Subject-specific Modelling of Knee Joint Motion for Routine Pre-operative Planning
Professor Howe Tet Sen, Duke-NUS
Abstract:
The knee joint is prone to injuries, such as osteoarthritis and torn ligaments, which often restrict its range of motion. To restore normal knee joint motion, complex surgical interventions are needed, whose outcome depends on the surgeons' experience as well as pre-operative planning. Even a slight mistake that happens during the surgery may result in abnormal knee joint motion and accelerate the degenerative process of the knee. Therefore, a subject-specific computational model of the knee joint is used in the pre-operative planning stage to predict possible surgical outcomes for a specific subject.
This thesis describes the research work of subject-specific knee joint motion modelling for routine pre-operative planning. Specifically, the research work consists of two tasks: (1) subject-specific modelling of knee joint motion, and (2) application of the model to determine optimal attachment sites of knee ligaments. The first task recovers accurate knee joint motion of specific three-dimensional (3D) bone models that are constructed from a set of computed tomography (CT) scan of a subject's knee, which is the minimum input requirement for routine clinical practice. The second task utilizes the recovered knee joint motion to determine the optimal attachment sites of the selected ligaments, such as medial patellofemoral ligament (MPFL) and anterior cruciate ligament (ACL). There are existing works that are related to each task, but they have some shortcomings. For the first task, most existing works are not suitable for routine clinical practice as they either expose a subject to excessive radiation (multiple CTs), require complex auxiliary data (motion capture data, electromyography (EMG) signals, etc.), or are too costly (multiple magnetic resonance imaging (MRI) scans). For the second task, most of the existing works consider a small number of candidate sites and a few joint poses, which may not be sufficient to locate the optimal attachment sites of ligaments for a specific subject.
To address the limitations of existing works, this thesis first presents a subject-specific knee joint motion generation algorithm that only requires specific 3D bone models at neutral knee pose and a few anatomical landmarks placed on the bone models to generate specific knee joint motion. For the second task, this thesis presents a site identification algorithm to efficiently and reliably locate a pair of attachment sites of MPFL that best satisfies an assessment criterion over a large number of candidate sites and detailed joint poses. Comprehensive experiments evaluated several aspects of the algorithms, and they show promising results. In conclusion, the research work showcases the applications of the subject-specific knee joint motion generation algorithm in routine pre-operative planning of knee surgery.