Thursday, December 4 | 9:40 a.m.-9:50 a.m. | R3-SSCH09-2 | Room E451A
While existing AI models have shown promise for detecting medical devices in adult populations, pediatric radiology remains underserved, according to this scientific presentation.
Amit Kharat, PhD, CEO and co-founder of AI startup Blitzy in Cambridge, MA, will present a model to fill in this gap, with a convolutional neural network (CNN) achieving an area under the receiver operating curve (AUROC) exceeding 0.95.
The model was developed from x-rays from 9,719 pediatric patients (average age, 7.1 years old) collected from 353 institutions and spanning over 20 different x-ray systems, with an average age of 7.1 years (0-11 years). The CNN was trained to detect the following devices and postsurgical material: sternal sutures, central venous lines, endotracheal tubes, and chest leads. Radiologists’ reports served as the reference standard.
The model achieved an overall sensitivity of 0.92, specificity of 0.81, and an AUROC of 0.86 across all medical device types. Performance was strong across individual device categories: sternal sutures (0.99 AUROC, 82% sensitivity, and 100% specificity), central venous lines (0.98 AUROC, 88% sensitivity, 96% specificity), endotracheal tubes (0.98 AUROC, 96% sensitivity, and 92% specificity), and chest leads (0.95 AUROC, 85% sensitivity, and 93% specificity).
“AI offers a scalable and sustainable approach to enhancing the accuracy, efficiency, and safety of pediatric chest imaging, with potential to significantly support radiologists and improve patient outcomes in clinical practice,” the group suggests.
Get all of the details on Thursday morning at McCormick Place.
