2023 ISAKOS Biennial Congress ePoster
Automated, Video-Based Analysis of Motion Patterns to Diagnose Anterior Cruciate Ligament Injuries in In-Game Video Footage
Attila Schulc, Lebanon, NH UNITED STATES
Chilan Bou Ghosson Leite, MD, PhD, Boston, MASSACHUSETTS UNITED STATES
Máté Csákvári, Lebanon, NH UNITED STATES
Luke Lattermann, Boston, MA UNITED STATES
Molly Zgoda, MS, Brighton, Massachusetts UNITED STATES
Evan M Farina, MD, Boston, Massachusetts UNITED STATES
Christian Lattermann, MD, Foxborough, MA UNITED STATES
Zoltán Tősér, Lebanon, NH UNITED STATES
Gergo B Merkely, MD, PhD, Boston, MA UNITED STATES
Brigham and Women's Hospital, Boston, MASSACHUSETTS, UNITED STATES
FDA Status Not Applicable
Summary
AI-based video analysis software may recognize motion patterns to diagnose ACL injuries in in-game video footage
ePosters will be available shortly before Congress
Abstract
Background
Failure to diagnose anterior cruciate ligament (ACL) injury during a game can delay adequate treatment and increase the risk of further injuries. Artificial intelligence (AI) has the potential to be an accurate, cost-efficient and readily-available diagnostic tool for ACL injury in-game situations.
Purpose
To develop an automated video analysis system that utilizes deep learning algorithms to identify biomechanical patterns associated with ACL injury using in-game video footage and to evaluate whether such a system assists orthopaedics and sports medicine specialists in diagnosing ACL injuries on video footage.
Study Design: Cross-sectional study. Level of evidence, 3.
Methods
Publicly available game videos of knee traumatic events were collected from athletes presenting or not having an ACL injury. Videos were trimmed, and injured athletes were recognized and tracked using a semi-annotation solution in a manner that three binary categories were identified: injury left foot ground contact and right foot ground contact. Three-dimensional (3D) poses and handcrafted features, including knee flexion, knee and hip abduction, and foot and hip rotation, were analyzed and tested with fully connected neural network (FCNN) and long short-term memory (LSTM) architectures. To evaluate if our system assists ACL injury diagnosis, two orthopaedic surgeons first watched plain clips and repeated the watching using the clip’s augmented version showing the 3D model.
Results
All four different models (FCNN and LSTM architecture normalized for 3D poses or handcrafted features) could accurately identify ACL injuries on in-game video footage, with slightly superior performance for LSTM models, particularly LSTM handcrafted. Additionally, our system helped diagnose ACL injuries and exclude ACL injuries when a traumatic event was watched in-game video footage, increasing the diagnosis precision.
Conclusion
Our AI-based analysis system effectively recognised biomechanical patterns associated with ACL injuries, precisely assisting orthopaedic surgeons in diagnosing these injuries in real-time video footage. Identification of biomechanical patterns associated with ACL injuries would contribute to recognising changes in those patterns during a season, potentially detecting players at risk for those injuries.