Abstract:
The shoulder complex is an intricate joint system of the human body that provides support and movement for the upper limb. It consists of four uniquely shaped bones that are responsible for three different articulations. An in-depth knowledge of shoulder motion would help clinicians, biomechanics researchers and physiotherapists to understand the complexities of the shoulder function, injury mechanisms and allow them to suggest preventive measures and improve treatment regimens. Determining shoulder complex kinematics accurately using the data from non-invasive methods such as optical motion capture (OMC) with skin-mounted markers is a challenging task. This is primarily due to the deformation of the soft tissue causing relative motion between the markers and underlying skeletal structure. This effect is known as soft tissue artefacts (STAs) and it leads to violation of rigid body assumptions, which can result in significant errors in joint angle calculations. Furthermore, the joint system’s anatomy also makes it difficult to infer kinematics of the clavicle reliably with skinmounted markers due to lack of well-defined bony landmarks. The use of homogeneously scaled generic models to simulate the motion and calculate joint kinematics is another source of error. The scaling and registration of a generic model is performed with the aid of anatomical landmarks. Since the investigator determines these landmarks, the accuracy of identifying them depends on the investigator’s experience and is therefore subjective. The methods and results presented in this thesis focused on investigating the uncertainties and errors in predicting the shoulder complex kinematics from motion data. The following three studies were conducted to determine these uncertainties and errors. 1. A novel method was developed for registering a subject specific model to the OMC data using the Microsoft Kinect device and medical imaging. 2. The degree in which STAs affect the motion of the shoulder was assessed using bi-planar xray videoradiography (BXV) technology simultaneously with OMC data. 3. In order to understand the nature of soft tissue deformation and investigate the effectiveness of linear methods for reducing its effects, two methods of estimating rigid body parameters were considered. The first method used singular value decomposition (SVD) of the meancentred cross-dispersion matrix while the other approach employed the polar decomposition (PD) of the deformation gradient tensor from the affine transformation. Using the proposed registration method, the need for palpated skeletal landmarks and scaling was eliminated. This has the benefit of accurately registering the model to the OMC data regardless of palpation experience. Additionally, the markers placed on the subject do not have to match any particular marker configuration because the registration process will automatically place the OMC markers onto the virtual model. This method also has the benefit of easily quantifiable registration errors. It was found that for all the tasks involving the shoulder complex considered in this study, the kinematics derived from the OMC data did not give an accurate account of the rigid body motion. It was also unclear what type of soft tissue deformation (shearing, scaling and homogeneous/ heterogeneous) was the major cause of the errors. The accuracy of the two methods of estimating rigid body parameters were compared against the motion from the BXV data to investigate the effect of STA on shoulder kinematics. It was found that the SVD method provided a better estimation of the rigid body parameters than that obtained from the PD method. This was due to the presence of the additional rotation from the PD of shear. In order to improve the accuracy of the PD method, a mathematical model describing the shearing and scaling was used in the calculations. The modified PD method showed improved reduction in STA for some tasks, and these improvements were comparable to the SVD method in terms of accuracy.