Researchers find a new way to automate the diagnosis of elbow fractures in children
A recent study conducted at Texas Children’s Hospital describes a new, automated method to efficiently and accurately diagnose elbow injuries and abnormalities in children. Published in “Radiology: Artificial Intelligence,” this is the first study to demonstrate the efficacy of using artificial intelligence methods to analyze radiographic images of pediatric patients. Findings from this study can significantly improve the point-of-care diagnosis and management of pediatric elbow injuries in primary care settings and community health centers.
Elbow injuries are very common among children and account for almost 10 percent of all pediatric fractures. As the popularity of youth recreational and competitive sports grows, the number of elbow injuries is expected to rise further.
Elbow injuries and abnormalities in children and adolescents are particularly difficult to diagnose. Unlike adults, the skeletal system of children and teenagers is still developing and therefore, has certain unique features which complicates the diagnosis. In children and teenagers, elbow joints are partly composed of cartilage, which means certain injuries cannot be detected using X-rays. Further, pediatric elbow joints have many growth centers that turn to bone (ossify) as the child grows, which means depending on the child’s age and unique developmental factors, the fracture patterns can vary significantly.
The purpose of this study was to find a way to simplify and automate the analysis of elbow radiographic images obtained from pediatric patients. To do that, the researchers tested the feasibility of using convolutional neural network (CNN), a machine-learning method most commonly used to analyze other kinds of radiographic images in adults, to reliably detect elbow injuries in children. The team analyzed more than 21,000 radiographic images of elbow injuries from pediatric patients and found the method was able to accurately and reliably identify acute pediatric elbow injuries from a series of radiographic images.
Most pediatricians and general radiologists in primary health centers, who are often front-line care providers for such cases, may not have the in-depth expertise needed to differentiate normal growth centers from elbow trauma or abnormalities. Moreover, in high-volume emergency centers/urgent care facilities, where pediatric radiologists may not always be available, there is a need to accurately and quickly diagnose these patients, which determines the subsequent clinical care the children receive. Although complications from these injuries are rare, failure to recognize and manage these injuries properly can lead to long-term growth arrest and potential deformity.
The team believes this new automated tool can greatly aid general radiologists in more accurately diagnosing various kinds of elbow trauma and abnormalities in children, improving clinical management of these cases and long-term outcomes for these children.
The study was conducted by Drs. Jesse Rayan and Nakul Reddy, who were at the time, in the diagnostic radiology residency program at Baylor. Dr. Herman Kan, chief of musculoskeletal radiology at Texas Children’s and associate professor of radiology at Baylor College of Medicine, along with Dr. Ananth Annapragada, professor and director of basic research in the Edward B. Singleton Department of Pediatric Radiology at Texas Children’s oversaw this study.