Soft tissue artifact causes significant errors in the calculation of joint angles and range of motion at the hip

Abstract

Soft tissue movement between reflective skin markers and underlying bone induces errors in gait analysis. These errors are known as soft tissue artifact (STA). Prior studies have not examined how STA affects hip joint angles and range of motion (ROM) during dynamic activities. Herein, we: 1) measured STA of skin markers on the pelvis and thigh during walking, hip abduction and hip rotation, 2) quantified errors in tracking the thigh, pelvis and hip joint angles/ROM, and 3) determined whether model constraints on hip joint degrees of freedom mitigated errors. Eleven asymptomatic young adults were imaged simultaneously with retroreflective skin markers (SM) and dual fluoroscopy (DF), an X-ray technique with sub-millimeter and sub-degree accuracy. STA, defined as the range of SM positions in the DF-measured bone anatomical frame, varied based on marker location, activity and subject. Considering all skin markers and activities, mean STA ranged from 0.3 cm to 5.4 cm. STA caused the hip joint angle tracked with SM to be 1.9° more extended, 0.6° more adducted, and 5.8° more internally rotated than the hip tracked with DF. ROM was reduced for SM measurements relative to DF, with the largest difference of 21.8° about the internal-external axis during hip rotation. Constraining the model did not consistently reduce angle errors. Our results indicate STA causes substantial errors, particularly for markers tracking the femur and during hip internal-external rotation. This study establishes the need for future research to develop methods minimizing STA of markers on the thigh and pelvis.

1. Introduction

Joint angle measurements provide insights into movement abnormalities for clinical gait analysis and scientific investigations [1,2]. Most often, tracking of markers adhered to the skin surface serve as the basis for calculating joint angles; however, skin marker motion capture suffers from soft tissue artifact (STA) [3]. Therefore, measuring STA and its influence on joint angles and range of motion (ROM) calculations are important for interpreting the results from clinical gait analysis and gait models.

Soft tissue artifact results from unequal movement of soft tissue layers, including muscle, tendon and dermis, between the bone and the skin surface. STA can arise from three main sources [3]—skin sliding relative to underlying bone, inertial effects of skin motion, and deformation caused by muscle contraction. Model constraints [4] and skin marker locations [5] have been purported to limit the effects of STA, but the effect of model constraints and marker locations on hip joint angles and ROM remain unclear as neither has been assessed in the hip relative to a true reference standard.

Historically, STA has been measured in-vivo using pins implanted in bone [6,7]. While these studies provided valuable information, bone pins require invasive procedures for placement and may affect tissue movement between the skin surface and bone by restricting sliding between the soft tissue interfaces. More recent advancements permit minimally invasive assessment of STA and its effects on joint kinematics using X-rays. X-ray studies to-date have often focused on imaging the knee joint [8] or only obtained multiple static positions of the hip [9]. A more recent dynamic imaging technique, termed high-speed dual fluoroscopy (DF), has been utilized to measure in-vivo joint motion. The advantage of DF as a reference standard is that bone motion is measured dynamically with sub-millimeter and sub-degree accuracy without the need to implant pins [10].

The purposes of this study were to: 1) measure STA of skin markers on the pelvis and thigh during walking, hip abduction and hip rotation, 2) quantify errors in tracking the thigh, pelvis and hip joint angles, and 3) determine whether model constraints mitigated errors. We hypothesized that all markers would undergo significant, direction-dependent STA and that model constraints would not have an effect on errors from STA.

2. Methods

2.1. Subjects

Eighteen subjects signed informed consent prior to participation in this University of Utah Institutional Review Board (IRB) approved study. Only subjects who were free from lower extremity pain, had not undergone previous lower limb surgery, and had a relatively low body-mass index (BMI) (< 25 kg/m²) were recruited. Subjects were screened with an anterior-posterior (AP) radiograph and excluded if they exhibited signs of hip dysplasia, femoroacetabular impingement, osteoarthritis, or other obvious anatomical deformities. Seven subjects were excluded based on these criteria, leaving 11 subjects who participated in the study (6 male). These subjects were recreationally active, young adults with the following characteristics (mean (SD)): age 23.2 (2.2) years, height 173.3 (10.4) cm, mass 63.8 (10.9) kg, BMI 21.1 (1.9) kg/m². The left or right side was chosen for imaging to yield a balanced number of left and right hips (6 right).

2.2. Activities

Subjects completed six activities: 1) standing (static), 2) level walk, 3) incline (5°) walk, 4) abduction to 45° then back to neutral, 5) internal hip rotation to end range of motion and 6) external hip rotation to end range of motion [11]. The stance of the subject was standardized during the static activity, wherein subjects stood with their feet hip width apart, toes pointing forward, and knees and hips at neutral (0°). The hip rotation activities were performed with the feet on the ground while angular changes were permitted at the ankle and knee joints. All trials were completed on a dual-belt treadmill (Bertec Corporation, Columbus, OH, USA). After the static, activities were performed in a random order. The speed for the walk trials was set to the subject’s self-selected over-ground walking speed [11]. To best represent errors in the transverse plane over the subject’s entire ROM, the internal and external hip rotations trials were combined, hereafter referred to simply as “rotation.” Two trials were acquired for all dynamic activities. The trial with the greatest range of motion and/or highest quality fluoroscopy images was analyzed. Image quality was assessed qualitatively by inspection prior to model-based tracking (NMF). For two subjects, no abduction trials were acquired, as the allowable time for radiation exposure, as set by the IRB, had expired.

2.3. Skin marker motion capture

Retroreflective spherical markers, 14 mm in diameter, were placed on the femur and pelvis to track dynamic motion [11]. For the pelvis, skin markers were placed bilaterally over bony landmarks at the anterior superior iliac spines (ASIS), posterior superior iliac spines (PSIS), and iliac crests (ILC), and the markers for the femur were placed on the lateral epicondyle, greater trochanter and a cluster of four markers on the lateral thigh [11]. The markers on the lateral thigh consisted of a stiff plate (14 cm × 12 cm) of four markers adhered with Velcro to a wrap around the thigh. The four markers were fixed to the plate in a diamond pattern 1.5 cm from the edge. The Velcro was wrapped around the thigh as superior as possible, which resulted in the marker cluster on the lateral thigh approximately midway between the greater trochanter and lateral knee markers. Markers were placed on subjects by the same study team member (NMF) after receiving training from a licensed Physical Therapist with 10 years of motion capture experience (KBF). Anatomical coordinate systems were defined based on International Society of Biomechanics recommendations [12]. Marker positions were measured with a 10-camera system (Vicon Motion Systems, Oxford, UK) running Nexus software (v1.8.5) at 100 Hz. Nexus software was also used to reconstruct and gap-fill 3D marker positions [11].

2.4. Dual fluoroscopy and model-based tracking

The DF custom system (Radiological Imaging Services, Hamburg, PA, USA) and model-based tracking used herein were previously validated to a bias and precision less than 1 mm and 1° [10]. High-speed cameras mounted to the back of the image intensifiers (i.e., x-ray receivers) acquired images at 100 Hz [11]. The 3D positions of bones were measured from dual fluoroscopy images using model-based tracking (DF-MBT) [13], which semi-automatically aligned projections from computed tomography (CT) surfaces with DF images. The DF-MBT solutions transformed bony landmarks identified on CT surfaces into the DF coordinate system [10].

2.5. Merging vicon and DF measurements

The Vicon and DF systems were temporally synced with an external voltage signal acquired by both systems simultaneously. The coordinate systems were synced spatially using a calibration cube that included metal beads at the center of retroreflective markers [14]. Bony landmark positions from DF-MBT measurements were transformed into the Vicon lab coordinate system and combined with skin marker (SM) positions in MATLAB (v7.10, The Mathworks, Inc., Natick, MA, USA). All SM and DF bony landmark positions were filtered in MATLAB using a 4th-order bidirectional low-pass Butterworth filter with a cutoff frequency of 6 Hz [10].

2.6. Kinematic models

Kinematic models were implemented in Visual3D (v5.02.11, C-Motion, Germantown, MD, USA). The reference model was a six degree of freedom model (“6 DOF”). A second model was analyzed that restricted femur motion to only rotations relative to the pelvis (“3 DOF”) via limiting the degrees of freedom under the Inverse Kinematics (IK) Constraints tab. To investigate whether excluding certain markers improved joint angle estimations, a third model excluded the greater trochanter and knee markers on the thigh and excluded the ILC markers on the pelvis (“Markers”). The Markers model included six degrees of freedom. All markers in all models were given equal weights. To eliminate differences between anatomical coordinate systems (ACS) generated by SM and DF, the pelvis and femur ACS as generated by calibration with DF landmarks were used for calculation of SM body segment positions and joint angles during dynamic activities. For all models, the femur’s proximal joint was set to the DF-measured geometric center of the femoral head [14].

2.7. Error calculations and soft tissue artifact metrics

When segment positions, joint angles, and range of motion were compared, DF served as the reference (i.e., SM – DF). To minimize the contribution of anatomical variability across subjects, all metrics during dynamic activities were offset by their value during the static trial. To quantify STA, the range of skin marker positions was calculated relative to the underlying bone anatomical coordinate system. This definition is also known as “peak-to-peak” values in the body segment axes [15]. A root-mean-squared (RMS) error was calculated between DF and SM solutions to assess the accuracy of the models.

2.8. Statistics

Data were plotted and presented as mean (lower and upper limit for 95% confidence interval, CI), unless otherwise noted. When differences for a single model and activity were analyzed, a Student’s t-test or Wilcoxon sign-rank test was employed in MATLAB, depending on the outcome of a Lilliefors test for normality. For comparisons that were made in multiple anatomical directions, P values were adjusted for multiplicity with the Holm step-down procedure [16]. When all dynamic activities were analyzed at once, multivariable linear regression was employed in Stata (v14.1, StataCorp LP, College Station, TX, USA). When interactions were observed between model type and activity, each activity was tested in isolation and P values were adjusted for multiplicity. For all tests, significance was set at P < 0.05.

3. Results

Qualitative plots of STA during dynamic activities demonstrated the amount of movement of skin markers relative to underlying DF-measured bone (Fig. 1). Markers that tracked the femur exhibited the largest STA, especially the greater trochanter marker. The greater trochanter marker underwent 5.2 (4.5 5.9) cm STA in the AP direction during rotation. During level walk, the thigh cluster markers showed relatively smaller STA in the AP direction (1.8 (1.5 2.1) cm); however, thigh marker STA was still relatively large during the rotation activity (3.6 (3.1 4.2) cm). Pelvis markers showed smaller STA than femur markers and more consistent magnitude of STA between markers. Marker STA varied between subjects, as evidenced by the large 95% CIs relative to mean values (Supplemental Table S-1). Nearly every marker demonstrated larger STA for certain anatomical directions, which was evident for all activities analyzed (Fig. 1 and Supplemental Table S-1).

Fig. 1.

Soft tissue artifact during dynamic activities. For each skin marker, six spheres were plotted at a distance equivalent to ± the mean range of skin marker positions along the three anatomical bone axes as measured using DF. A longer distance between spheres represents more skin marker movement in that direction relative to the underlying bone. Markers were plotted relative to a representative subject’s CT bone surfaces. Note that the spheres’ diameter matches the diameter of skin markers used during motion capture (14 mm), and that the entire length of the femur bone was not imaged with CT to limit radiation exposure. See Supplemental Table S-1 for quantitative measurements of mean and 95% CI ranges during each dynamic activity. See text for marker abbreviations. S = superior. I = inferior. M = medial. L = lateral. A = anterior. P = posterior.

Segment position error and joint angle error demonstrated significant differences between SM and DF measurements during the static and dynamic activities (Fig. 2 and Supplemental Table S-2). During the static trial, the pelvis and femur positions as measured with SM were anterior of the DF-measured segments (0.7 (0.4 0.9) cm and 0.5 (0.2 0.8) cm, respectively), and the hip joint angle for SM measurements was externally rotated relative to DF (7.8 (3.8 11.8)°). When considering all dynamic activities, the joint angles were underestimated by SM measurements, with the hip joint more extended (−1.9 (−2.5 −1.2)°), more adducted (−0.6 (−1.0 −0.2)°), and more internally rotated (−5.8 (−7.4 −4.3)°).

Fig. 2.

Segment position and hip joint angle errors during static and dynamic activities. Errors were defined as the difference between segment positions and joint angles tracked by skin markers relative to the same segment and angles tracked with dual fluoroscopy (i.e., SM – DF). Horizontal bars represent the mean across all subjects, and error bars represent ± 95% CI. Static errors were defined on the static calibration trial (top row). To eliminate differences between anatomical coordinate systems generated by skin markers and dual fluoroscopy bony landmarks, the pelvis and femur ACS as generated by calibration with DF landmarks were used for calculation of SM body segment positions and joint angles during dynamic activities. To eliminate errors due to anatomical variability between subjects, errors during dynamic activities (bottom row) were offset by the value during the static trial. An asterisk (*) indicates a significant difference between SM and DF (P < 0.05). See Supplemental Table S-2 for errors during each dynamic activity. FE = flexion-extension. AB = abduction-adduction. Rot = internal-external rotation.

The overall patterns of joint angles plotted dynamically over the time-normalized gait cycle were similar between SM and DF (Fig. 3). The largest difference was observed about the internal-external rotation axis, where the SM measurements were more internally rotated. Even though joint angles were qualitatively similar between measurement systems, joint angle range of motion was consistently reduced for skin marker measurements relative to dual fluoroscopy (Fig. 4). The joint angle range about the flexion-extension axis as measured by skin markers was reduced during the level walk and incline walk activities by 7.3° and 6.6°, respectively. For the internal-external rotation axis, the SM joint angle range was smaller by 21.8° for the rotation activity. Joint angle ranges were different about all axes during the abduction activity, including a difference of 2.1° about the abduction-adduction axis.

Fig. 3.

Hip joint angles during level walk. Mean joint angles were plotted for dual fluoroscopy measurements (black lines) and skin marker measurements (lighter colored lines). Thinner lines above and below the mean represent ± 95% CI. The gait cycle was defined from foot contact (1%) to the next ipsilateral foot contact (100%). DF = dual fluoroscopy. SM = skin markers.

Fig. 4.

Hip range of motion as measured with skin markers and dual fluoroscopy during dynamic activities. Bar height represents mean ROM with error bar ± 95% CI for dual fluoroscopy measurements (black lines) and skin marker measurements (lighter colored lines). An asterisk (*) indicates a significant difference between DF and SM measurements (P < 0.05). DF = dual fluoroscopy. SM = skin markers. FE = flexion-extension. AB = abduction-adduction. Rot = internal-external rotation.

Neither the degrees of freedom of the model nor marker configuration reduced RMS error during the walk and abduction activities (Fig. 5). During the rotation activity, the 3DOF model had lower RMS error for the abduction-adduction axis, but higher error about the internal-external rotation axis, as compared to the 6DOF model. The Markers model also had lower error about abduction-adduction axis and higher error about the internal-external rotation axis.

Fig. 5.

Root-mean-squared (RMS) errors for joint angles as a function of model type during dynamic activities. Bar height represents mean RMS error across all subjects with error bar ± 95% CI for the 3 DOF model (darker shade), 6 DOF model (no shade) and Markers model (lighter shade). An asterisk (*) indicates a significant difference from the 6 DOF model (P < 0.05). FE = flexion-extension. AB = abduction-adduction. Rot = internal-external rotation. DOF = degrees of freedom.

4. Discussion

We found that STA was large and differed by anatomical direction, dynamic activity and subject. Joint angles differed between skin markers and dual fluoroscopy measurements during static and dynamic activities. In fact, the amount of STA was similar in magnitude to the positional and angular errors that resulted from static calibration (see Fig. 2). Range of motion was considerably underestimated by skin markers during dynamic activities. Restricting the hip to rotations only (3 DOF) and tracking only the thigh and ASIS/PSIS markers (Markers) did not universally reduce RMS angle errors. Thus, we confirmed our hypotheses that all markers would undergo significant, direction-dependent STA and that model constraints would not unequivocally reduce errors from STA.

Our overall assessment of STA was similar to that reported previously [3,17], namely, that STA varied by marker location, activity and subject. When averaging across all thigh markers, Akbarshahi et al. [8] measured slightly smaller errors (0.6 cm, 0.8 cm and 1.0 cm in the AP, SI and ML directions) than the thigh cluster markers in our study (1.8 cm, 1.6 cm and 1.0 cm, respectively). It is possible that these discrepancies arise due to marker location, as the study by Akbarshahi et al. [8] tracked six markers over a larger area on the surface of the thigh. A study by Barre et al. [18] measured STA for 80 markers on the thigh during gait and found a maximum error (2.5 cm) within 1 cm of the largest error in our study (1.8 cm in the AP direction).

While many studies have published results on tibia position errors and knee joint kinematics during dynamic motion, to our knowledge no previous study published results that measured dynamic pelvis position errors and hip joint angle errors in-vivo. A study by Hara et al. [9] measured pelvic STA at multiple static postures and found the largest STA in ASIS markers (1.7 cm) at 90° hip flexion, which matches the linear distance STA that we measured for the ASIS markers during level walking albeit over smaller joint angle changes (see Supplemental Table S-1). A recent study by Camomilla et al. [19] investigated STA at two static postures that mimicked the beginning and end of mid-stance. The previous study’s results were approximately 1 cm smaller than ours for the anterior pelvic landmarks (0.8 cm vs. 1.7 cm, respectively) and the posterior pelvic landmarks (0.3 cm vs. 1.3 cm, respectively). The larger STA was likely due to the relatively larger joint angle ranges completed by the subjects in the current study and the dynamic nature of our measurements.

In general, we found that altering the degrees of freedom of the model did not reduce errors in hip joint kinematics. Our findings in this regard are similar to other studies that found model iterations did not reduce errors from STA [4,20]. Conversely, other investigators found that marker configuration and choice affected STA and estimates of joint kinematics [5,21,22]. The contrast could be due to the fact that our study focused on three common marker locations on the thigh and pelvis. Thus, the reader should be cautioned against extending the conclusions of the current study beyond the marker locations investigated. Furthermore, in an attempt to account for differences in bony anatomy, we offset joint angles by the values calculated during the standardized static activity. Femoral version, in particular, varies widely among non-pathologic hips [23]; if we had not offset angles, differences in version angle would lead to different hip internal-external rotation joint angles for the same pose. Similar studies did not offset angles, which may be a source of differences between our study and prior work.

Perhaps our most clinically relevant finding was that range of motion was reduced for the SM measurements as compared to DF measurements. For all activities, SM-measured range of motion was underestimated in the primary movement direction. Based on our results, we recommend that investigators use caution when interpreting range of motion from skin marker measurements, especially those representing internal-external rotation in the transverse plane.

Reductions in range of motion support the notion that “skin sliding” represents the major contributor to STA [24]. While skin deformation due to inertial effects and muscle contraction contributes to STA as well [3], the reduced ROM measured with skin markers suggests that bone can move underneath the skin without an equivalent movement on the skin surface, especially for the femur about the internal-external rotation axis. The results presented here motivate future work to minimize STA and compensate for its effects [3].

While the DF-MBT technique boasts sub-millimeter and sub-degree bias and precision, the technique and this study were not without limitations. The radiation exposure for the subjects, including CT and DF, was 10.72 mSv. This amounts to 21% of the annual exposure limit for a radiation worker, or nearly three years of background radiation in the Salt Lake City, Utah, area. Being at elevation, Salt Lake City residents are exposed to higher levels (average > 4 mSv) due to increased cosmic radiation [25], as opposed to the worldwide average (2.4 mSv) [26]. The reader should also note that the annual dose limit for the general public from licensed operation is 1.0 mSv [27]; however, this limit excludes background radiation and voluntary participation in medical research studies. In addition, our subject population was relatively limited to 11 subjects, though, as far as STA studies are concerned, this puts our study at the higher end of subject participation [3]. Direct kinematics (DK), a method used in the conventional gait model to calculate joint angles, was not used in the current study to obtain joint angles. Previous research [28], however, showed that there are only minor differences between joint angles obtained with DK and IK if the same anatomical model is used. Furthermore, our population was limited to subjects with a relatively low BMI; thus, the results reported in our study likely represent the low end of STA.