Knee injuries are the most common lower limb pathologies in competitive or recreational sports. Post-trauma consequences lead to lower performances and pauses in activity. Moreover, despite specific prevention guidelines, the recurrence rate, whether from a rupture of the ligament graft or a contralateral injury, remains very high.
Based on an experimental network platform developed in collaboration between the universities of Bordeaux, Grenoble, and Marseille, which brings together complementary expertise in movement analysis, signal processing, and mathematical modeling, this thesis project aimed to improve knee injury prevention through a multifactorial approach. Relying on a database of 96 individuals built during the doctoral research, we hypothesized that accounting for multifactorial data related to injury (biomechanical and psychological data) and integrating them into a model based on artificial intelligence algorithms would allow better monitoring of injury progression and recovery over time, particularly to determine the optimal time for returning to sport.
The first study highlighted the reproducibility of data obtained from the tests performed. The second study demonstrated that it is possible, using an artificial intelligence model and biomechanical and psychological data, to classify individuals based on their history of knee injury. Finally, the third study showed that using all measurement points rather than characteristic variables allowed artificial intelligence models to achieve better performances. Thus, all these studies highlight the predictive capability of the measurements made during the protocol to recognize knee injury sequelae. Additionally, the use of interpretability algorithms in the second and third studies allowed us to identify the functional variables that most influenced the models in classifying individuals.
This work therefore optimizes the timing of return to sport and the rehabilitation process by recognizing the presence of knee injury sequelae in individuals and identifying the most important functional variables to work on.