AIRFrac: Artificial Intelligence Radiographic Point-of-Care Decision-Making Aid for Prehospital Fracture and Dislocation in Military and Civilian Populations
Abstract
Fractures and dislocations are the most common orthopedic trauma injuries among Service Members in both combat and non-combat situations. Effective treatment relies on imaging diagnosis, which requires a well-trained, board- certified radiologist to read and interpret the images. However, studies have found that as workload of radiologists increases over the years, many fractures were missed in the civilian emergency department in the night and when trainees were involved in the initial interpretation. Overlooked fractures left untreated may severely aggravate, resulting in long-term consequences such as osteoarthritis (joint narrowing) or even disability. In response, automated fracture detectors have been developed by applying Artificial Intelligence (AI) technologies. Several AI fracture detection systems have cleared U.S. Food and Drug Administration (FDA) approval for clinical use. Wide adoption and improvement in diagnosis quality are yet to be seen, but they demonstrate the potential of the AI in assisting diagnosis of fractures. Such AI systems may potentially be deployed in military health care to alleviate the workload of well-trained medical personnel. Moreover, they can be deployed in the battlefield near the point-of-injury to provide rapid diagnoses of fractures and prioritize which injured Warfighters need immediate attention and which are cleared with fractures, performed without the presence of experienced, well-trained medical personnel. The potential benefit is particularly eminent in mass casualty events. The project aims at addressing the fiscal year 2022 PRORP Applied Research Award Focus Area Retention Strategies - Battlefield Care: Strategies that can be utilized at or near the point of injury to allow an injured Service Member to remain on the battlefield or on mission without the need for evacuation by developing an effective AI detector of orthopedic trauma injuries. The existing AI fracture detectors, however, may not be ready for deployment in the battlefield for many reasons. First, it is not clear if their systems work well if the images are acquired with portable X-ray machines available in the forward field medical aid stations. Their systems were tested on fixed X-ray machines in a hospital. Portable X-ray machines may not have the same image quality as those fixed machines. Second, their systems may require high-end computers to run and may not work with a tablet used by military medics. Also, none of them was evaluated at localizing where fractures present but only at the case level – whether this case has a fracture or no fracture, which is not sufficient to provide an actionable treatment recommendation. Lastly, none of them considers dislocations, subluxations, and other orthopedic trauma injuries. These are important considerations for an AI fracture detector to benefit the point of injury care in deployed environments. The proposed project will develop AIRFrac, a novel AI software system for fracture and dislocation detection. AIRFrac will also detect dislocations because an automated AI detector will be the most useful for assisting screening of patients with mild and moderate injuries. Confirmation of the presence of dislocations in addition to fractures will be beneficial for these patients and our preliminary result with dislocation detection was encouraging. AIRFrac will be trained by deep machine learning algorithms with a large data set of fracture and dislocation cases sampled from the trauma center of UC San Diego Health and eight VA medical centers in the western states. The cases will match the intended population of Service Members. AIRFrac will be developed on top of previous research of the study team on AI for radiology supported by DOD. Preliminary results with foot and ankle showed that our idea can achieve a fracture detection rate comparable to the state-of-the-art AI systems. Fractures and dislocations of 17 body parts will be cons
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 04, 2024
- Source ID
- HT94252310320
Entities
People
- Chun-Nan Hsu
Organizations
- United States Army
- University of California, San Diego