Comprehensive Summary
This study aimed to analyze the effectiveness of an AI tool, BoneView, at detecting fractures in the upper extremity in pediatric patients. Data was retrospectively collected from November 2023 to July 2019 to find patients aged 2 – 18 years with a fracture in different regions. For reference images, radiographs were sent to two different pediatric radiologists to ensure presence of a fracture, 826 of which made it through and were then analyzed by the AI model. This model had a dedicated feature to detect elbow joint effusions (EJE), which can be a key indicator of an elbow fracture. The AI detected fractures with a sensitivity of 89% and specificity of 91%, with the highest individual sensitivity and specificity at the wrist, with 96% and 94%, respectively. The lowest sensitivity was found in the fingers, at 73%, while the lowest specificity was found in the elbow, at 65%. The EJE dedicated feature did not perform as well, with a sensitivity of 79% and specificity of 51%. While the AI performed well overall, it struggled in several key areas and was significantly less accurate than trained pediatric radiologists.
Outcomes and Implications
This AI tool could be used in settings where access to healthcare is limited by providing rapid and accurate diagnoses without the input of a radiologist. While it can't currently reliably detect fractures alone, it could play an important role in improving current pediatric radiologist accuracy. Future iterations of this tool should focus on improvements in both specificity and accuracy of the lower scoring regions before being widely utilized by the medical community.