Mass Personalisation of Preoperative Planning for Total Knee Arthroplasty. Evaluation of Statistical Models to Predict Patient-Specific Shape Information

Total knee arthroplasty (TKA) is considered the standard of care to alleviate pain and restore joint function in subjects with severe osteoarthritis (OA). The number of TKA procedures is expected to grow rapidly in the coming years, targeting an increasingly younger and more demanding patient population. At the same time, government expenditure on health is continuously under pressure to keep public healthcare sustainable. It has been shown that the obtained alignment of the joint can influence the postoperative outcome and patient satisfaction. Three-dimensional (3D) preoperative planning combined with patient-specific instrumentation (PSI) enables the surgeon to prepare for the procedure, improving surgical accuracy, efficiency, and postsurgical outcome. Further incentivised by Getting It Right First Time initiatives to reduce the rate of dissatisfactory postoperative outcome and the additional cost of revision surgery, the use of preoperative planning has grown substantially in the last decade. Nevertheless, obtaining the required patient-specific information remains a major barrier that prevents the adoption of personalised preoperative planning on a large scale. In this work, we aimed to address a number of these barriers using statistical shape models (SSM). An SSM describes the anatomical shape variation in a population using a low-dimensional set of parameters. We hypothesise that these models can be leveraged to derive, reconstruct and augment patient-specific information to facilitate the adoption of mass personalisation of preoperative planning and instrumentation for TKA. Currently, the preoperative planning workflow still involves substantial manual work to derive the required patient-specific data from the medical images. Not only does this increase the cost and lead time of a preoperative plan, but it also requires the availability of trained personnel to process the vast amount of procedures executed worldwide. While much attention has been given to the extraction of 3D models of the anatomy, i.e. image segmentation, few have investigated the automation of anatomical landmark annotation. These anatomical reference points are required to derive the relevant reference axes which determine the orientation and placement of the implant components. We demonstrated that shape models can robustly derive these reference axes with an accuracy similar to the existing observer variability in the current manual process. The creation of a preoperative plan with accompanying PSI requires the acquisition of 3D images of the patient. Current options include computed tomography (CT) or magnetic resonance (MR) imaging. While CT imaging is cheaper, easier to process and more accessible, MR imaging is often preferred due to the lack of ionising radiation. The main contributor to the effective dose of CT imaging for TKA is the irradiation of the radiosensitive tissue near the hip. Imaging of the hip is required to determine the mechanical axis of the femur, an important parameter when planning the procedure. In this thesis, we demonstrated that shape models can accurately reconstruct the coronal orientation of this axis based on the shape of the distal femur and information derived from the topogram, a low dose projection image. Imaging of the hip is therefore avoided, which substantially lowers the effective radiation dose delivered to the patient. Another advantage of MR imaging is the visibility of cartilage tissue, which allows this structure to be segmented to assess the damage caused by OA. This information is required to support a kinematic alignment approach to TKA, which recently gained popularity due to positive results regarding postoperative patient satisfaction. Furthermore, the availability of the cartilage tissue enables more stable designs of PSI that support on the articulating surfaces. We investigated the use of SSMs to augment the available CT data with information on the patient's tibiofemoral cartilage. Accounting for the geometry and relative position of both bones in the tibiofemoral joint, the SSM predicted the cartilage with significantly higher accuracy compared with the application of a population average cartilage distribution. However, the obtained surface error is too large to confidently assess the patient-specific cartilage damage. Therefore, further research is required before CT can fully compete with MR imaging to design PSI and to plan a kinematic alignment in total knee replacement. These applications were realised with effective general-purpose tools to build and apply SSMs. A pairwise non-rigid warping method was developed based on an iterative elastic deformation algorithm. The proposed warping method outperformed the original implementation and performed similarly to the state-of-the-art method optimised for shapes of cylindrical topology. The developed iterative, mesh-based fitting approach efficiently dealt with partial data and was employed in each of the three applications. In this work, three successful proofs of concept were presented of how SSMs can derive, reconstruct and augment patient-specific information to address current barriers that prevent mass personalisation of preoperative planning for TKA. However, we did reach the limits of these population-based models to represent and predict patient-specific information. These limits can be overcome using complementary technologies such as deep learning to develop robust, yet accurate algorithms to support more extensive patient-specific workflows and as such facilitate the road towards mass personalisation of knee replacement surgery.

Publication year:2021

Accessibility:Open