Comprehensive Grading System of Inflammatory Features of the Lumbar Facet Joints on Magnetic Resonance Imaging: Reliability

Abstract

Study design:

Retrospective.

Objective:

We aimed to describe a magnetic resonance imaging (MRI)-based grading system of inflammatory features of the lumbar facet joints using an atlas and assess its reliability.

Summary of Background Data:

Chronic low back pain is often caused by facet joint arthropathy. Inflammatory features are often evident on MRI. While several grading systems of facet arthropathy have been described, there is scant data on the reliability of these systems, and none focus exclusively on inflammatory features.

Methods:

We describe a grading system that assesses facet joint effusion, bone marrow edema, and soft tissue edema. Each feature was graded from 0–3 (facet edema) or 0–2 (bone marrow edema intensity and extent, soft tissue edema intensity and extent). Four spine experts graded MRIs of 50 subjects at the bilateral L3/4, L4/5, and L5/S1 levels. All subjects had symptomatic facet arthropathy and received therapeutic facet joint injections. We assessed the intra- and inter-reader reliability of each feature at each joint and summarized across all six joints.

Results:

The mean age of subjects was 56 years (SD = 17), and 48% were female. The injections occurred at the L3/4 level in 12% of cases, at L4/5 in 88%, and at L5/S1 in 80% of cases. The intra-reader reliability kappa’s for each feature ranged from 0.42 to 0.81. In contrast, the inter-reader reliability kappa values for each feature ranged from 0.37 to 0.54.

Conclusion:

MRI inflammatory features of the lumbar facet joints are often noted in patients with low back pain. The proposed grading system is reliable and could serve as a research tool for studies assessing the clinical relevance and prognostic value of these features.

INTRODUCTION

Low back pain is the leading cause of disability worldwide¹, with most adults developing an episode of low back pain in their lifetime.² The estimated annual cost of low back pain in the United States is $100 billion.³ Facet joint arthropathy is among the most frequently observed anatomical abnormalities in patients with axial, non-radicular back pain.⁴

Facet joints are synovial joints in the spine and commonly develop degenerative changes.⁵ Estimates of the prevalence of facet arthropathy in patients with chronic low back pain range from 15 to 67%.^4-5;7–9 However, the clinical significance of facet arthropathy identified on imaging is sometimes uncertain, as facet arthropathy is observed on computed tomography (CT) imaging in 37% of asymptomatic individuals.⁶ Therefore, it is crucial to identify which imaging findings are most highly correlated with symptoms. Imaging findings of facet arthropathy include morphologic abnormalities of the facet joints, such as hypertrophy of the articular process, subarticular bone erosions, and osteophytes. Additionally, peri-facet signal abnormalities evident on Magnetic Resonance Imaging (MRI) include facet joint effusion, bone marrow edema, and soft tissue edema, which are present in 14 to 70% of patients with low back pain.^{9; 14; 15–17} We describe these imaging findings as markers of inflammatory features, though the extent and type of inflammation are not entirely understood. In a similar fashion, bone marrow edema and facet joint effusion are often described as markers of inflammation in patients with knee osteoarthritis.^18–19 These peri-facet signal changes are reported to correlate with the side of pain and the injected spine level.^9;16

Several grading systems have been developed to quantify lumbar facet joint morphological and inflammatory changes.^9–14 However, there is scant data on the reliability of these measures, and existing grading systems include overlapping features.²⁰ Here, we propose an MRI-grading system of facet joint inflammatory changes that relies on frequently described inflammatory parameters, including facet joint effusion, bone marrow edema, and soft tissue edema. In addition, we propose an atlas with examples to provide visual guidance to grade features of facet joint inflammation for further research and potentially serve as a predictive tool in clinical settings.

We describe an MRI-based grading system of inflammatory features of the lumbar facet joints using an atlas with examples. This grading system showed moderate inter-reader reliability when used by experienced clinicians to interpret lumbar spine MRI studies in subjects with low back pain.

MATERIALS AND METHODS

Development of the grading system

The grading system was developed as a multidisciplinary collaboration among a musculoskeletal radiologist with eight years of experience, a musculoskeletal radiologist fellow, two rheumatologists with 33 and 13 years of experience, one interventional spine and pain management physiatrist with 20 years of experience, one orthopedic surgeon with 11 years of experience, a resident in physiatry and a fourth-year medical student.

The grading system captures the severity of facet joint effusion, bone marrow edema, and peri-facet soft tissue edema. We chose rubrics for grading these features based on prior work^9–14, with some modifications: we defined facet joint effusion using criteria proposed by Longmuir et al. and Rodriguez et al.^11;14; bone marrow using criteria proposed by Friedrich et al. and Lehman et al.^10;13; and peri-facet soft tissue edema using measures proposed by Czervionke et al. and Lehman et.⁹⁻¹⁰

To create the atlas, we searched the image archives at our institution to identify representative MR images for each grade of severity for each feature. The multidisciplinary research team developed brief descriptions of the severity levels of each feature and created schematic diagrams with an iterative approach. Representative MRI example images and schematics were combined into a Microsoft PowerPoint file (Microsoft Corp., Redmond, WA, USA).

The grading system

Table 1 summarizes the descriptors for the features included in the grading system. Table 1 and the Supplemental File provide examples of MRIs with graded inflammatory features.

Table 1.

Lumbar facet inflammatory changes grading system

Inflammatory feature	Atlas description	Points assigned	Corresponding Image in Supplemental File#
Facet joint effusion (FJE)			2
None	No fluid signal is evident. The articular surfaces of the inferior and superior articular processes are in direct opposition.	0
Mild	A thin bright line on T2-weighted images in either or both axial or sagittal planes. Less than or equal to the thickness of subchondral bone.	1	3 and 4
Moderate	A prominent bright line on T2-weighted images in either or both axial or sagittal planes. Greater in thickness than subchondral bone.	2	5 and 6
Severe	There is gapping of the articular surfaces of the inferior and superior articular processes, which are no longer in direct apposition. Fluid may extend into and distend the posterior joint capsule.	3	7 and 8
Bone marrow edema (BME) – intensity
None	Normal	0
BME intensity - mild	Mild edema-like signa, with T1-weighted images remaining normal.	1	10 and 11
BME intensity - marked	Pronounced edema-like signal, either near fluid signal intensity or evident on T1-weighted images.	2	12 and 13
Bone marrow edema - extent
None	Normal	0
BME extent- mild	Involving the superior or inferior articular processes only, measuring less than one square centimeter.	1	10 and 11
BME extent- marked	Extending greater than 50% through the pedicle or measuring more than 1 square centimeter.	2	12 and 13
Soft tissue edema (STE) - intensity
None	Normal	0
STE intensity- mild	Mild edema-like signal may be wispy in appearance.	1	15 and 16
STE intensity-marked	Pronounced edema-like signal near fluid signal intensity, may appear similar to discrete fluid.	2	17 and 18
Soft tissue edema – extent
None	Normal	0
STE extent- mild	Remains confined to the facet level (e.g., for L4-L5, does not extend to the L5-S1 facet).	1	15 and 16
STE extent- marked	Extends beyond the level of the facet.	2	17 and 18

^#= The numbers in this column refer to the Supplemental File PowerPoint slide numbers.

Facet joint effusion (FJE)

FJE is assessed on sagittal Short Tau Inversion Recovery (STIR) and axial T2-weighted MRI. Fluid appears bright on T2-weighted MRI with fat suppression, while subchondral bone appears dark on all pulse sequences. The severity is graded based on the width of the fluid signal, as shown in Table 1. In the case of a discrepancy between the grades for sagittal and axial images, the highest grade is used.

Bone marrow edema (BME)

BME is characterized by a bright signal within bones on T2-weighted images with fat suppression (e.g., STIR sequences). BME may involve the inferior or superior articular processes or the pedicles. Signal intensity and extent are graded independently.

Soft tissue edema (STE)

STE is characterized by a bright signal in soft tissue structures surrounding the facet joints on T2-weighted images with fat suppression (e.g., STIR). Signal intensity and extent are graded independently.

Pilot Testing and Calibration Session

After developing our grading system with verbatim descriptors and examples of each MRI feature, we conducted pilot testing. Each team member reviewed the atlas examples and definitions, and as a team, we reviewed several MRIs to discuss distinctions between severity grades. We then performed a calibration session in which team members independently graded five MRIs. We examined the extent of agreement for each feature. We discussed discrepancies and performed a similar calibration session on ten additional MRI cases.

We developed a REDCap (Research Electronic Data Capture) database to record the readers’ grades. Readers practiced grading with the REDCap database before reading the cases for the reliability study.²¹

Data Source for reliability sample

We identified cases for the reliability sample from the institution’s Research Patient Data Repository (RPDR), a clinical data registry that gathers medical record data from the healthcare system.

Patient selection

With approval from the Mass General Brigham (MGB) Institutional Review Board (IRB, protocol number: 2022P001022), we used the hospital data system (RDPR). This clinical data registry gathers medical record data from the healthcare system to identify cases for the reliability sample.²² We searched for patients older than eighteen who had a lumbar facet joint injection performed at Brigham and Women’s Hospital or Brigham and Women’s Faulkner Hospital (both in Boston, MA, USA) and had a prior lumbosacral spine MRI at a Mass General Brigham-affiliated site.

We excluded subjects for the following reasons: history of cancer or metastasis localized in the spine, lumbar spine surgery or trauma, use of intravenous steroid injection within six months before the intervention (given that steroid injections provide short-term pain relief and reduce inflammatory mediators) ^23–24, spine congenital anomalies (e.g., spina bifida or spinal dysraphism), cauda equina syndrome, tumor located in the lumbar spine, discitis or osteomyelitis in the lumbar spine, inflammatory conditions (e.g., rheumatoid arthritis), osteoporosis noted on MRI reports, multiple sclerosis, no STIR sequence available, MRI performed outside of the Mass General Brigham hospital network, or poor image quality. Finally, the musculoskeletal radiologist reviewed the cases to ensure image quality before starting the reliability study.

All imaging studies were clinical exams performed on several MRI scanners and locations across the Mass General Brigham Hospital network using a standard clinical protocol at either 1.5 or 3.0 Tesla. While there was some variability in protocols across scanners, all studies included axial T1 and T2-weighted, sagittal T1, T2, and short tau inversion recovery (STIR) sequences. Axial images were performed with an “upper” and “lower” stack to remain in the plane, with the discs accounting for lumbar curvature. All exams were performed without intravenous contrast. Fifty cases were selected for the reliability study.

Reader sessions and classification

The four authors in the reliability study were experienced in interpreting spine MRIs. They included the musculoskeletal radiologist, the musculoskeletal radiology fellow, the orthopedic surgeon, and the spine physiatrist. Two authors graded half of the cases twice to measure intra-reader reliability. All readers were provided with the atlas when grading images. All the reading sessions of the reliability study were performed independently on de-identified lumbar spine MRIs on a secure Web-based picture archiving and communication system.²⁵ Readers were blinded to any identifying patient health data. For each case, the readers assessed the level of inflammation of the facet joints at the L3-L4, L4-L5, and L5-S1 levels. They provided separate grades for FJE, BME, and STE of the right and left facet joints at each spine level. The results were tabulated with a multiple-choice interface in REDCap and then transferred to SAS (Version 9.4, SAS Institute, Cary, NC, USA) for statistical analysis.

Statistical Analysis

Each feature was assessed on six facet joints for each subject (right and left facets at L3–4, L4–5, and L5-S1). We created a subject-level score for each feature by summing the ratings of all six joints for each feature. We also created a total score by adding the subject-level scores for each feature across all five features.

We assessed intra-reader agreement for two readers who read 25 cases twice, separated by about three weeks. We calculated (1) the proportion of these 150 pairs (25 subjects × 2 sides × 3 joints) that agreed exactly and the proportion that agreed within one grade for each feature and the total score, (2) weighted kappa coefficient (≤ 0.20: slight agreement, 0.21–0.40: fair agreement, 0.41–0.60: moderate agreement, 0.61–0.80: substantial agreement, 0.81–1.0: almost perfect agreement²⁶) for each feature, (3) the intraclass correlation coefficient (ICC) for the combined score for each feature and (4) the ICC for the total score for each subject.

We used a similar strategy to calculate inter-reader agreements. Four readers graded 50 subjects. We calculated (1) the proportion with exact agreement and agreement within one grade for the 300 ratings (50 subjects × 2 sides × 3 joints) across the six pairs of readers for each feature and the total score, (2) the pairwise kappa for the 300 ratings across the six pairs of readers for each feature, (3) the ICC for the combined score for each feature, and (4) the ICC for the total score for each subject.

We chose the sample size of 50 subjects (300 distinct joints) to provide reasonable precision, as reflected in the 95% confidence interval around the ICC estimates.

RESULTS

Screening process

Figure 1 shows the screening process of the subjects included in the initial RPDR search. The initial search from RDPR included 380 subjects. Nine of these subjects had died and were excluded, leaving 371 subjects. We reviewed 153 records and excluded 103 that met at least one of our exclusion criteria, leaving us with 50 subjects.

Figure 1. Diagram of the screening process of the included subjects.

^= We screened subjects sequentially using our exclusion criteria until we included at least 50 subjects.

Demographics

Table 2 summarizes the demographic information of our cohort. The mean age was 56 years (SD = 17), and 48% of subjects were female. The injection occurred at the L3/4 level in 12% of the cases, at L4/5 in 88%, and at L5/S1 in 80% of the cases. However, multiple levels were injected in many cases, and 56% were injected in bilateral facet joints.

Table 2.

Patient cohort demographics

Characteristics	Values (n = 50)
Mean age, years (SD)	56 (16.7)
Sex
Female	24
Male	26
Race
White	45
Black	0
Asian	2
American Indian/Alaskan Native	0
Hawaiian/Pacific Islander	0
Other	1
Unknown	2
Ethnicity
Hispanic	0
Not Hispanic	39
Unknown	11
Level of injection*
L2/L3	2
L3/L4	6
L4/L5	44
L5/S1	40
Side of injection**
Right	13
Left	9
Bilateral	28

^*The sum of patients with injections at different lumbar spine levels does not add up to the total number of participants included in this exercise (n = 50) because some patients had injections at multiple levels of the spine.

^**The sum of patients with injections at different lumbar spine levels does not add up to the total number of participants included in this exercise (n = 50) because some patients had injections at the right, left, or at bilateral facet joints.

Frequency of inflammatory features

Table 3 summarizes the grade distribution for each inflammatory measure noted by the four readers. Most facet joint effusion ratings were grade 0 (68.9%), followed by grade 1 (28.4%). Bone marrow edema intensity and extent and soft tissue edema intensity and extent were also frequently graded as grade 0 (93.4% and 93.4%, 86.7%, and 86.7%, respectively). The total score (i.e., the sum of the inflammation scores) ranges from 0 to 9 for each facet joint, with 89.3% graded from 0–2.

Table 3.

Distribution of frequency of the inflammatory features and total inflammation score at the joint level

Feature	Grade	N (Total N=1,200α)	%
Facet joint effusion	0	827	68.9
	1	341	28.4
	2	20	1.7
	3	12	1.0
Bone marrow edema – intensity	0	1121	93.4
	1	69	5.8
	2	10	0.8
Bone marrow edema – extent	0	1121	93.4
	1	53	4.4
	2	26	2.2
Soft tissue edema – intensity	0	1040	86.7
	1	142	11.8
	2	18	1.5
Soft tissue edema – extent	0	1040	86.7
	1	134	11.2
	2	26	2.2
Total score	0–2	1072	89.3
	3–4	93	7.8
	≥5	35	2.9

^αThe distribution of the features was calculated based on the total number of observations. (i.e., Right, and left facet joints for L3/4, L4/5, and L5/S1 levels across 50 subjects. Four raters graded each.)

Intra-reader agreement

Table 4 summarizes the results of the intra-reader agreement scores. Exact agreement in ratings by the senior musculoskeletal radiologist for the inflammatory features ranged from 88.7% (facet joint effusion) to 95.3% (soft tissue edema extent), with kappa statistics ranging from 0.52 (bone marrow edema intensity) to 0.78 (soft tissue edema extent). The percentage of exact agreement for the total scores was 76.7%. The agreement within one grade of difference was 100.0% for all the features and reached 88.7% for the total inflammation score. The ICC for the combined score from the six joints ranged from 0.65 to 0.84, with the highest value for bone marrow edema extent, and the ICC for the total score was 0.72.

Table 4.

Intra-rater reliability assessed by agreement, weighted Kappa, and Intraclass Correlation Coefficient (ICC) values for each inflammatory feature, and the total score of inflammation summed across the right and left facets at the joint level of L3/4, L4/5, and L5/S1

	Reviewer
	Reader #1 *				Reader #2 *
	Exact agreement	Agreement within 1	Kappa (95% CI)	ICC (95% CI)	Exact Agreement	Agreement within 1	Kappa (95% CI)	ICC (95%CI)
Facet joint effusion	88.7%	100.0%	0.72 (0.61–0.83)	0.65 (0.46–0.78)	78.7%	100.0%	0.56 (0.43–0.69)	0.79 (0.66–0.87)
Bone marrow edema – intensity	91.3%	100.0%	0.52 (0.29–0.74)	0.72 (0.55–0.83)	96.0%	99.3%	0.71 (0.49–0.92)	0.72 (0.56–0.83)
Bone marrow edema – extent	92.0%	100.0%	0.62 (0.40–0.85)	0.84 (0.73–0.90)	96.7%	100.0%	0.81 (0.63–0.98)	0.82 (0.71–0.89)
Soft tissue edema – intensity	94.7%	100.0%	0.74 (0.57–0.92)	0.73 (0.57–0.83)	94.7%	98.7%	0.42 (0.13–0.71)	0.69 (0.52–0.81)
Soft tissue edema – extent	95.3%	100.0%	0.78 (0.62–0.95)	0.70 (0.53–0.82)	94.7%	98.7%	0.42 (o.13–0.71)	0.69 (0.52–0.81)
Total score	76.7%	88.7%	-	0.72 (0.56–0.83)	73.3%	93.3%	-	0.76 (0.62–0.86)

^*Reader 1 refers to a senior musculoskeletal radiologist attending, while reader 2 refers to a musculoskeletal radiology fellow.

^**% with agreement and kappa were evaluated using the 150 joints from 25 subjects, while ICCs were evaluated using the sum of the six joints from 25 subjects

For the gradings of the musculoskeletal radiology fellow, the proportion of exact agreement ranged from 78.7% (facet joint effusion) to 96.7% (bone marrow edema extent), with kappa statistics ranging from 0.42 (soft tissue edema extent or intensity) to 0.81 (bone marrow edema extent). The percentage of exact agreement for the total scores was 73.3%. The agreement within one grade of difference ranged from 98.7% to 100.0% and 93.3% for the total inflammation score. The ICC for the combined score from the six joints ranged from 0.69 to 0.82, with the highest value for bone marrow edema extent, and the ICC for the total score was 0.76.

Inter-reader reliability

Table 5 summarizes the results of the inter-reader agreement scores. The exact agreement among the readers ranged from 75.1% (facet joint effusion) to 93.0% (bone marrow edema extent), and the agreement within one grade of difference ranged from 98.2% to 99.7% and 79.9% for the total inflammation score. The kappa statistics ranged from 0.37 to 0.54, showing moderate agreement in general, with the highest kappa for bone marrow edema extent. The ICC for the combined score ranged from 0.39 to 0.74, with the highest ICC for bone marrow edema extent, and the ICC for the total score was 0.63.

Table 5.

Inter-rater reliability assessed by agreement, weighted Kappa, and Intraclass Correlation Coefficient (ICC) values for each inflammatory feature and the total score of inflammation summed across the right and left facets at the joint level of L3/4, L4/5, and L5/S1

	Exact agreement***	Agreement within 1	Kappa (95% CI)	ICC (95% CI)
Facet joint effusion	75.1%	98.2%	0.45 (0.22–0.69)	0.48 (0.24–0.66)
Bone marrow edema – intensity	92.9%	99.7%	0.47 (0.23–0.71)	0.66 (0.47–0.79)
Bone marrow edema – extent	93.0%	99.3%	0.54 (0.29–0.78)	0.74 (0.59–0.84)
Soft tissue edema – intensity	82.7%	99.6%	0.37 (−0.05–0.79)	0.46 (0.21–0.65)
Soft tissue edema – extent	82.6%	98.5%	0.37 (−0.12–0.86)	0.39 (0.14–0.60)
Total score	59.6%	79.9%	-	0.63 (0.44–0.77)

^***= This exercise included the readings of four clinicians experienced in interpreting spine Magnetic Resonance Imaging. It included the musculoskeletal radiologist, the musculoskeletal radiology fellow, the orthopedic surgeon, and the spine physiatrist.

The percentage with agreement and kappa were evaluated using the 300 joints from 50 subjects, while ICCs were evaluated using the sum of the six joints from 50 subjects

DISCUSSION

The current study summarizes the multidisciplinary development and reports on the reliability of a comprehensive MRI grading system for the most frequently described inflammatory features of the lumbar facet joints, including facet joint effusion, bone marrow edema, and soft tissue edema. The inter-reader agreement between four expert readers was generally moderate for the three features, and the intra-reader agreement was almost perfect.

Of the individual inflammatory features, the extent of bone marrow edema showed the highest agreement among and within readers. Similarly, Friedrich et al. developed a grading system that categorizes the severity of bone marrow edema and found an inter-reader agreement kappa value of 0.81.¹³ Conversely, Czervionke et al. assessed a grading system for soft tissue edema of the facet joints and found an inter-reader agreement kappa value of 0.41.⁹ Rodriguez et al. described a grading system for facet joint effusion and found a kappa value ranging from 0.87–0.98.¹⁴ Our findings parallel those found in studies assessing inflammatory features of knee osteoarthritis. For example, Hunter et al. found an inter-reader kappa of 0.54 for tibial bone marrow edema lesions, the same value we found for the extent of bone marrow edema in the facet joint of the lumbar spine.²⁷ Our findings also reveal similar reliability of MRI assessment of other spine pathologies, such as spinal stenosis. For example, Schizas et al. assessed a qualitative grading system of the severity of lumbar spinal stenosis and found an intra- and inter-reader agreement of 0.65 and 0.44, respectively.²⁸

These inflammatory changes may be transient and presumably antedate structural damage. For example, bone marrow edema is often described in the literature as a marker of inflammation related to the progression of knee osteoarthritis.^18;29–30 Other studies suggest knee joint effusion synovitis may result from joint structural abnormalities in established osteoarthritis.¹⁹ Another study found that the amount of bone marrow edema of the hip as measured on MRI correlates with the severity of pain, radiographic findings, and the number of microfractures.³¹

We found a moderate-to-substantial intra-reader agreement and a fair-to-moderate inter-reader agreement for the individual inflammatory features. The within and between-reader agreement within one grade was almost perfect for all the features. The extent of bone marrow edema had the highest agreement within and between readers.

We acknowledge several limitations in our study. First, our sample is mostly comprised of non-Hispanic Whites, limiting our generalizability. Second, the images evaluated were taken from different scanners and hospitals and read in various environments (with different screen sizes and lighting conditions). High-field MRI, like 3.0T, provides a better signal-to-noise ratio (SNR), improving image quality.^32–34 Thus, variations in magnet strength across scanners used in this study may have influenced inter- and intra-reader reliability.³⁵ We also note that reliability statistics are influenced by the range of severity in the sample. Many of the subjects in our sample had relatively little pathology, which may have led to lower kappa’s and ICCs than would be expected in a sample with greater variability in imaging scores.³⁶ Lastly, the small sample for the intra-rater reliability may not be a representative sample. Strengths of the study include the first grading system compiling the most frequently described inflammatory features of the lumbar facet joints. Also, our results show that this system is reliable across multiple levels of experience and specialties. These features are frequently noted in patients with low back pain, and the system may serve as a predictive tool in clinical settings.

In conclusion, this study describes a reliable grading system for facet joint inflammatory features on lumbar spine MRI that may serve as a research tool for future studies evaluating the diagnostic and prognostic value of these inflammatory features. Future studies could also consider using artificial intelligence to improve the quality of these results.