Facet Joint Distance Measurement Using Digital Tomosynthesis While Standing

Abstract

The zygapophyseal (facet) joint plays a critical role in load transmission and stability of the spine, and facet degeneration is a common consequence of aging and osteoarthritis. The ability to accurately measure facet space is important, as decreased facet space is associated with facet degeneration and lower back pain. Although grading systems exist for assessing facet joint space narrowing, static imaging fails to characterize changes in the facet gap under load that play a role in segmental stability. Current methods for estimating the dynamic behavior of the facet joint are either inaccurate, radiation costly, or clinically impractical. In the current study, we demonstrate the feasibility of a novel method for 3D measurement of facet joint space using digital tomosynthesis (DTS) imaging in supine and standing positions. Facet gap measurements were found to be strongly correlated with (r to 0.98) and accurate (<20 μm error for median facet gap) relative to microcomputed tomography reference values. In a pilot in vivo demonstration with seven participants, the effect of physiological loading was detectable, with median facet joint space being larger in standing as compared to supine images (p < 0.0001). The presented approach may be useful in directly characterizing changes in the facet joint relevant to segmental stability that are not readily assessed via current clinical imaging methods.

INTRODUCTION

The zygapophyseal, or facet, joint is a critical structural component in articulation of the spinal column and plays an important role in spinal stability and load transmission. The facet joint complex is comprised of the adjoining superior and inferior articular processes surrounded by a fibrous synovial capsule. Facet joints are thought to carry some portion of the axial load, with estimates varying from 3% to 33%, and higher (47% to 70%) in the case of arthritic joints and disc degeneration (Adams and Hutton, 1983; King et al., 1975; Yang and King, 1984).

Facet joint narrowing, degeneration, and arthropathy are common consequences of aging and osteoarthritis. In terms of radiographic assessment of these changes, computed tomography (CT) is the preferred modality (Cohen and Raja, 2007; Faure et al., 2016; Kalichman and Hunter, 2007), and offers an improvement over plain radiography because the joint can be viewed in multiple planes. Magnetic resonance imaging (MRI) and single photon emission computed tomography (SPECT) offer further improvements especially regarding soft tissue effects of facet degeneration including synovitis, effusion, and marrow edema (Faure et al., 2016). Grading systems for facet narrowing, with normal dimensions defined based on the typical thickness of the cartilaginous layers, exist for CT and have been evaluated for use with MRI (Pathria et al., 1987; Weishaupt et al., 1999). Assessment of facet gap distance is typically performed by averaging several lengths measured from static images (Chung et al., 2012; Pathria et al., 1987; Weishaupt et al., 1999), and more complete characterization of the facet gap has been described using 3D distance mapping of the joint space via CT imaging (Prado et al., 2022; Simon et al., 2012). The ability to accurately measure facet space is important, as decreased facet space is associated with facet degeneration and lower back pain (Rahimzadeh et al., 2017; Shi et al., 2017; Simon et al., 2012) and increased facet space may be associated with spondylolisthesis and segmental instability (Chung et al., 2012; Hasegawa et al., 2010). The association with spondylolisthesis is particularly significant, as spondylolisthesis is the most common indication for lumbar fusion surgery, with prevalence in females over age 69 as high as 39% (Wang et al., 2017). However, spondylolisthesis alone cannot fully explain spinal instability: intraoperative mechanical loading studies suggest that facet opening may indicate instability where the segment would otherwise be considered stable on the basis of spondylolisthesis (Hasegawa et al., 2014).

Measuring the facet space under load may provide information in regard to segmental stability beyond that achieved through static measurements alone. Chronic excess facet joint capsular strain may be a consequence of overload or repetitive strain injury and is implicated in the development of facet arthropathy and joint effusion (Cohen and Raja, 2007) and chronic overstretching of the richly innervated joint capsule may be a source of pain during sustained postures (Adams and Hutton, 1983). This laxity in the ligamentous capsule may be estimated by the change in the gap between the adjoining facet surfaces, but static imaging methods fail to accurately characterize these changes in the facet gap under load that play a role in segmental stability. The extent to which measurements made from static supine imaging positions are associated with dynamic changes in facet joint space is currently unknown, and little is known in general regarding how the facet joint space changes in vivo under physiological load. Directly measuring this change would require imaging the joint under unloaded and weight-bearing configurations, and facet gap measurements from flexion-extension radiographs fall short in accurately characterizing spinal instability (Viswanathan et al., 2018). To date, in vivo 3D measurement of the facet gap has involved either comparison of paired static CT images acquired under flexion/extension conditions or dynamic in vivo stereo-radiogrammetry. In paired CT analysis, translations in the facet joint have been estimated by co-registering serial images of the spine in postures (not weight-bearing) provoked by a positioning jig with 4.5–6.9 mm of facet joint translation found to occur between flexion and extension positions for L4/5 and L5/S1 levels (Svedmark et al., 2012). Dynamic stereo-radiogrammetry studies suggest increased facet distance between supine and standing positions (Kozanek et al., 2009), with an overall trend toward facet gap opening during weight lifting (Byrne et al., 2018). Although useful for research applications, current in vivo measurement techniques may be impractical in a clinical setting due to high radiation exposure or technical complexity.

Digital tomosynthesis (DTS) imaging presents as an ideal modality for 3-dimensional measurement of the facet joint space because of several favorable features. First, DTS has relatively low radiation dose exposure, about 12–17% that of clinical CT (Mettler et al. 2008, Zhang et al. 2014). For systems with an adjustable tilt table or wall-mounted bucky, the facet joint can be visualized in supine and/or standing configurations as a series of sharply high resolution planes (150–300 μm voxel size) acquired under a patient’s own body weight. Tomosynthesis images also have strong contrast between dense structures and noise, making the modality particularly well suited for delineating cortical bone edges (Gang et al., 2010). Tomosynthesis views can also conceivably be acquired in arbitrary positioning, not necessarily aligned with the imaging table, allowing imaging of the facet joint in various directions or flexion/rotation configurations. DTS has been shown to better characterize subtle degenerative changes in the vertebral facet joint such as erosions and sclerosis than plain x-rays (Joo et al., 2017). The use of DTS imaging has also been demonstrated for joint space measurements in the tibiofemoral joint (Kalinosky et al., 2011), hand (Hatano et al., 2017), and in load-bearing conditions (Ha et al., 2019; Oravec et al., 2019; Teramoto et al., 2014), but not the vertebral facet joint.

Therefore, in the current study, we performed direct 3D measurement of the facet joint space from DTS images with the specific objective to 1) assess measurement accuracy error using a microcomputed tomography reference and 2) evaluate in vivo feasibility with the specific hypothesis that facet joint space increases when standing.

METHODS

Measurement accuracy study:

A phantom study was performed to establish the measurement accuracy error of DTS facet distance measurements relative to a high resolution standard. A Sawbones radiographic lumbar spine model with high resolution radiopaque coating was used as a geometric reference (Model 1352–48, Sawbones, Vashon Island, WA). The use of Sawbones specimens allowed assessing measurement accuracy over a range of facet-to-facet distances approximating in vivo geometry. Each pair of vertebrae (L1–2, L2–3, L3–4, and L4–5) were imaged and reconstructed at 62 μm isotropic voxel size using a custom microcomputed tomography (μCT) system described previously (Oravec et al., 2020). The Sawbones model was also imaged en bloc in a clinical digital tomosynthesis system (Shimadzu Sonialvision Safire II, Kyoto, Japan). A series of 74 projection images were acquired over 40° tube angulation, and images were reconstructed in a stack of coronal planes (0.28 mm pixel size, 1mm slice thickness) and further resampled to isotropic voxel size using linear interpolation. Throughout DTS and μCT imaging, the segment was constrained using radiolucent positioning fixtures to minimize motion and ensure alignment with image axes.

Reconstructed DTS and μCT images were segmented in MIMICS (Version 24, Materialise NV, Belgium). Briefly, for all vertebrae in the segment (L1–5), an initial segmentation mask was created using livewire edge delineation with automatic mask interpolation between coronal slices. The 3D geometric model created from the initial segmentation mask was refined using 3D contour editing tools and saved as a stereolithography (STL) mesh. Pairs of meshes (L1/2, L2/3, L3/4, and L4/5) were imported for analysis in Meshlab (Cignoni et al., 2008). The 3D polygon selection tool was used to define margins of each facet surface, which was then extracted into unique layers. For each pair of vertebrae, a nearest-distance map was calculated between all vertices on the left and right facet surfaces of the superior vertebra with respect to the adjacent inferior facet surface (Figure 1). Average and median of the distance distribution were calculated for each facet joint to assess measurement accuracy. For diagnostic purposes, standard deviation, coefficient of variation, median, the first (Q1) and third (Q3) quartiles, interquartile range (IQR: Q3-Q1), and nonparametric coefficient of variation (NCV: IQR/median) were also calculated. During analysis, the reader was blinded to the vertebral level and side.

Figure 1:

Participants were imaged in lying (A, top) and standing (A, bottom) configurations. As described for measurement accuracy studies, reconstructed DTS images (B, top) were segmented in MIMICS (B, bottom) and from each facet pair (red arrows), distance metrics were derived from nearest-distance maps for each pair of facet surfaces in supine (C, top) and standing images (C, bottom). In this case, distance maps indicate a larger median facet gap when standing (left/right average = 0.86 mm) as compared to the supine position (left/right average = 0.43 mm).

In vivo feasibility study:

For an in vivo proof of concept, images from seven participants (6 women, 1 man; ages 58–69 years) were utilized from another study executed with goals unrelated to the current study. Images were selected to include participants with L1 and L2 vertebral levels available in images, and to exclude participants with scoliosis, a history of vertebral fracture at the levels of T12-L3, or history of bone disorders other than osteoporosis. The utilized images were a convenience sample of seven participants that fulfilled these criteria. All study procedures were performed after institutional IRB approval.

Participants were positioned in a supine configuration on the DTS system tilt table. Fluoroscopy scout views were used to ensure consistent centering on T12-L1 vertebral levels within the 15×17” field of view. Projection views were acquired using a 3.2 mAs radiographic technique with kV set to match the fluoroscopy scout view. The tilt table was then brought to a vertical position and images were acquired while standing using the same technique. To minimize movement between lying and standing scans, a chest restraint strap was utilized during imaging (Figure 1). Coronal planes were reconstructed at 0.28×0.28 mm in-plane pixel size with 1 mm slice thickness and resampled to isotropic voxel size (0.28 mm³) prior to segmentation. DTS reconstructions were segmented in MIMICS and distance metrics were recorded from pairs of L1 and L2 vertebrae in Meshlab as described for phantom studies. During analysis, the reader was blinded to the image type (supine or standing) and participant number.

Statistical analysis:

The distance distribution for each joint was tested for normality using Shapiro-Wilk or Kolmogorov-Smirnov tests. Distributions of facet-to-facet distances within each joint were nonnormal for Sawbones (both μCT and DTS, p<0.01 to p<0.0001) and in vivo cases (p<0.003 to p<0.0001), except for one in vivo joint (p>0.25). Therefore, nonparametric measures were deemed more appropriate to quantify central and dispersion characteristics of facet-to-facet distances within a joint. The median was examined as the metric representing central tendency, IQR as the dispersion, and NCV as the relative dispersion. 25^th (Q1) and 75^th (Q3) percentiles of the distribution were included to understand the influence of lower and higher magnitude distances on findings.

Analysis of data from phantom:

Regression was used to examine the relationship between DTS and the corresponding μCT measurements. Bland-Altman plots were used to assess agreement between DTS and μCT measurements of median distance (Bland and Altman, 1986). Systematic and proportional bias were evaluated relative to reference values using one-sample t-test for mean differences and correlation between differences and mean. Overall accuracy error was expressed as the average difference between DTS and μCT measurements.

Analysis of in vivo data:

In order to establish the effect of standing on the facet joint gap, mixed ANOVA models were constructed with loading (standing, supine), side (left, right) and their interactions as fixed effects and participant as a random effect. If an effect was found to be nonsignificant, analysis was repeated with nonsignificant effects removed. Mixed ANOVA was used to test the effect of loading on facet joint asymmetry (i.e., the absolute difference in distance metrics between left and right sides). In order to assess the dependence of facet space when standing on baseline facet space, the change in median distance for the side (left, right) with the largest change between supine and standing (in all cases an increased gap distance) was regressed upon its supine width.

All analyses were performed in JMP (v10, SAS Institute, Inc., Cary, NC) with statistical significance considered at p < 0.05.

Based on the available in vivo sample (N=7), with the assumption that supine and standing measurements are moderately correlated (R=0.5), a change in facet-to-facet distance from supine to standing was expected to be statistically demonstrable with a power of 0.8 for an effect size greater than 1.27 (Kreidler et al., 2013).

RESULTS

Correlations between DTS and reference μCT measurements were strong: r = 0.77 to 0.98 (Figure 2), and accuracy error for median facet-to-facet distance was under 20 μm (Table 1). The mean difference in median facet-to-facet distance between DTS and μCT (0.019 mm) was not significantly different from zero (p = 0.90), and all measurements fell within the 95% limits of agreement (dashed lines, Figure 3). The slope of the regression of differences upon means was nonsignificant (p = 0.26).

Figure 2:

Correlation plots for (A) median, (B), 25^th percentile, (C), 75^th percentile, (D) interquartile range, and (E) nonparametric coefficient of variation measured from Sawbones specimens.

Table 1:

Descriptive data and accuracy error for facet-to-facet distance measurements of Sawbones specimens.

Variable	μCT (Av±SD)	DTS (Av±SD)	Accuracy error (μCT – DTS)
Median (mm)	1.47±1.24	1.45±1.05	0.019
25th Percentile (mm)	1.03±0.92	1.06±0.83	0.033
75th Percentile (mm)	2.01±1.66	1.79±1.19	0.225
IQR (mm)	0.99±0.77	0.73±0.44	0.258
NCV	0.73±0.18	0.58±0.24	0.147

Figure 3:

Bland-Altman plot for median distance. The mean difference between DTS and μCT (solid line) measurements was small relative to changes due to standing without evidence of proportional bias. Limits of agreement are presented as dashed lines.

For in vivo trials, the effect of side (left, right) and its interaction with loading (standing, supine) were not significant (p = 0.18–0.87). There was also no significant effect of loading when the absolute difference between left and right was considered (no left-right asymmetry change due to loading; p = 0.46–0.86). When these nonsignificant effects were removed from the model, the effect of loading was significant for median, NCV, and 25^th and 75^th percentiles of the distribution, but not IQR (Table 2, and see Figure 4 for median and NCV). The change in median facet joint width due to standing was not correlated with the joint space measured in the supine configuration (p = 0.84, r = 0.1).

Table 2:

Descriptive data for in-vivo facet-to-facet distance measurements.

Variable	Standing (Av ± SD, mm)	Supine (Av ± SD, mm)	p-value
Median	1.25±0.41	0.73±0.27	<0.0001
25thPerc	1.02±0.34	0.53±0.22	<0.0001
75thPerc	1.48±0.48	0.98±0.33	<0.0003
IQR	0.46±0.20	0.45±0.16	0.71
NCV	0.37±0.10	0.65±0.22	<0.0001

Figure 4:

(A) Increased median facet-to-facet distance and (B) decreased relative dispersion of facet-to-facet distance were observed during standing.

DISCUSSION

In the current study, we assessed the feasibility of direct 3D measurement of facet joint space using digital tomosynthesis (DTS) imaging. Collectively, the current results indicate that measurements of facet joint spaces from DTS were strongly correlated with reference values from μCT and may be useful in detecting changes due to loading in vivo.

Measured values from in vivo supine (median 0.27 to 1.24 mm) and standing (median 0.78 to 2.06 mm) DTS images are in the lower range of reported values for measurements made using a similar technique applied to CT images (1.48 ± 0.44 for the medial zone to 1.93 ± 0.51 mm for the central zone) (Simon et al., 2012). This is consistent with previous work in which we noted considerable differences in distance metrics between DTS and CT, where measurements of thin structures (<3 mm) using CT may be overestimated due to partial volume effects (Dougherty and Newman, 1999; Newman et al., 1998) and are considerably larger than DTS (Johnsson et al., 2010; Oravec et al., 2015). In the context of a feasibility study, it was important to establish that distances are even measurable considering the substantial difference in resolution between DTS and μCT. The mean difference between DTS- and μCT-measured median distances was small with no statistically demonstrable evidence of systematic or proportional bias (Figure 3). Measured values for DTS median facet-to-facet distance were found to be accurate relative to reference values from μCT (1.45 mm vs 1.47 mm; Table 1). In addition to the median, DTS was strongly correlated with μCT at both the upper and lower quartiles of the distribution, indicating that the strength of the correlation observed for median was not simply dictated by good correlations in larger distances.

The approach presented in this study is relevant to the interaction between degenerative changes in the components of the so-called three-joint complex consisting of the intervertebral disc and adjoining facet joints. Although facet joint pathology is often thought to be secondary to disc degeneration (Butler et al., 1990; Fujiwara et al., 1999), there is considerable controversy in the literature regarding this assumption. Many studies suggest that disc degeneration does not fully explain facet joint dysfunction and acknowledge that facet dysfunction in fact precedes disc degeneration. This concept dates back decades, and several cadaveric studies conclude that there is no relationship between disc degeneration and facet joint osteoarthritis at thoracolumbar (Malmivaara et al., 1987) and lumbar levels (Lewin, 1964; Swanepoel et al., 1995). Indeed, in a clinical study, 92% of patients with discogenic pain (assessed by provocative discography) did not have facet joint pain (assessed by positive facet block), and conversely, only 3% of those obtaining relief from facet joint blocks had no evidence of discogenic pain (Schwarzer et al., 1994). Regardless of which process occurs first (disc degeneration or facet dysfunction), excess facet gap opening is thought to exacerbate degenerative processes in the disc (Dunlop et al., 1984; Haher et al., 1994). Disc degeneration may in turn alter the relative contribution of the facet joints in maintaining segmental stability (Schmidt et al., 2013). The facet gap measurement approach presented in the current study is expected to contribute to our understanding of these processes.

Facet instability is an important consideration in treatment planning, and current diagnostic methods rely on indirect measurements of dynamic instability such as facet fluid sign (measured from MRI) or relative position of vertebral bodies (i.e., spondylolisthesis measured from static images or via flexion-extension radiographs). Although facet joint measurement from radiographic imaging in static positions may provide valuable information in assessing changes in aging and low back pain (Rahimzadeh et al., 2017; Shi et al., 2017; Simon et al., 2012), functional changes in the facet space under physiological load are not readily assessed via static imaging. In the current study, the lack of correlation between baseline gap and difference under load suggests that measurements performed in the static supine condition may not adequately represent the dynamic behavior of the facet joint. It is reasonable to expect that direct measurement of joint space change under physiological load, either alone or in combination with other clinically measurable metrics related to facet instability, would provide a more complete characterization of degenerative changes prior to treatment.

In the current study, the comparison of supine and standing images indicates the facet joint space is larger for standing images, with no significant evidence of left-right asymmetry. Dispersion (IQR) was found to be nonsignificant, and the difference in relative dispersion (NCV) is therefore owing to change in the median. Evidence in the literature relating to facet gap changes under weightbearing conditions is sparse, likely due to the technical challenges in quantifying these differences, however two studies were identified that measured facet translation during physiological loading. Both studies utilized an in-vivo model-based tracking approach (dynamic in vivo stereo-radiogrammetry) wherein the differences in local coordinate systems of each facet surface, owing to dynamic motion or differences in static positions, were used to derive relative translations and rotations of the facet surface. In one study comparing facet motion between supine (MRI) and standing (estimated via biplane fluoroscopy) configurations, adjacent superior facet surfaces were observed to translate primarily in the cranial direction relative to the fixed inferior level, consistent with facet gap opening (Kozanek et al., 2009). Similarly, a trend was identified toward facet gap opening during dynamic weight-lifting activities between flexed and upright standing positions (Byrne et al., 2018). In comparison to available studies, the current results agree with the notion that facet joint space increases when standing, suggesting DTS facet gap measurement may be a clinically accessible and cost-effective alternative to dynamic in vivo stereo-radiogrammetry.

Several limitations in our study must be considered. First, analyses were performed on a single level (L1/2) opportunistically identified from an ongoing study. It is reported that the facet orientation varies considerably within the spine, with lower lumbar levels more aligned with the coronal plane (Masharawi et al., 2004), and significant differences have been described for segment-specific facet joint kinematics (Byrne et al., 2018). The extent to which the observed changes in facet joint distance for L1/2 due to standing is applicable to other levels therefore remains unknown. Second, although manual contouring by an expert analyst is typically considered the reference standard for evaluating success of image segmentation (Buie et al., 2007), the method is quite time consuming (approximately 2 hours per segment) and requires a trained eye to produce a visually accurate segmentation mask given the blur from objects outside the plane of interest inherent to tomosynthesis imaging (Kim et al., 2018). Third, Due to the nature of tomosynthesis imaging, global thresholding approaches are not always successful (Kim et al., 2017; Vikhe and Thool, 2016). In future work, alternative image pre-processing schemes or AI segmentation methods could be evaluated to improve efficiency of the method. It must be recognized that our sample size was small and findings relating baseline space and changes due to standing should be confirmed in a larger sample. It should also be noted that the measured gap changes were very small. However, the changes due to standing were, on average, more than 20-fold greater than the level of accuracy error. Although changes due to standing were large relative to accuracy error, they were small relative to the limits of agreement in differences between DTS and μCT. In future work, a larger sample for evaluation of measurement agreement could lead to narrower limits of agreement due to a more precise estimate of the standard deviation of differences, improving the reliability of conclusions regarding potential measurement bias. Finally, it should be noted that the distance measurements considered in the current study are an aggregate measure derived from the field of displacements across the entire joint surface. Although differences due to physiologic loading appear detectable for these measures, it is possible that accuracy of DTS relative to μCT may vary by subregions in the joint space field. In future work we will explore the reliability of inter-modality registration to evaluate accuracy of DTS relative to μCT on a subregional basis.

In conclusion, our study results support the feasibility of a method for direct measurement of facet joint space in supine and standing positions using digital tomosynthesis images. Facet joint space measurements were accurate relative to reference values from micro-CT and changes in facet joint space due to loading were detectable. As a pilot in vivo demonstration, changes in facet joint space due to physiological loading were found to be detectable. This was an opportunistic proof of concept study using normal patient images from a different study. Future work will apply this technique to assess changes in facet joint space for patients with lower back pain and spondylolisthesis and include lower lumbar levels.