Background
Dissatisfaction amongst total knee arthroplasty (TKA) is the result of a complex relationship between the patient anatomy, prosthesis design and position, and other patient-specific factors. Prosthesis malalignment has been linked to poor patient outcomes in which coronal and axial malalignment has been most closely studied [
1,
2]. To have confidence in the correlation between component alignment and outcome, the method used to determine component placement must be accurate and reliable.
Component alignment refers to the angular difference between the prosthetic components and patient-derived antero-posterior (AP), medio-lateral (ML), and superior-inferior (SI) anatomic axes. This measurement has traditionally been the focus of post-operative analysis in TKA due to the ease of measurement [
3‐
5]. Component placement refers to the translational movement of the prosthetic components along these patient-specific axes. Due to difficulty in identifying the origin of these axes and accurately determining translation in space, component placement has been less well investigated. To understand the holistic effect of the TKA components on knee kinematics, both the alignment and placement must be taken into account. Here, we term the combination of component alignment and placement as ‘component position’.
The pre-operative state of the patient is a critical source of missing data from most analyses which prevents accurate reporting of component position. Bony resections cannot be accurately determined from a post-op analysis alone, and as a result, there is very little data available on the outcome of TKA as a result of the modification of the anatomy [
6,
7], highlighting the need for improved post-operative analysis techniques. Nevertheless, studies have investigated a range of movement and maximum flexion as a function of the posterior condylar offset (PCO) [
8‐
10]. In these publications, a greater PCO resulted in higher maximum flexion due to reduced steric hindrance. Pre- and post-operative measurements however were limited by the use of ML X-rays, indicating that the relationship must be strong to overcome such errors.
Alteration of the joint line and flexion/extension gaps is associated with a change in joint kinematics [
11] and patient outcome [
12]. In these studies, patients with less change to the coronal joint line reported improved WOMAC and Knee Society Clinical Rating Scores. Identification of such changes however can be difficult, as the joint line and joint gaps can be modified without affecting the appearance of the component alignment [
5]. To better understand the effect of bone resections and joint line and gap modification, accurate pre-operative geometry data is required. Similarly, Bengs and Scott [
13] found that increasing the patella button thickness without increasing the patella resection decreased the maximum passive flexion. Identification of appropriate patella resection for a given button thickness would not be possible with traditional post-operative analysis techniques.
Traditional methods of assessing TKA component alignment, including short leg X-rays [
14‐
16], long leg X-rays [
17], and post-operative 3D imaging only [
18‐
22], have been shown to suffer inaccuracies from anatomic variability and projection errors and difficulty in identifying patient-specific landmarks from the post-operative imaging. To improve landmarking and component placement accuracy, a pre-operative CT is required. Fortunately, CT imaging is rapidly becoming a standard of care in pre-operative planning for TKA [
23] and is available for a wide range of patients. Pre-operative CT imaging allows a volumetric registration of the pre-operative and post-operative bones and component geometries in 3D space eliminating any anatomic assumptions and projection errors. The models can then be used to determine bony resections and component placement. A method to compare the pre-operative state of the knee to the post-operative component position and bone resections, in which accuracy has not been affected by component flare, has not yet been achieved.
Here, we introduce a method of 3D reconstruction which utilises both a pre-operative and post-operative CT scan to determine the post-operative component position in TKA. The method may be extended to any joint replacement and is termed here the Australian Universal Resection, Orientation, and Rotation Analysis (AURORA) protocol. Landmarks and bone models unaffected by component flare obtained from the pre-operative scan are transformed into the post-operative frame of reference. Component position as defined by the landmarked patient-specific axes and bony resections are reported. The reproducibility and reliability of this method are presented and compared to other post-operative analysis techniques.
Discussion
The maximum component alignment differences from the mean within this study are low compared to previous literature and provide a confidence interval up to tenfold narrower when compared to protocols in which individual CT slices were investigated [
18,
19,
22,
28], or only post-operative CT scans were available [
29,
30]. The maximum error of < 1° is similar to the protocols using more advanced techniques, such a computational edge detection; however, these studies did not include ICC coefficients, so an assessment of the repeatability was not possible [
31]. The highest deviation from the mean was the tibial IE rotation at 0.9° and 0.7° for the two cases, with a confidence interval of 0.6° and 0.4°, respectively. These values represent an eightfold improvement in accuracy compared to previous attempts to measure tibial rotation [
32]. Previous attempts have reported difficulty in measuring tibial IE rotation due to the variability in the landmarks required to define a useful axis [
33]. By combining the pre-op and post-op CT, the landmarks that define the AP axis can be identified more easily than using post-op CTs alone. Although there may still be some debate over which landmarks are the most appropriate, this method allows points to be defined that accurately reproduce an anatomic axis across multiple subjects. The origin of all axes may be redefined based on future literature if needed.
The resulting resection level measures of the femur and tibia also show high reproducibility, with the highest deviation seen for the medial tibial plateau resection at 0.5 mm and 0.3 mm between the two cases and confidence intervals of 0.6° and 0.4°, respectively. The magnitude of the error here, however, is only slightly above the other resections, indicating that there may not be a systematic reason for reduced accuracy when placing this component. Previous attempts have been made to investigate the effect on TKA outcome arising from resection levels. These studies have mainly focussed on the femur, particularly the posterior condylar offset [
8,
34,
35]. These techniques however have primarily relied upon fluoroscopic images and planar X-rays which were discussed previously to be inaccurate, limiting the reliability of such studies.
Across both femoral and tibial component alignments and bony resections, this 3D pre-operative registration process shows excelled reliability, in which all ICC values report greater than 0.93. The lowest reported ICC value of 0.93, resulting from the femoral axial IE rotation measure, is primarily due to the difficulty of post-operative registration of the femur component. The posterior condyles, which dominate the axial rotation positioning, of the APEX implant used in this study are thicker than the distal condyles (11 mm vs 9 mm) and tibial tray (~ 3 mm). As such, the CT flare is greater in these regions, reducing the accuracy of the registration. The ICC values reported here are consistently similar to or higher than other post-operative analysis techniques [
7,
22,
36] indicating this method is not only accurate but suitable for routine post-processing by multiple users.
The high reproducibility and reliability of calculating both component alignment and bony resections performed by surgeons can lead to a better understanding of the influences of component alignment and component placement. The current literature has thoroughly reviewed the influence of component malalignment and poor patient outcomes [
37‐
39]. Missing from all of these analyses, however, is an understanding of the patient’s pre-operative anatomy, leading Hadi et al. [
37] to conclude that there is a dubious link between component malalignment and patient outcomes. From this post-operative analysis, we can begin to determine how the bony resections and the combination of component placement and alignment influence the outcome on a patient-specific level in greater detail. For example, the use of reliable bone resection measures from pre-operative bones may provide insight into the change of a patient’s soft tissue profile post-surgery. From the pre-operative CT scans, comparative ligament lengths and change in length resulting from component alignment and placement can be investigated from landmarked attachment sites. CT scans in this analysis, however, are performed in a non-functional supine position, such that the distance between ligament attachment sites may not be representative of the functional length of the ligament. Functional imaging may be introduced to this workflow in the future to this issue without a change in post-processing techniques.
The proposed 3D registration process for post-operative analysis involves additional pre-operative CT imaging compared to other processes [
6,
18]. Though this increases X-ray exposure to the patient, pre-operative planning, generally requiring a CT scan, is becoming the standard of care for TKA [
23], such that the pre-operative scans are not for post-operative analysis alone. The protocol used here is a low-dose CT, with radiation exposure less than the typical yearly background radiation and similar to protocols currently in use [
6]. All patient movement identified in pre-operative scans occurred in the mid-femur and mid-tibia regions, indicating that protocols which did not include the mid-femur and tibia sections would report inaccurate component placement. The resulting error in component position if these scans were used is the subject of further study.
Manual translation and rotation of the pre-operative bones and component geometries into the post-operative CT scan is reasonably labour intensive, requiring on average 60 min to complete, before the registration is quality control checked by a second engineer with additional experience. Further refinement of the proposed post-operative analysis process could include the use of an automated registration method. A preliminary automated registration process using the iterative closest point (ICP) method [
30] was performed on these cases. The registration time was observed to reduce to approximately 2 min, from which the results were then fine-tuned by one engineer and quality control checked by a second engineer, representing a 30-fold decrease in time. Further development of the ICP method to optimise parameters around fitting regions of interest, reliability, and time for analysis may allow accurate post-operative analysis to be part of routine care and is the subject of future studies.
Joint infection and component loosening are a cause of dissatisfaction and revision surgery. Joint infection can be identified by swelling of the joint and pathology reports; however, these are not always conclusive. Combining component position as determined using the AURORA protocol with SPECT imaging could identify bone metabolism associated with infection or component movement [
40]. Although current methods integrating SPECT imaging with CT do not improve the accuracy of determining component placement, such methods may be used to augment a pre-operative and post-operative CT 3D reconstruction to add metabolic activity.
The proposed post-operative 3D registration method described here has some limitations. The current time taken for this analysis as mentioned is approximately 60 min; this represents a high engineering burden and must be reduced to improve use in routine analysis. Commercially, TKA component geometry varies between medical device manufacturers, forming a significant part of their IP portfolio, as such, the component geometries must be obtained from the implant companies, which may be difficult—limiting the generalisability of this technique to engineering firms with a close relationship with implant companies. The reproducibility analysis performed here utilises two cases processed at multiple time points by multiple engineers of equal training. To better understand the reproducibility, particularly when processing outlier or severely pathological anatomy, a greater number of cases should be analysed.
Other methods to assess component position such as bi-planar X-rays followed by 2D to 3D registration offer a number of advantages over a CT, such as providing long leg assessments in a functional state [
41]. Such techniques, however, may require fluoroscopic agents [
42] and may only capture the region around the knee and are performed on apparatus less widely available than a traditional X-ray or CT machines, limiting its use [
31].