Trauma to the orbital region accounts approximately for 3% of all craniofacial injuries, with a mean hospitalization cost of $35,500 per case (Ko et al., 2013; Mims and Wang, 2019). Management of orbital fractures is challenging as a substantial amount of individual judgment is required to determine the need for surgery. The definitive indications for surgery include muscle entrapment, visible enophthalmos (sunken eye), diplopia (double vision) or hypo-globus (vertical asymmetry of the globes). Otherwise, the need for surgery is based on subjective factors such as the relative size of the fracture, which is used as a proxy for the risk of developing enophthalmos once healing has progressed. Other factors considered include calculating the change in orbital volume, presence, and extent of tissue herniation, etc. However, these methodologies lack robust sample sizes or controls for population-based variation. In addition, sophisticated systems to standardize orbital fracture surgical intervention are impractical for real-time clinical application and have low reproducibility and not adequately validated against clinical outcomes. In essence, complications arising from untreated orbital fractures pose a unique predictive analytics problem, and the need for an effective and informed risk assessment tool to help determine the need for surgical intervention is highly desired. The power of predictive analytics utilizing Artificial Intelligence (AI) provides a unique opportunity to address this issue. An innovative AI-based system will be developed to localize and quantify orbital fractures and link these measurements to multiple factors contributing to change in volume associated with orbital fracture injuries. Ultimately the analytical tools developed through this proposal will be used to develop predictive tools to link these characteristics with the likelihood of developing poor clinical outcomes like enophthalmos, hypoglobus and late-stage diplopia. This study will aid surgeons in making more informed and efficient determinations about the need for surgery in ambiguous cases. The feasibility of AI and machine learning algorithms has been proven in multiple medical applications ranging from automated breast cancer detection in mammograms to predicting the risk of fatal heart attacks in patients with chronic heart disease (Suh et al., 2020; Nag et al., 2017). An AI-based system of orbital fracture analysis will increase the ability of clinicians to accurately diagnose and determine metrics for orbital fracture repair protocols. This study will provide a proof-of-concept framework for AI applications in craniofacial trauma by proposing the following hypotheses:
Hypothesis 1: A deep learning model can independently identify the presence of orbital fractures and localize, characterize, and quantify the fractures.
Hypothesis 2: A machine-learning algorithm can reliably estimate changes in orbital volume following a fracture and evaluate the variable etiology of orbital volume change.
