Machine learning Approaches to Radiological Dating of Infant Clavicle Fractures"

点击翻译按钮获取中文摘要

Project Summary Child abuse is a major public health problem, with 8.9/1000 children abused per year (or 656,243 victims in 2019). Infants (≤1-year-old) are particularly vulnerable to abuse, and constitute 45.4% of children who die from inflicted injuries. Most of these infants are found to have skeletal injuries at the time of death. Skeletal survey radiographs play a pivotal role in cases of suspected infant abuse for they not only establish the presence of fractures, they also identify fractures of different ages—a hallmark of infant abuse. In infant abuse investigations, dating of skeletal injuries from radiographs is vital to reaching a clear timeline of traumatic events. Investigators typically collect information to determine the presence of potential perpetrators in the victim’s environment at the time these injuries occurred. Thus, the course of subsequent legal proceedings may depend upon the dates of these fractures. Although clinical experience, along with the pertinent scientific literature, permit the radiologist to draw limited conclusions from the radiographs, the task of precisely dating injuries remains challenging. In short, there is an unmet and critical need for precise and objective dating of fractures in cases of suspected infant abuse. In this study, we propose to develop an estimation algorithm that harnesses the power of machine learning (ML)—a branch of artificial intelligence based on the idea of computers learning from data— to help date infant fractures. To this end, we will compare the accuracy performance of an ML-based algorithm to that of expert radiologists in fracture dating and will validate its clinical utility. In summary, this proposal presents a new and provocative utilization of ML: To create and clinically validate a new radiology decision support tool to further the ability of radiologists to date infant fractures with unprecedented precision and accuracy. Not only will this approach enhance our ability to date individual fractures, but in the case of multiple injuries, to determine if the fractures occurred at different times. If successful, we expect this work will aid in reaching more accurate clinical and social assessments leading to improved case outcomes.