Tibio-femoral kinematics of the healthy knee joint throughout complete cycles of gait activities

Accurate assessment of 3D tibio-femoral kinematics is essential for understanding knee joint functionality, but also provides a basis for assessing joint pathologies and the efficacy of musculoskeletal interventions. Until now, however, the assessment of functional kinematics in healthy knees has been mostly restricted to the loaded stance phase of gait, and level walking only, but the most critical conditions for the surrounding soft tissues are known to occur during high-flexion activities. This study aimed to determine the ranges of tibio-femoral rotation and condylar translation as well as provide evidence on the location of the centre of rotation during multiple complete cycles of different gait activities. Based on radiographic images captured using moving fluoroscopy in ten healthy subjects during multiple cycles of level walking, downhill walking and stair descent, 3D femoral and tibial poses were reconstructed to provide a comprehensive description of tibio-femoral kinematics. Despite a significant increase in joint flexion, the condylar antero-posterior range of motion remained comparable across all activities, with mean translations of 6.3–8.3 mm and 7.3–9.3 mm for the medial and lateral condyles respectively. Only the swing phase of level walking and stair descent exhibited a significantly greater range of motion for the lateral over the medial compartment. Although intra-subject variability was low, considerable differences in joint kinematics were observed between subjects. The observed subject-specific movement patterns indicate that accurate assessment of individual preoperative kinematics together with individual implant selection and/or surgical implantation decisions might be necessary before further improvement to joint replacement outcome can be achieved. 2020 The Authors. Published by Elsevier Ltd. This is an open access article under theCCBY license (http:// creativecommons.org/licenses/by/4.0/).


Introduction
Kinematics of the human knee are guided by an interconnected system of bones, soft tissue structures and muscles acting around the joint. The resulting joint function in healthy knees is known to exhibit motion in all six degrees of freedom, presenting a complex set of translations and rotations (Andriacchi et al., 1998). Accurate knowledge of tibio-femoral movement can hence provide critical information for assessing knee joint functionality (Akbarshahi et al., 2010), but also lay the foundations for under-standing joint pathologies and musculoskeletal interventions. Total knee arthroplasty (TKA) or ligament reconstruction surgeries generally aim to restore pain-free functionality to the joint, based on physiological kinematics. However, the motion patterns of the native tibio-femoral joint itself are still discussed controversially (Dennis et al., 2001;Gray et al., 2019;Komistek et al., 2003;Koo and Andriacchi, 2008;Kozanek et al., 2009). Nevertheless, novel implant designs have been introduced that aim to mimic healthy joint kinematics, even though a comprehensive understanding of tibio-femoral kinematics during different gait activities remains missing.
While early studies used cadaveric material to examine tibiofemoral motion, optical motion capture using skin-mounted markers has become commonplace for assessing dynamic joint function. However, these measurements are known to be critically affected https://doi.org/10.1016/j.jbiomech.2020.109915 0021-9290/Ó 2020 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). by soft tissue artefact (STA) (Andriacchi et al., 1998;Cappozzo et al., 2005;Taylor et al., 2005). Errors associated with STA can be avoided using pins screwed directly into the subjects' bones (Lafortune et al., 1992;Reinschmidt et al., 1997), but such procedures are highly invasive and therefore limited to small cohort studies. To directly assess tibio-femoral kinematics in a less invasive manner, single-plane (Dennis et al., 2001;Galvin et al., 2019;Grieco et al., 2016;List et al., 2017;Moewis et al., 2016;Moro-oka et al., 2008) and dual-plane (Anderst et al., 2009;Guan et al., 2016;Kozanek et al., 2009;Li et al., 2008) fluoroscopic analyses have provided valuable and accurate data. The predominant drawback of most systems, however, relates to the small size and static nature of the intensifier, hence restricting measurements to a limited field-of-view. Consequently, most studies have analysed quasi-static activities such as knee bending, squatting and sit-tostand or assessed only a portion of the gait cycle (Dennis et al., 2001;Galvin et al., 2019;Grieco et al., 2018;Komistek et al., 2003;Kono et al., 2018;Kozanek et al., 2009;Moro-oka et al., 2008;Yamaguchi et al., 2009). To overcome the limitations of static imaging set-ups, procedures to enhance the fluoroscopic field-ofview have been investigated (Li et al., 2008). Moreover, mobile fluoroscopic systems have emerged that allow tracking of the knee during dynamic movements (Yamokoski and Banks, 2011) and consecutive cycles of gait activities (Guan et al., 2016;List et al., 2017).
Due to their strong radiographic image contrast and associated ease of 2D/3D registration to fluoroscopic images, the kinematics of total knee replacements have been studied in detail (Banks and Hodge, 1996;Dennis et al., 2001;Grieco et al., 2018;Guan et al., 2017;Schütz et al., 2019aSchütz et al., , 2019b. The movement patterns of the natural knee have, however, proven far more difficult to access. While most available studies have evaluated the loaded stance phase of walking (Benoit et al., 2007;Dennis et al., 2001;Komistek et al., 2003;Koo and Andriacchi, 2008;Kozanek et al., 2009;Reinschmidt et al., 1997), only few studies have successfully examined tibio-femoral kinematics over a complete cycle of walking (Andriacchi et al., 1998;Gray et al., 2019;Lafortune et al., 1992;Li et al., 2008). While consistent findings were presented for joint flexion, with a first peak observed at contralateral toe-off and a maximum knee flexion shortly after toe-off (Andriacchi et al., 1998;Benoit et al., 2007;Farrokhi et al., 2012;Gray et al., 2019;Kozanek et al., 2009;Lafortune et al., 1992;Li et al., 2009;Li et al., 2008;Reinschmidt et al., 1997), contradictory patterns were reported for the other two rotations. Most studies observed an internally rotated tibia at heel-strike, followed by further internal rotation during early stance phase, reaching a first peak shortly after contralateral toe-off (Andriacchi et al., 1998;Farrokhi et al., 2012;Gray et al., 2019;Kozanek et al., 2009;Lafortune et al., 1992;Li et al., 2008;Reinschmidt et al., 1997). The tibia is then thought to rotate externally through mid-stance, before rotating internally once again prior to toe-off (Gray et al., 2019;Kozanek et al., 2009;Lafortune et al., 1992;Li et al., 2008;Reinschmidt et al., 1997). Contrary to this common understanding of knee motion, tibial external rotation during early stance phase was described in two studies (Benoit et al., 2007;Komistek et al., 2003), where tibial internal rotation during mid-stance was followed by external rotation from 66% of the stance phase until toe-off. In the few studies that were able to assess the unloaded swing phase, peak tibial internal rotation was found shortly after toe-off, followed by external rotation until just before heel-strike, with tibial internal rotation occurring in preparation for the following heel-strike (Andriacchi et al., 1998;Gray et al., 2019;Lafortune et al., 1992). Throughout the complete gait cycle, minimal changes in abduction/adduction were observed, but with clear subjectspecific differences (Andriacchi et al., 1998;Kozanek et al., 2009;Lafortune et al., 1992).
Only few studies have examined the antero-posterior (A-P) translation of the two femoral condyles relative to the tibia individually. Here, a larger translation has generally been observed for the lateral compared to the medial condyle during the loaded stance phase (Dennis et al., 2001;Komistek et al., 2003), as well as over complete gait cycles (Gray et al., 2019), indicating an overall medial centre of rotation (CoR). In contrast, Koo and Andriacchi (2008) observed more translation of the medial compared to the lateral condyle during the loaded stance phase of gait, which was later supported by Kozanek and co-workers (2009).
Beyond these reports, quantifying knee motion during more challenging activities of daily living such as downhill walking and stair descent are critical for understanding knee function. In the limited literature, no consistent trend of A-P translation of the medial versus the lateral condyle was reported for either of the two activities, and analysis of the unloaded swing phase remains missing (Dennis et al., 2001;Farrokhi et al., 2014;Komistek et al., 2003).
As a result, tibio-femoral movement patterns during functional gait activities are still discussed controversially. Importantly, comparison of condylar A-P translation is complicated by the method used for analysing the movement: the tibio-femoral contact path (Dennis et al., 2001;Gray et al., 2019;Komistek et al., 2003), geometric condylar centres (Gray et al., 2019;Kozanek et al., 2009), or points on the transepicondylar axis ) have all been used to describe condylar translation, plausibly also contributing to the inconsistent observations. However, it also remains unclear whether the observed variations resulted from differences in performing the activities, the cohorts studied, or simply because of variability in soft tissue structure or anatomy. Therefore, the aim of this study was to determine the ranges of tibio-femoral rotation and condylar translation during complete cycles of free level walking, downhill walking and stair descent, as well as provide evidence on the location and possible movement of the centre of joint rotation in the transverse plane in healthy knees.

Subjects
Ten subjects (5 female/5 male, aged 23.7 ± 3.2 years, BMI 21.7 ± 2.2 kg/m 2 ) provided written, informed consent to participate in this study, which was approved by the local ethics committee (KEK-ZH-Nr. 2016-00410). All subjects underwent a clinical knee examination (including testing of ligament condition) by an experienced knee surgeon, and subject-specific limb alignment (mean varus angle of 0.6 ± 1.4°) was evaluated using a biplanar X-ray system (EOS imaging, France) in a standing position (Folinais et al., 2013). Subjects with significant problems at the lower extremities or exhibiting a limb alignment >3°varus/valgus were excluded from this study.

Measurement Set-Up
Tibio-femoral kinematics were assessed using the ETH moving fluoroscope (List et al., 2017), synchronised with a 22 camera optical motion capture system (Vicon MX System, Oxfords Metrics Group, UK) as well as five floor-mounted force plates (Kistler AG, Switzerland), while an additional two mobile force plates instrumented the ramp and stairs respectively (Fig. 1). The subjects performed all motion tasks in a pair of self-selected sports shoes.

Motion tasks
After a standing trial and activity-specific familiarisation trials, all subjects were measured throughout at least five valid cycles each of straight level walking, downhill walking (10°declined slope), and stair descent (three steps, each 18 cm in height) -in this study collectively considered to be gait activities. Trials were considered valid when the knee remained within the field-of-view of the image intensifier throughout the stance as well as the swing phase, and all force plates were hit correctly. The ground reaction forces were used to determine heel-strike and toe-off with a threshold of 25 N. For downhill walking, the second heel-strike, marking the end of swing phase, was defined based on the trajectory of a heel marker. Overall, for each participant the mean dose area product (DAP) was 7.9 ± 1.2 Gycm 2 (diameter image intensifier 30.5 cm) leading to a total average effective dose of 0.11 ± 0.02 mSv for the fluoroscopic measurement.

3D tibio-femoral kinematics
Subject-specific volumetric models were obtained based on a CT scan of each subject's knee (~20 cm proximal/distal of the joint line, resolution 0.5 Â 0.5 mm, slice thickness 1 mm), followed by manual segmentation of the data using the open-source software MITK-GEM (Pauchard et al., 2016). All fluoroscopic images were distortion corrected and the optical projection parameters (focal distance, principal point) were determined (Foresti, 2009). The 3D position and orientation of the femur and tibia were evaluated using a semi-automatic 2D/3D registration software , with reported mean absolute errors of <1°for rotations around all three axes, <0.6 mm for in-plane, and <7.1 mm for out-of-plane translations using artificial bones with radiopaque image properties.
Local femoral and tibial anatomical coordinate systems were established to calculate relative tibio-femoral rotations ( Fig. 2A and B) (Grood and Suntay, 1983). The functional flexion axis (FFA) was defined based on kinematics derived from the fluoroscopic imaging data of a maximal deep knee bending trial using the symmetrical axis of rotation approach (Ehrig et al., 2007). Condylar A-P location was calculated based on a medial (FFA_P med ) and lateral (FFA_P lat ) point on the functional flexion axis of the femur (Fig. 2C). Translation of each femoral condyle was then described as the movement of these points with respect to the mid-coronal plane of the tibia. All A-P translations were scaled using a subject-specific coefficient, scaled to a ratio of the mean condylar width divided by the subject's condylar width (Gray et al., 2019). To allow comparison and interpretation throughout complete gait cycles, tibio-femoral kinematic data was linearly interpolated to 101 data points. To allow comparison of joint kinematics against other studies, tibio-femoral translation is additionally presented using the nearest point (N_P) approach (same normalisation and interpolation) on the surface of each femoral condyle (N_P med and N_P lat ) relative to a plane parallel to the tibial articular surface, as described by Asano and co-workers (2001).
Finally, the CoR in the transverse plane was calculated using the symmetrical centre of rotation estimation based on the locations of FFA_P med and FFA_P lat (Ehrig et al., 2006).

Statistics
A total of four mixed-model analysis of variances (ANOVAs) with subject as random effect were performed. Three mixedmodel ANOVAs were performed to investigate the influence of the task and gait-phase on the ranges of tibio-femoral rotation using the ranges of rotation (flexion/extension, ab/adduction, tibial internal/external) as the dependent variables, with task (three Fig. 1. Measurement set-up during level walking. Radiographic images of the knee were captured using a moving fluoroscope, imaging at 25 Hz (1 ms shutter time) with an image resolution of 1000 Â 1000 pixels (List et al., 2017;Foresti, 2009). The moving fluoroscope was equipped with five optical markers on the x-ray source and six on the image intensifier to track its position (List et al., 2017). Fifty-five skin markers, placed at specific anatomical locations (List et al., 2013), were tracked at 100 Hz using a 22 camera system (Vicon MX system, Oxford Metrics Group, UK). Five force plates embedded in the floor (decoupled from surrounding ground) and two mobile force plates mounted on the ramp and stairs (Kistler Instrumentation, Switzerland) were used to measure undisturbed ground reaction forces at a sampling frequency of 2000 Hz. levels: level walking, downhill walking, stair descent) and gaitphase (two levels: stance phase, swing phase) as independent variables. One ANOVA was applied to test the effect of task, side, and gait-phase on the range of A-P translation (FFA_P approach). Here, the range of A-P translation was set as the dependent variable, while task (three levels: level walking, downhill walking, stair descent), gait-phase (two levels: stance phase, swing phase) and side (two levels: medial A-P, lateral A-P) were set as the independent variables. The least significant differences approach was used for post-hoc comparisons and the levels of significance were corrected for multiple comparisons, starting at a = 0.05, using Bonferroni correction. No statistical tests were performed for the A-P translation of the N_P approach. Statistics were conducted in SPSS (SPSS 24, IBM, USA).

Results
Mean average velocities over the complete gait cycle were 0.8 7 ± 0.06 m/s during level walking, 0.81 ± 0.07 m/s during downhill walking, and 0.58 ± 0.05 m/s for stair descent.
As expected, a significant increase in range of joint flexion was observed from level walking to downhill walking and stair descent (Table 1A). For all three activities, the knee joint was slightly flexed at heel-strike, followed by an activity-specific flexion/extension behaviour, with peak knee flexion occurring shortly after toe-off (level walking: 62.8 ± 5.5°; downhill walking: 70.1 ± 4.7°; stair descent 95.0 ± 5.7°) (Fig. 3).
After heel-strike, the tibia generally rotated internally during the stance phase, reaching a maximum internal rotation shortly before toe-off (level walking: À9.9 ± 6.9°, downhill walking: À10.6 ± 7.7°, stair descent: À12.3 ± 7.6°) followed by an external rotation in preparation for the subsequent heel-strike (Fig. 3). Initial tibial internal/external rotation at heel-strike was highly subject-specific, with individual values ranging from À14.5°internal to 11.1°external rotation between the three activities. However, each subject exhibited consistent internal/external rotation characteristics, with a mean standard deviation (SD) of 1.5-3.4°across all subjects.
Repeatable ab/adduction was found for the individual subjects (mean subject SDs: 0.9-1.8°), but these were highly individual, and no clear ab/adduction patterns could be established over a complete gait cycle for any of the activities (Supplementary material: Fig. S1).

Condylar translation
During standing, the mean FFA_P med (À10.6 ± 7.0 mm) and FFA_P lat (À8.7 ± 5.7 mm) over all subjects were located posterior to the tibial mid-coronal plane.  (Ehrig et al., 2007). The origin was defined as the midpoint between the most medial and lateral point of the femur on the FFA. The medio-lateral axis (Z Femur ), was oriented along the FFA, pointing medially. A cylinder (depicted as multiple circles) was then fitted to the most proximal 10 cm of the femoral shaft, defining the shaft axis. The antero-posterior axis (X Femur ) was orthogonal to Z Femur and the shaft axis. In right knees, X Femur pointed posterior, while in left knees it was directed anteriorly. The longitudinal axis (Y Femur ) was orthogonal to both X Femur and Z Femur to complete the right-handed coordinate system. (B) Tibial anatomical coordinate system pictured in a left tibia. A plane parallel to the posterior tibial slope was constructed, resecting the condylar cartilage. Two circles were manually fitted in the plane, such that the maximum possible circle perimeter fitted to the tibial cortex. The midpoint between the two circles' origins was set as the origin of the tibial anatomical coordinate system. Similar to the femur, a cylinder was fitted to the most distal 10 cm of the tibial shaft (depicted as multiple circles). However, for the tibia, the longitudinal axis (Y Tibia ) was consistent with the orientation of the shaft axis, but passed through the origin. The antero-posterior axis (X Tibia ) was orthogonal to Y Tibia and the line connecting the circle centres. Similar to the femur, the X Tibia pointed posteriorly in right knees, but anteriorly in left knees. The medio-lateral axis (Z Tibia ) was orthogonal to the X Tibia and Y Tibia to complete the right-handed coordinate system. (C) The nearest points (N_P med & N_P lat ) between the femoral condyles and the tibial plane were calculated as the weighted means of the ten nearest points (mean point separation 0.4 mm) for the medial and the lateral condyles. Two points on the functional flexion axis (FFA_P med & FFA_P lat ) were then defined based on the projection of N_P med & N_P lat during standing onto the FFA. These points were used to describe condylar antero-posterior (A-P) translation relative to the tibia.
Following a short posterior translation after heel-strike, FFA_P med moved anteriorly, peaking shortly before toe-off for both level and downhill walking. This motion was followed by a short posterior and again anterior translation, reaching a second peak before returning to its initial position at heel-strike (Fig. 4). For all three activities, a significant difference was found for the amount of A-P translation of the FFA_P med between the stance and the swing phases (Table 1B). The FFA_P lat moved posteriorly after heel-strike, reaching a minimum around toe-off. Extension of the knee during the unloaded swing phase was then associated with an anterior translation of the FFA_P lat for all three activities and all subjects (Fig. 4). Overall, FFA_P med and FFA_P lat showed Fig. 3. Tibio-femoral flexion/extension (flex/ex) and tibial internal/external rotations (tib int/ext) throughout complete cycles of level walking (top), downhill walking (middle) and stair descent (bottom). The mean across all subjects (black line), standard deviations across all subjects (grey shading), as well as means for every subject (dark grey lines) are presented. The average instance of toe-off across all subjects is indicated as a vertical dashed line.

Table 1
Range of motion for (A) all three tibio-femoral rotations. Flex/ex: flexion/extension, int/ext: tibial internal/external rotation, ab/add: abduction/adduction (B) Antero-posterior (A-P) translation for FFA_P med and FFA_P lat (C) A-P translation for N_P med and N_P lat . Mean and standard deviations across all 10 subjects are presented for the loaded stance and unloaded swing phases of level walking, downhill walking and stair descent.

RoM [°]
Rotation during the loaded stance phase comparable ranges of A-P translation, with only the swing phase of level walking and stair descent showing significant differences (Table 1B). Intra-subject repeatability was high for both FFA_P med and FFA_P lat A-P translation, with overall mean SDs of only 0.8-2.0 mm across all repetitions of each different task (Figs. S2-S4). For comparison, the N_Ps were located anteriorly (N_P med : 4.7 ± 5.4 mm; N_P lat : 11.3 ± 9.7 mm) to the mid-coronal plane while standing. As opposed to the FFA_P, a considerable difference between the A-P translations of N_P med compared to N_P lat was found for both phases of all three activities (Table 1C). While N_P med showed a consistent pattern of A-P translation for all subjects over all activities, high inter-subject variability was observed for N_P lat , especially during mid-stance of level walking and shortly before heel-strike in all activities (Fig. 4). However, the mean intra-subject variability (1.2-3.0 mm) was comparable to both FFA points throughout the entire gait cycles (Figs. S2-S4).

CoR in the transverse plane
Over all subjects, a slightly medial mean CoR was observed for the entire gait cycle of all activities with the swing phase (level walking: 13.7 ± 14.4 mm; downhill walking: 11.7 ± 9.0 mm; stair descent: 14. 6 ± 9.8 mm, all medial) nearly always presenting a more medial CoR than the stance phase (level walking: 7.2 ± 7.4 mm; downhill walking: 6.7 ± 8.8 mm; stair descent: 9.4 ± 14.3 mm, all medial) (Fig. 5, Tables S1-S3). Intra-subject variability of the location of the FFA_P med and FFA_P lat within a task was high (Figs. S5-S34), and resulted in SD's ranging from 0.9 to 23.3 mm for the mean location of the subject-specific CoR over a gait phase (Tables S1-S3).

Discussion
A comprehensive understanding of human knee joint kinematics is critical for guiding the design of total knee replacement com-ponents that aim to mimic healthy joint motion. Until now, movement of the tibio-femoral joint has been discussed controversially, but only rarely analysed over multiple cycles of different activities. For the first time in healthy knees, moving fluoroscopy has been used in this study to determine the in vivo tibiofemoral rotations and translations that occur throughout complete cycles of downhill walking and stair descent, in addition to free level walking, using the same subject cohort as well as the same measurement set-up. Only a limited task-dependency was observed for ranges of A-P translation, despite clear differences in knee flexion angles (Table 1). Overall, the data have revealed a predominantly central to medial CoR in the transverse plane, even though it is located laterally in some individuals and during some phases of different activity cycles.
The flexion/extension patterns during level walking found in this study were consistent with those reported in the literature (Gray et al., 2019;Lafortune et al., 1992). Furthermore, our tibiofemoral internal-external rotation patterns were comparable to the limited evidence available on complete gait cycles during level walking (Gray et al., 2019), even though the absolute values differed. This variation likely originates from differences in the definition of the femoral and tibial anatomical coordinate systems (Gray et al., 2019;Lafortune et al., 1992). In our study, similar ranges of A-P translation were observed for both compartments (using FFA_P med and FFA_P lat ), but with larger A-P translations on the lateral side of the joint for the unloaded swing phase of level walking and stair descent. Such data is consistent with previous studies that suggested a medial pivot during unloaded flexion (Iwaki et al., 2000;Pinskerova et al., 2004). The magnitude of the overall translations of the medial compartment (FFA_P med ) were comparable to values presented in the literature, whereas for the lateral compartment (FFA_P lat ), contradictory results were found (Gray et al., 2019;Kozanek et al., 2009). As the subject-specific FFA was optimised over the full range of motion during deep knee bending, the true instantaneous flexion axis at a specific instant of time might vary from the mean FFA, possibly introducing an A-P translation error around the flexion axis. Further investigation is clearly required in order to understand the influence of such secondary errors, including their role relative to different activities and flexion axis approaches.
To allow cross-study comparison, we have also presented condylar kinematics using the nearest point approach. While the range of A-P translation for some subjects was in line with the literature (Gray et al., 2019), others exhibited a substantially larger range of A-P translation. An assessment of the femoral geometries revealed a relatively flat region on the lateral condyle (similar to previous observations in cadaveric studies (Iwaki et al., 2000)) for some subjects, plausibly resulting in a ''jump" of the N_P lat during phases of knee extension (Fig. 4, right). As a result, it seems that this method is sensitive to local contact conditions, but does not necessarily describe the relative movement between the tibia and the femur. Additional research into the influence of subjectspecific anatomy on the translation of the contact points is thus needed before a thorough understanding of the large variability in A-P translation between subjects can be gained.
The mean CoR was located on the medial tibial plateau for all three activities. Nevertheless, the location of trial specific mean CoRs varied among subjects and tasks (Tables S1-S3). These results are generally in line with previous findings (Dennis et al., 2001;Gray et al., 2019), although a mean lateral CoR has also been presented (Koo and Andriacchi, 2008;Kozanek et al., 2009). Calculation of the mean CoR over the complete gait cycle seems to be more robust than for either the stance or swing phase individually, resulting in SDs of less than 5 mm for most of the subjects, in line with previous research (Dennis et al., 2001). Furthermore, individual movement patterns were clearly observed, thus emphasising the need to analyse multiple trials for individual subjects.
Overall, the majority of studies fail to capture the unloaded stance phase, which, based on the results of our study, differs from the loaded stance phase. Analysis of this phase of gait provides clear insights into the largest joint rotations that are critical for understanding soft tissue loading as well as possible overloading (Hosseini Nasab et al., 2019, 2020. Accurate assessment of complete cycles of daily activities could therefore be key for planning subject-specific component implantation to achieve pain-free post-operative outcome. Walking with the moving fluoroscope potentially influenced the natural gait patterns of our subjects in the present study, especially as walking speeds are known to be reduced compared to free level walking (Stacoff et al., 2005). However, the changes in gait parameters have been shown to be small, and gait patterns remain comparable to those exhibited during slow walking (Hitz et al., 2018). While kinematics captured in this study may not be completely natural, they still represent the state-of-the-art in understanding joint motion over complete cycles and for different daily activities. The predominant drawback, however, is the subject's exposure to radiation, necessitating that data collection using such technologies is kept to a minimum.
Overall, comparison with the existing literature is challenging, mainly due to the variety of approaches used. While the geometric centre axis as well as the transepicondylar axis are widely used, their definition is highly dependent on the protocol and experience of the examiner. Within this study, we aimed to minimise the observers input to receive an independent axis that is only based on the subject's deep knee bending movement (using the fluoroscopic based reconstruction of the bone segments). Only the tibial coordinate system in our study was defined based on subject-specific anatomy, but it is indeed entirely plausible that individual differences could have added variation to the presented results. Although the current single-plane set-up in combination with the semi-automatic 2D/3D registration possesses high accuracy in the in-plane direction , a combination of errors for different segments could plausibly lead to larger total relative errors. The greatest source of error is likely to arise from the contralateral knee crossing through the image twice during each gait cycle, hence cumulatively limiting the accuracy of the 2D/3D registration at these instances. In addition, out-of-plane registration errors could play a role in the determination of the CoR in the transverse plane. To overcome this limitation as well as to improve out-of-plane accuracy, a dual-plane set-up could provide considerable advantages, however this would increase the overall radiation dosage applied to the subject.
This study provides the first analysis of knee kinematics during multiple complete cycles of free level walking, downhill walking and stair descent. The results suggest that the CoR is primarily located medially, but varies between subjects and gait phases. It therefore seems that the recent movement towards medial congruent implant designs  is warranted, but further investigation is clearly needed to elucidate the role of limb alignment and anatomical inter-subject variability on knee joint kinematics, particularly with regard to the loading and possible overloading of the surrounding soft tissues (Hosseini Nasab et al., 2019). Importantly, however, the subject-specific movement patterns observed in this study indicate that individual implant selection and/or surgical implantation decisions might be necessary before further improvement to joint replacement outcome can be achieved, possibly supported by the accurate assessment of each individual's pre-operative knee joint kinematics to guide clinical decision-making.