population, citing a need for more data and further validation:
“AI-based lesion detection software can be developed and utilized for pediatric chest radiographs after further validation. AI-based lesion detection software needs to be validated in younger children with larger data to assure safe usage, and adult-oriented software can be a starting point for this.”
View the full research in Scientific Reports.
Shin, H., Son, NH., Kim, M. et al. Diagnostic performance of artificial intelligence approved for adults for the interpretation of pediatric chest radiographs. Sci Rep.