Non-invasive computer-assisted measurement of knee alignment

Abstract

The quantification of knee alignment is a routine part of orthopaedic practice and is important for monitoring disease progression, planning interventional strategies, and follow-up of patients. Currently available technologies such as radiographic measurements have a number of drawbacks. The aim of this study was to validate a potentially improved technique for measuring knee alignment under different conditions. An image-free navigation system was adapted for non-invasive use through the development of external infrared tracker mountings. Stability was assessed by comparing the variance (F-test) of repeated mechanical femoro-tibial (MFT) angle measurements for a volunteer and a leg model. MFT angles were then measured supine, standing and with varus-valgus stress in asymptomatic volunteers who each underwent two separate registrations and repeated measurements for each condition. The mean difference and 95% limits of agreement were used to assess intra-registration and inter-registration repeatability. For multiple registrations the range of measurements for the external mountings was 1° larger than for the rigid model with statistically similar variance (p = 0.34). Thirty volunteers were assessed (19 males, 11 females) with a mean age of 41 years (range: 20–65) and a mean BMI of 26 (range: 19–34). For intra-registration repeatability, consecutive coronal alignment readings agreed to almost ±1°, with up to ±0.5° loss of repeatability for coronal alignment measured before and after stress maneuvers, and a ±0.2° loss following stance trials. Sagittal alignment measurements were less repeatable overall by an approximate factor of two.

Inter-registration agreement limits for coronal and sagittal supine MFT angles were ±1.6° and ±2.3°, respectively. Varus and valgus stress measurements agreed to within ±1.3° and ±1.1°, respectively. Agreement limits for standing MFT angles were ±2.9° (coronal) and ±5.0° (sagittal), which may have reflected a variation in stance between measurements. The system provided repeatable, real-time measurements of coronal and sagittal knee alignment under a number of dynamic, real-time conditions, offering a potential alternative to radiographs.

• Knee alignment
• non-invasive
• infrared tracking
• femoral-tibial angle

Introduction

Knee joint alignment is an important parameter that has been extensively investigated in the context of osteoarthritis (OA). Radiographic and magnetic resonance imaging (MRI) studies have provided evidence that coronal malalignment is associated with an increased incidence [1] of tibiofemoral OA and risk of progression [2–5]. The importance of coronal implant orientation in reconstructive surgery of the knee has been widely accepted, with the recognition that mechanical alignment outside ±3° can lead to early prosthesis loosening [6], with reported failure rates of 67% for varus knee prostheses versus 29% for knee prostheses in a neutral position [7], together with increased polyethylene wear and poor overall function [8], [9]. Accurate measurement of knee alignment is therefore important for the monitoring of patients with OA, the subsequent planning of surgical interventions, and the assessment of treatment outcomes.

The standard measurement of knee alignment often relies on clinical evaluation in conjunction with radiographs that center on the knee joint. However, human assessment of angles is known to be poor [10], and the accuracy of alignment estimates under these circumstances may be no better than on the order of ±5° [11]. The use of knee radiographs has been found to be an inaccurate measure of mechanical lower limb alignment [12], so their role in assessing knee alignment for planning intervention strategies and for postoperative evaluation may be limited. Full-length hip-knee-ankle radiographs have therefore been increasingly adopted to provide more reliable pre- and postoperative information and are widely considered the gold standard for measuring knee alignment. In spite of enabling measurement of the mechanical femoro-tibial (MFT) angle, these radiographs are susceptible to limb positioning errors, with apparent variations in alignment being produced as a result of knee flexion or rotation [13], [14]. Computed tomography (CT) imaging can overcome these positional artefacts by providing a 3D evaluation of the lower limb anatomy, but is unable to provide weight-bearing information as subjects are required to be supine during the imaging procedure. Further drawbacks of both imaging modalities include limited availability, exposure of the pelvis to ionizing radiation, and the lack of more normal physiological control data from populations not typically exposed to them, such as children and non-arthritic subjects with knee ligament injuries.

Because of the limitations of radiographs and CT scans, several alternative clinical measures of alignment have been reported, and include techniques ranging from direct visual estimation to measurement adjuncts such as calipers, manual goniometers and plumb-line methods [15], [16]. These methods are inexpensive, avoid radiation exposure, and are relatively quick to perform while providing instant measurement results. However, the reported errors are potentially too large for use in planning and follow-up of surgical interventions such as replacement arthroplasty and corrective osteotomy, where higher levels of accuracy are often required [16].

Outside the clinic situation, a number of new technologies using infrared tracking have been introduced intraoperatively to provide surgeons with quantitative measurement tools that permit real-time assessment of lower limb kinematics [17–19]. These systems have high levels of precision and can achieve angular and tibiofemoral gap measurements within 1° or 1 mm, respectively [20], [21]. At present, these quantitative measurement techniques have restricted scope due to their reliance on the rigid bony fixation of trackers. Adapting this technology for non-invasive patient assessment is challenging due to the soft tissue artefacts associated with the external mounting of trackers. Previous investigations to quantify the relative movement of external marker sets relative to underlying bones have reported large potential errors and have questioned the value of these methods for accurate kinematic analysis [22], [23]. However, these functional methods of determining rotational joint centers and resultant mechanical lower limb alignment are often applied in the context of gait analysis or involve active joint movement with contraction of the underlying muscles. A more recent study [24] sought to minimize these potential artefacts by measuring static standing lower limb alignment with position capture and skin markers, along with external anatomical landmarks. The reliance on anthropometric measurements to predict joint center location may have accounted for the only moderate correlation with corresponding long-leg radiographs in an experimental set-up not readily adaptable to an out-patient clinic.

Given the subjective nature of clinical examination and the limitations of the various measurement techniques reported to date, there is potential to improve current methods of assessing knee joint alignment. This paper reports the validation of a non-invasive system for measuring lower limb alignment based on a commercially available infrared tracking technology with kinematic registration. Our hypothesis was that repeatable, real-time measurements of mechanical knee alignment under a number of conditions could be obtained in a clinical situation.

Materials and methods

Infrared tracking system

An image-free navigation system (OrthoPilot®, BBraun Aesculap, Tuttlingen, Germany) was chosen due to its being in current clinical use. The system consisted of an optical localizer, active infrared (IR) trackers, a pre-calibrated probe to digitize anatomical landmarks, and a foot pedal that enabled “hands-free” data recording. High tibial osteotomy (HTO) software (OrthoPilot® HTO version 1.5, BBraun Aesculap, Tuttlingen, Germany) was used for the kinematic determination of hip, knee and ankle centers and the resultant generation of coronal and sagittal MFT angles. Coronal alignment was defined with varus negative and valgus positive, whilst sagittal alignment was defined with hyperextension negative and flexion positive.

Rigid tracker mounting model

A metal lower limb model was designed and manufactured to provide optimum conditions for measuring knee alignment. This consisted of metal rods representing a femur, tibia and foot with rigidly attached tracker mounts and mechanical hip, knee and ankle joints with the required range of movement for registration of their rotational centers (Figure 1).

Figure 1. Leg model with rigid tracker mountings.

Non-invasive tracker mounting

Tracker mountings for the thigh, calf and mid-foot regions were developed using curved metal base plates and broad straps made from standard-strength elastic webbing (542, E&E Accessories, Kingston upon Thames, UK). A variety of lengths were made with a sequence of eyelets at either end to connect to the base plate and enable further adjustment of strap size (Figure 2). Straps were firmly applied as tightly as each subject could tolerate. Although the trackers could be deflected from their applied position by deliberate application of force, they then returned to this position once the deflecting force was removed. It was therefore expected that accidental contact with the trackers, as long as this did not occur at the point of data acquisition, would not affect system accuracy or repeatability. Care was taken throughout testing to avoid contact with the trackers or tensioning the connecting cables, which could also lead to tracker deflection.

Figure 2. External tracker mountings with curved base plates and adjustable elasticated straps. (a) Mounting of trackers on thigh and calf. (b) Foot tracker mounting.

Tracker stability testing

To quantify the soft tissue artefacts of the non-invasive mountings, the repeatability of the measurement of coronal knee alignment for both the leg model and the right lower limb of a slim, female volunteer was determined. The volunteer was asked to relax whilst lying supine on an examination couch to ensure that all movements were passive. The registration process which would be employed intraoperatively in the normal use of the software was then performed. The registration began with the identification of the kinematic center of the hip joint, which required a slow, controlled circumduction of the thigh. The maneuver was performed in this manner to avoid moving the pelvis and subsequently altering the location of the rotational center of the femoral head. Excessive movement of the pelvis or trackers could have resulted in a wider, “non-spherical” spread of acquired hip joint center (HJC) points that was outside the required precision of the system [25]. This would have resulted in rejection of the HJC acquisition and the instruction to repeat the circumduction maneuver until the spread of measured points was within the required threshold. The kinematic ankle center was determined next by attaching a tracker to the dorsum of the foot and then dorsi-flexing and plantar-flexing the ankle (Figure 2b). The rotational center of the knee joint was then acquired by flexing and extending the knee between 0 and 90° as well as rotating the tibia on the femur at 90° of flexion. Following a single registration, the trackers were left in position and 20 consecutive MFT angle recordings were made with the rigid leg model stationary and with the volunteer instructed to remain as still as possible. The full registration process was then repeated a further 20 times on 13 different days to quantify additional soft tissue artefacts associated with removal and re-attachment of the trackers. Statistical analysis was performed using SPSS version 17 (SPSS Inc., Chicago, IL) and F-tests were used for comparison of the variances of the repeated data sets

Repeatability testing

All experimental procedures were approved by the University Ethics Committee and, after giving informed consent, 30 volunteers were recruited (19 males and 11 females) with a mean age of 41 years (range: 20–65 years) and a mean body mass index (BMI) of 26 (range: 19–34). Participants confirmed no acute knee symptoms and no history of joint replacement. Basic demographic data were recorded prior to assessment of the right lower limb. Two kinematic registration processes were performed using the appropriate passive clinical maneuvers described above. After each registration, the immediate coronal and sagittal alignments in full extension were recorded with the lower limb supported at the heel and the subject having been told to relax. Following this, coronal and sagittal alignment was measured with subjects having been asked to assume their normal bipedal stance. Returning the participant to the supine position, the coronal and sagittal alignment measurements were then performed twice, and subsequent to this five manual stresses were applied to the knee joint by a single clinician to determine varus and valgus angular displacements. During these stress maneuvers, the knee was held at between 0° and 5° of flexion as indicated by the on-screen measurement of the sagittal MFT angle. If the knee could not extend to 0° the stress measurements were performed within a 5° window of flexion from the maximum extension angle. Finally, the coronal and sagittal alignment measurements were again repeated twice. Thus, five coronal and sagittal MFT angles were determined, before and after standing and before and twice after five bouts of varus-valgus stressing. The clinician was blinded to all the recorded alignment measurements except for the initial supine coronal MFT angle following registration. Occasionally, this measurement after the second registration did not agree to within 2° of the first registration, in which case the registration process was repeated. The limit of 2° was based on the acceptance of a small anticipated loss of accuracy due to soft tissue artefacts in comparison to the reported 1° accuracy for invasive use [21].

The mean difference and the Bland-Altman 95% limits of agreement [26] of supine coronal MFT angles taken consecutively, and before and after standing and collateral stress, within each trial were measured. This was used as an indirect measure of any intra-registration tracker movement that may have occurred during manipulation of the lower limb or as a result of the subject actively moving between supine and standing positions. The mean difference and 95% agreement limits were also used to assess inter-registration agreement of MFT angles measured supine, standing, and following applied collateral stress. Bland-Altman plots were generated for the inter-registration comparative data sets. When more than one measurement of a variable was taken within a trial the median value was used.

Results

Tracker stability

A comparison of the rigid and non-invasive mounts is shown in Table I. Consecutive readings of coronal alignment following a single registration demonstrated standard deviations of 0.07° and 0.13° for the rigid leg model and volunteer, respectively, and the variances were found to be statistically different (p < 0.01). For multiple registrations on different days the overall range was 1° larger for the non-invasive volunteer mounting, but the SD was still less than 1° for both tracker mounting methods with no statistically significant difference in the variance of the groups. Straps were well tolerated by the individual volunteer, easily allowing extended testing to be carried out over a 45-minute period.

Table I.  The mean and standard deviation (SD) of each set of tests was used to compare the difference in repeatability of the rigid leg model and the non-invasive tracker mounting on the volunteers.

	Single registration		Multiple registrations
	Leg model	Volunteer	Leg model	Volunteer
n	20	20	20	20
Mean (SD) (°)	2.1 (0.07)	1.4 (0.13)	1.6 (0.5)	1.5 (0.7)
Range (°)	2.0–2.3	1.1–1.6	0.9–2.8	0.3–2.5
F-test	p = 0.008		p = 0.34

Repeatability

The total time for the testing, including registration and data collection, was approximately 15 minutes, during which no discomfort was reported. The overall cohort had a mean supine coronal MFT angle of 0.1 ± 2.5° and a corresponding sagittal MFT angle of −1.7 ± 3.3° (mean ± SD). The intra-registration agreement of MFT angle measurements is shown for each of the two sets of registrations in Table II. Repeat coronal alignment readings with the volunteer lower limbs stationary agreed to almost ±1° for both the first and second registrations. For the first registration there was an approximately ±0.5° loss of repeatability for coronal alignment measured before and after collateral stress maneuvers, and a less significant loss of ±0.2° following stance trials. These small losses in coronal MFT angle repeatability were not seen with the second registration, which gave a consistent agreement of approximately ±1°. Sagittal alignment measurements were less repeatable overall by an approximate factor of two and were generally no more precise for consecutive stationary readings.

Table II.  Mean difference and 95% limits of agreement of repeat supine alignment measurements in extension with leg stationary and before and after both standing and collateral stress maneuvers (all measurements in degrees)

	Registration 1		Registration 2
	Mean difference	±1.96 SD	Mean difference	±1.96 SD
Coronal MFT angle consecutive	0.03	1.2	−0.02	1.1
Coronal MFT angle before & after stance	−0.1	1.4	0.07	1.1
Coronal MFT angle before & after stress	0.2	1.7	0.2	1.0
Sagittal MFT angle consecutive	0.2	2.2	−0.1	2.1
Sagittal MFT angle before & after stance	0.5	2.8	0.7	2.6
Sagittal MFT angle before & after stress	−0.3	2.2	−0.9	1.7

The agreement between the two registrations (Table III) indicated a repeatability of approximately ±1° for all the supine coronal alignment measurements, including the change with applied stress. On three occasions, a third registration process was required to obtain two registrations with a difference in supine coronal MFT angle of 2° or less.

Table III.  Inter-registration agreement of supine and standing coronal and sagittal MFT angles, and relative change following varus-valgus stress (measurements in degrees).

MFT angle	Mean difference	±1.96 SD
Supine coronal	−0.2	0.8
Supine sagittal	0.2	1.2
Change with varus stress	−0.3	1.3
Change with valgus stress	−0.2	1.1
Standing coronal	0.2	2.9
Standing sagittal	0.1	5.0

Standing alignment measurements showed less agreement for both coronal (±3°) and sagittal (±5°) MFT angles. These results are illustrated with Bland-Altman plots in Figure 3.

Figure 3. Bland-Altman plots showing mean difference (solid black line) and 95% limits of agreement (dotted grey lines) of MFT angular measurements for two trials: (a) supine coronal; (b) supine sagittal; (c) with varus stress; (d) with valgus stress; (e) standing coronal; and (f) standing sagittal.

Discussion

The quantification of knee alignment is a routine part of orthopaedic practice and is important for the monitoring of disease progression, the planning of interventional procedures, and the follow-up of patients. Currently available technologies and measurement techniques have a number of drawbacks including inaccuracy, limb positioning artefacts, and radiation exposure. This study developed a system that addresses some of these issues.

The stability of the IR tracker mountings permitted non-invasive kinematic measurement of knee alignment. For a single volunteer, the non-invasive attachments compared well with the rigid mountings of the leg model. The variance of volunteer measurements for repeated consecutive MFT angles on one registration was statistically greater that for the rigidly fixed mounting, but this difference is of doubtful clinical significance given that both set-ups demonstrated precision well within 1°. For repeated registrations, the SD for the non-invasive mounting was a third higher than for the leg model, and the actual range was 1° larger with no statistical difference between the two. This result was perhaps surprising, given that the leg model had a rigid hinge for a knee joint with no collateral movement and therefore a more consistent MFT angle. The only minor source of variation between trials on different days was the coupling mechanism between the trackers and fixation screws. In comparison, although care was taken to place the tracker base plates in a consistent anatomical location, the volunteer straps may not have been identically applied in terms of both position and tightness. This situation mimicked actual clinical use of the system for multiple assessments over time, where no indication of previous tracker position would be available. Furthermore, the small amount of natural collateral laxity of the volunteer knee could potentially have resulted in real differences in alignment on different days.

Further evaluation of the non-invasive tracker mountings was provided by the assessment of multiple volunteers. Following registration, the lower limb coronal and sagittal MFT angles could be repeatedly measured in real time, permitting an intra-registration assessment of tracker stability following stance and varus-valgus stress. These limb movements could have potentially modified the tracker position, but qualitatively they appeared stable throughout and remained in position for the duration of the measurements, with no complaints of discomfort. This observation of stability was reflected in the results for consecutive coronal MFT angle measurements in comparison to those obtained before and after stance and collateral stress of the knee. All repeatability was within levels of clinical relevance. For sagittal alignment the measurements were less repeatable overall within both sets of registrations, with the poorest agreement of up to almost ±3° being seen before and after stance. However, this may reflect a true difference in sagittal MFT angles rather than a change in tracker position. Some volunteers were noted to have poor relaxation, which often improved over the course of the assessment, thereby reducing resistance to full extension from the hamstring muscles. This resulted in a tendency for knees to become more extended towards the end of the trials, which could potentially explain the greater variation in sagittal measurements in comparison to coronal MFT angles, which were less likely to be affected by muscle tone.

The limits of agreement between the two sets of registrations were approximately ±1° for all supine alignments, including change with applied stress. For the initial supine coronal alignment measurements, only three gave inconsistent results that required repetition, and all repetitions were acceptable. Therefore, although the registration process was open to error, this was an infrequent occurrence, and a simple repeat protocol enabled it to be identified every time. The potential variation in applied manual load to the knee did not result in a loss of repeatability, as would perhaps have been anticipated. This may be explained by the consistency of the clinician performing the collateral stress maneuvers [27], which may have shown greater inter-observer variation if different examiners were assessed. Standing alignment measurements showed less agreement for both coronal (±3°) and sagittal (±5°) MFT angles. This may represent a true difference in alignment as a result of stance variation between trials, as volunteers were only instructed to stand on both legs as normal rather than assuming a position of maximum extension with their knees “locked” straight. Therefore, the variation in standing knee extension angle could be due to this lack of control of limb position. In comparison, the supine measurements were performed in a more reproducible manner by supporting the lower limb under the heel, and this was reflected in the narrower agreement limits illustrated with Bland-Altman plots. The ±5° scale of the vertical axis (except for standing sagittal measurements) was chosen to reflect the typical repeatability of other methods of assessing both sagittal [10] and coronal [24] knee alignment, including human variations in joint angle estimation [11]. However, it should be noted that considerably greater intra-observer estimates of knee flexion and extension angles have been reported, with critical differences between measurements of 7.1° to 21.4° [28].

The use of externally mounted markers and a motion capture system was not an entirely novel approach to measuring lower limb alignment. Mündermann et al. [24] used reflective marker sets and four high-speed cameras to measure static mechanical lower limb alignment, but reported only a moderate correlation (R²= 0.544) with the corresponding long-leg radiographs and a discrepancy of more than 5.3° for 10% of cases. However, the hip, knee and ankle joint centers were determined from anthropometric measurements, which are widely accepted as being inaccurate, particularly for the hip joint [29–32]. The experimental set up in terms of anatomical landmark identification, marker placement, multiple camera positioning, and data capture analysis also presented several limitations for a clinically adaptable measurement tool. In contrast, the system developed in this study consisted of a single portable camera unit with corresponding IR trackers that should be secure and visible but do not require specific anatomical placement. The kinematic registration process took approximately five minutes with on-screen guidance for performing simple joint movements to determine their rotational centers. The subsequent MFT angle was generated from kinematic data alone without the potential associated errors of anatomical landmark registration [33]. Hip joint center location errors were minimized by the use of a commercial software system (OrthoPilot® HTO version 1.5, BBraun Aesculap, Tuttlingen, Germany). The software contained an algorithm that rejected the points in space acquired during thigh circumduction if their spread was too large or the distribution was non-spherical [25]. The passive movements for kinematic registration were therefore required to be slow and controlled, which contrasts to other studies of functional joint center determination using active movements or gait [22], [23].

The immediate generation of real-time on-screen coronal and sagittal MFT angles presented a number of potential advantages over other measurement systems. Firstly, it enabled dynamic measurements of alignment to be made following applied stress or weight bearing with immediate visualization of angular displacement. The ability to measure the resultant change in the coronal MFT angle from a supine resting position when a collateral stress is applied has a potential clinical application in improving the measurement of relative varus and valgus knee laxity. Current methods are either subjective [34] or rely on adjuncts such as X-ray measurements of tibiofemoral gap opening [35] which are prone to potential radiographic errors associated with limb positioning [13], [14]. For weight-bearing conditions the measurements did not require strict rotational control of the lower limb, and the coronal MFT angle was recorded with the associated knee flexion angle. This IR system could therefore potentially offer a viable alternative to long-leg radiographs whilst also overcoming some of the previously discussed limitations.

This validation study also has its limitations. The measurements were made by a single clinician (J.V.C.) involved in the development of the system without an assessment of inter-observer variation. The true volunteer knee alignments were unknown, so validation of the measurement tool was based on repeatability rather than comparison to a measurement standard. However, the IR measurement system is validated for use with rigid tracker attachments. It could therefore be inferred that repeatable measurements are also accurate, as in order for measurements to be repeatable, soft tissue artefacts must be minimal. In addition, it could be argued that the acknowledged long-leg radiographic gold standard has more potential variation [14] than the IR system, and that disagreement between measurements may not reflect true inaccuracies [36]. Although several obese subjects were enrolled in the study, none were morbidly obese and no subject reported discomfort when performing the necessary kinematic maneuvers. The registration process may be less reliable in a typically more obese osteoarthritic population [37], [38] with potential for pain upon joint movement.

In summary, a non-invasive tool for measuring coronal and sagittal knee alignment under a number of dynamic, real-time conditions was developed and validated. The portability of the system makes it a potential tool for out-patient assessment, and the system provides an alternative to long-leg radiographs without the need for radiation exposure. The measurement of supine, standing and stress alignment on both asymptomatic and osteoarthritic subjects may help to further our understanding of the complex kinematics of the knee.

Declaration of interest: Dr. Picard has patents/licensing agreements with BBraun Aesculap. The other authors certify that they have no commercial associations (e.g., consultancies, stock ownership, equity interest, patent/licensing arrangements, etc.) that might pose a conflict of interest in connection with the article. Our department has received research funding from BBraun Aesculap.