Abstract Number: OC 07.4
Meeting: ISTH 2021 Congress
Background: Ultrasound provides a non-invasive diagnostic tool for assessing hemarthrosis in persons with hemophilia. Artificial intelligence (AI) algorithms applied to ultrasound images detect joint recess distensions with high accuracy in adults, however, it is unclear whether they can be applied to pediatric ultrasound images.
Aims: Assess the accuracy of an existing adult algorithm in pediatric patients and if necessary, train the algorithm to detect joint recess distensions in pediatric knee joints. Additionally, this study aimed to develop a tool for electronically guiding transducer positioning to support the use of remote ultrasound.
Methods: A dataset of 179 pediatric knee ultrasound exams (three planes: suprapatellar, medial- and lateral-parapatellar), balanced for the presence of joint recess distensions, was manually labelled to filter false positives. The existing adult algorithm was tested on both pediatric controls and cases before being retrained and retested on the pediatric dataset. An AI-based deep-learning algorithm was designed to direct users to optimal transducer positioning using audio guidance.
Results: Accuracy using the adult algorithm with pediatric images was 62%. After retraining, the binary convolutional neural network classifier detected pediatric joint recess distensions in the test set with an accuracy of 82% (83% sensitivity; 81% specificity; Figure). An ultrasound-scanning algorithm for assisting untrained users with transducer positioning for optimal image capture was successfully developed as a proof of concept.
Conclusions: Adult ultrasound AI algorithms could not be directly applied to pediatric patients, likely due to larger amounts of anechoic cartilage in immature bones, which appear similar to fluid in ultrasound images. Pediatric-specific training of an AI computer vision system resulted in high accuracy for detecting distended recesses in pediatric knees; expected to increase with additional training. Adaptation of this algorithm to elbows, ankles, and blood detection will follow. Self-guided ultrasound scanning combined with the algorithm may present an opportunity for remote management (virtual care) of hemarthrosis.
To cite this abstract in AMA style:Tyrrell P, Blanchette V, Mendez M, Paniukov D, Brand B, Zak M, Roth J. Detection of Joint Effusions in Pediatric Patients with Hemophilia Using Artificial Intelligence-assisted Ultrasound Scanning; Early Insights from the Development of a Self-management Tool [abstract]. Res Pract Thromb Haemost. 2021; 5 (Suppl 1). https://abstracts.isth.org/abstract/detection-of-joint-effusions-in-pediatric-patients-with-hemophilia-using-artificial-intelligence-assisted-ultrasound-scanning-early-insights-from-the-development-of-a-self-management-tool/. Accessed September 24, 2021.
« Back to ISTH 2021 Congress
ISTH Congress Abstracts - https://abstracts.isth.org/abstract/detection-of-joint-effusions-in-pediatric-patients-with-hemophilia-using-artificial-intelligence-assisted-ultrasound-scanning-early-insights-from-the-development-of-a-self-management-tool/