Abstract
Background
The accurate positioning of the total knee arthroplasty affects the survival of the implants(1). Alignment of the femoral component in relation to the native knee is best determined using pre- and post-operative 3D-CT reconstruction(2). Currently, the scans are visualised on separate displays. There is a high inter- and intra-observer variability in measurements of implant rotation and translation(3). Correct alignment is required to allow a direct comparison of the pre- and post-operative surfaces. This is prevented by the presence of the prostheses, the bone shape alteration around the implant, associated metal artefacts, and possibly a segmentation noise.
Aim
Create a novel method to automatically register pre- and post-operative femora for the direct comparison of the implant and the native bone.
Methods
The concept is to use post-operative femoral shaft segments free of metal noise and of surgical alteration for alignment with the pre-operative scan. It involves three steps. Firstly, using principal component analysis, the femoral shafts are re-oriented to match the X axis. Secondly, variants of the post-operative scan are created by subtracting 1mm increments from the distal femoral end (Fig1). Thirdly, an iterative closest point algorithm is applied to align the variants with the pre-operative scan.
For exploratory validation, this algorithm was applied to a mesh representing the distal half of a 3D scanned femur. The mesh of a prosthesis was blended with the femur to create a post-operative model. To simulate a realistic environment, segmentation and metal artefact noise were added. For segmentation noise, each femoral vertex was translated randomly within +−1mm,+−2mm,+−3mm along its normal vector. To create metal artefact random noise was added within 50 mm of the implant points in the planes orthogonal to the shaft. The alignment error was considered as the average distance between corresponding points which are identical in pre- and post-operative femora.
Results
Figure 2 shows, that when the implant zone is completely ignored, the error reaches a minimum plateau to below 1mm level. Different levels of segmentation noise had low impact on error value.
Conclusions
These preliminary results obtained within a simulated environment show that by using only the native parts of the femur, the algorithm was able to automatically register the pre- and post-operative scans even in presence of the implant. Its application will allow visualisation of the scans on the same display for the direct comparison of the perioperative scans.
This method requires further validation with more realistic noise models and with patient data. Future studies will have to determine if correct alignment has any effect on inter- and intra-observer variability.
For figures, please contact authors directly.
