MRI for axial SpA: Diagnosis, disease activity assessment, and recent advances

Magnetic resonance imaging (MRI) is a sensitive imaging modality to detect early inflammatory changes in axial spondyloarthritis (SpA). Over a decade has passed since the inclusion of MRI assessment in the 2009 Assessment of SpondyloArthritis International Society (ASAS) classification criteria for axial SpA. Evidence and clinical experience of MRI in axial SpA have accumulated rapidly since. This has led to a better understanding of the clinical utility of MRI in early diagnosis, disease activity assessment, and monitoring of treatment response in axial SpA. Furthermore, technological advancements have paved the way for the development of novel MRI sequences for the quantification of inflammation and image optimization. The field of artificial intelligence has also been explored to aid medical imaging interpretation, including MRI in axial SpA. This review serves to provide an update on the latest understanding of the evolving roles of MRI in axial SpA.

review reported a diagnostic delay ranging from 0.67 to 8 years in axial SpA. 5 With its ability to detect inflammation before the development of structural lesions, MRI has enabled the early diagnosis of axial SpA.
Over the years, different sets of classification criteria have been proposed, including the European Spondyloarthropathy Study Group and Amor criteria. 6,7Classification criteria were designed to select homogenous patients for research purposes and should only be applied to patients with an established diagnosis of axial SpA.The Assessment of SpondyloArthritis International Society (ASAS) classification criteria for axial SpA were developed in 2009 and included a clinical arm and an imaging arm. 8Patients are classified into the "imaging arm" in the presence of sacroiliitis by radiography or by MRI plus at least one SpA feature, or the "clinical arm" in the presence of HLA-B27 plus at least two SpA features.According to the ASAS criteria, positive MRI sacroiliitis was defined as (1) the presence of definite bone marrow edema/osteitis highly suggestive of sacroiliitis, (2) located in the typical anatomical areas (subchondral or periarticular bone marrow), and (3)   meeting the required amount of signal (at least two consecutive slices if only one signal per MRI slice; one slice may be sufficient if more than one signal per slice). 9The overall performance of the imaging arm (±clinical arm) of the ASAS criteria for axial SpA was reported to have a pooled sensitivity of 57% and a specificity of 96%. 10 A study based on longitudinal data from the Devenir des Spondyloarthropathies Indifférenciées Récentes (DESIR) cohort (a large national multicenter cohort of patients with early inflammatory back pain of more than 3 months and less than 3 years) demonstrated excellent predictive validity of the ASAS criteria at baseline assessment when tested against the rheumatologist's diagnosis of axial SpA. 11The results highlight the utility of the ASAS criteria among patients with early disease onset.
MRI also serves as one of the imaging modalities in juvenile SpA. 12 Similar to adults with SpA, MRI can detect sacroiliitis in juvenile SpA before changes are present in radiographs. 13In a prospective cross-sectional study of 40 children with newly diagnosed juvenile SpA, active MRI sacroiliitis was found in 20% of patients, and most cases were asymptomatic. 14Gadolinium contrast may be considered in selected patients to detect synovitis, which may occur in the absence of bone marrow edema, after weighing against possible contrast-related side effects such as injection-site discomfort and hypersensitivity. 15More research is needed to define and standardize the best approach to imaging and the role of MRI in children with juvenile SpA. 16

| MRI interpretation in SpA and mimickers of sacroiliitis
The introduction of MRI has improved early diagnosis of axial SpA.However, a substantial proportion of healthy individuals may demonstrate bone marrow edema lesions on MRI.Using the Spondyloarthritis Research Consortium of Canada (SPARCC) score of ≥2 as the cutoff, one study showed that positive MRI sacroiliitis was found in 57.1% of postpartum women, 16.7% of runners, and 25.5% of healthy individuals. 17Mechanical stress on the sacroiliac joint has been proposed as an underlying mechanism, but it does not fully account for the development of bone marrow edema. 18The extent and anatomical distribution of bone marrow edema lesions may provide useful information.Deep bone marrow edema lesions (defined as a homogenous, unequivocal increase in signal at least 1 cm from the articular surface) were more specific among patients with axial SpA (Figure 1). 17Furthermore, bone marrow edema lesions occur more frequently in the intermediate-middle segment in axial SpA and predominantly affect the antero-middle segment in non-SpA participants (Figure 2).Bilateral bone marrow edema lesions are more common in SpA compared with non-SpA, except for women with postpartum back pain. 19Given the high prevalence of non-specific bone marrow lesions among non-SpA patients, MRI should be interpreted based on the contexts of individual patients, considering various clinical and serological features, to avoid overdiagnosis of axial SpA. 20

| Utility of sacroiliac joint structural lesions and spinal MRI in axial SpA
In addition to bone marrow edema, different MRI inflammatory and structural lesions of the sacroiliac joint have been defined. 8Structural F I G U R E 1 Bilateral sacroiliitis on short tau inversion recovery magenetic resonance imaging.Deep bone marrow edema lesions (defined as a homogenous, unequivocal increase in signal at least 1 cm from the articular surface) are shown in bilateral sacroiliac joints (white arrows).

F I G U R E 2
The intermediate-middle segment of the sacroiliac joint that is predominantly affected in axial spondyloarthritis.lesions in the sacroiliac joint include erosions, periarticular fat depositions, and ankylosis.Several studies have examined the role of structural lesions in the classification of axial SpA.A proposed cutoff of erosions ≥3, fatty lesions ≥3, or erosions/fatty lesions ≥5 (based on <5% prevalence among non-SpA patients) was evaluated in previous studies and demonstrated good performance either as an addition to or as a substitute for radiographs in the ASAS axial SpA classification 21 ; further studies are needed to confirm its usefulness in clinical practice.A recent study carried out by the ASAS MRI group proposed data-driven definitions of active and structural lesions of the sacroiliac joints in SpA, defined as achieving a positive predictive value of ≥95%. 22Either bone marrow edema in ≥4 quadrants at any location or at the same location in ≥3 consecutive slices contributed to the best cutoffs for a definite active lesion typical of axial SpA.The optimal cutoffs for definite structural lesion included ≥3 quadrants with erosion or ≥5 with fat lesions, erosion at the same location for ≥2 consecutive slices, fat lesions at the same location for ≥3 consecutive slices, or the presence of a deep (defined as >1 cm depth) fat lesion.
Since the inflammatory process in axial SpA does not limit itself to the sacroiliac joint and may affect different spinal segments, several studies have evaluated the diagnostic utility of spinal MRI in axial SpA.The presence of ≥3 corner inflammatory lesions (CILs) has been proposed as the definition of a positive spinal MRI for axial SpA by the ASAS/Outcome Measures in Rheumatology working group. 23rlier studies have shown that when used alone, whole-spine CILs have poor diagnostic utility. 24Spinal MRI also adds little incremental diagnostic value when added to sacroiliac joint MRI assessment. 24A study has shown that among patients without sacroiliitis on MRI or radiographs, 8%-13% might be diagnosed by spinal MRI. 25 Thoracic and whole spine MRI had similar diagnostic performance using the proposed cutoff of ≥5 whole spine CILs and ≥3 thoracic spine CILs. 25 In a study of 238 patients with axial SpA, it was shown that ≥3 fatty corner lesions (FCLs) in T1-weighted MRI could be used for diagnosis without additional MRI sacroiliac joints. 26

| Correlation between clinical disease activity index and MRI inflammation
Currently, there is limited reliable biomarker for axial SpA.HLA-B27 is a useful marker and predicts poor prognosis, but it is not specific for axial SpA.Clinical disease activity indices, such as the Bath Ankylosing Spondylitis Disease Activity Index (BASDAI) and the Ankylosing Spondylitis Disease Activity Score (ASDAS), are commonly used to evaluate disease activity in routine clinical practice in patients with axial SpA. 27,28BASDAI is a self-score index that includes fatigue, pain (in axial and peripheral joints), swelling, and the level and duration of morning stiffness.The ASDAS is a composite index that is also based on a similar set of subjective symptoms, including back pain, duration of morning stiffness, patient global assessment, and peripheral pain and swelling.In addition to selfrated parameters, ASDAS also includes serum inflammatory markers, namely C-reactive protein (CRP) or erythrocyte sedimentation rate (ESR), which serve as an objective measure of inflammation.
Inflammation on MRI has been shown to be correlated with inflammatory cellularity documented by tissue biopsy. 29Therefore, researchers are using MRI as a "gold standard" for comparison.Traditional inflammatory markers, including CRP and ESR, correlate poorly with MRI-detected inflammation. 30Many studies have found conflicting results regarding the correlation between clinical disease activity indices and MRI inflammation.Two-year data from the DESIR cohort demonstrated a significant association between MRI inflammation in the sacroiliac joint and ASDAS in male patients but not in female patients. 31In a study of 149 patients with axial SpA, Byravan et al. 32 reported a significant correlation between active sacroiliitis and BASDAI.However, the study showed no correlation between BASDAI and other MRI features including, chronic sacroiliitis, active axial disease, and other MRI changes (costovertebral, costotransverse, and sternoclavicular inflammation).Another study showed that ASDAS was associated with both the extent and intensity of spinal inflammation in patients with detectable MRI inflammation only. 33An MRI of the SI joint and/or the spine may be considered based on the physician's discretion to complement clinical and biochemical assessment at the initial stage and to provide an objective assessment of activity for disease monitoring. 34,35

| MRI inflammation as predictors for treatment response to biological therapies
MRI also plays an important role in treatment decisions.Among patients with a rheumatologist's diagnosis of axial SpA, biological therapies are recommended for those with persistently high disease activity despite conventional treatments. 36Given the treatment costs and potential side effects, careful selection of patients who would benefit most from biological therapies remains important.Studies based on data from randomized controlled trials have shown that the presence of SI and spinal inflammation on MRI was associated with a higher likelihood of treatment response with TNF inhibitors. 37Other predictive factors of response to TNF inhibitors included shorter disease duration, HLA-B27 positivity, and raised CRP.Similar findings have been shown with anti-IL17 treatment. 38sed on the evidence, it is recommended that patients considered for biological therapies should have elevated CRP, the presence of inflammation on the MRI SI joint, or radiographic sacroiliitis at baseline before treatment initiation.MRI also serves as an objective assessment of treatment response.A recent systematic review and meta-analysis evaluated the differences in MRI changes in SpA patients treated with anti-TNF agents compared with patients without biological treatment. 39sed on the SPARCC score, a significant difference in reduction of MRI sacroiliac joint inflammation among patients treated with TNF inhibitors was found.A tendency toward reduction of MRI spinal inflammation was also observed among patients treated with TNF inhibitors, even though it did not reach statistical significance.Furthermore, reductions in MRI inflammation demonstrated a correlation with improvements in clinical disease activity indices in most of these studies.

| Potential roles of MRI in determining treatment targets in SpA
A treat-to-target (T2T) approach has gained importance in the management of different rheumatological conditions.Treatment goals usually focus on disease remission (or low disease activity), prevention of disease complications and treatment-related side effects, and to improve patient's functional outcomes.In rheumatoid arthritis, a Boolean-based and index-based remission definitions have been adopted in the T2T management. 40Compared with RA, a T2T treatment strategy in SpA is less extensively evaluated.A set of recommendations for treatment targets in axial SpA and peripheral SpA was formulated in 2017. 41Clinical remission or inactive disease (defined as ASDAS <1.3 in axial SpA) was recommended as the treatment target.
Low or minimal disease activity (defined as ASDAS <2.1) may be an alternative treatment target.However, a randomized trial evaluating the efficacy of T2T approach in axial SpA failed to demonstrate a significant difference in the primary endpoint between the T2T arm and the usual care arm. 42Treatment intensification and biologic prescriptions were more frequent in the T2T arm.
Since MRI provides a valuable objective measure of inflammation and predicts treatment response to biological therapies, it remains to be answered whether there is a role for MRI assessment in determining treatment target in axial SpA.The frequency of clinical and MRI remission was compared based on the 3-year results of the ABILITY-1 trial, a randomized controlled trial evaluating adalimumab in non-radiographic axial SpA. 43Among patients with positive MRI who achieved ASAS inactive disease, only 40% and 44% achieved MRI remission (both in the sacroiliac joint and spine) at years 1 and 2, respectively.A study by Zheng et al. 44 evaluated the frequency and predictive value of the flare of MRIactive sacroiliitis among patients in clinical remission.Active MRI sacroiliitis, found in 55.8% of patients in clinical remission, was associated with a significant risk of disease flare.Further studies are anticipated to evaluate the impact of persistent MR inflammation on long-term physical function or structural progression among axial SpA in clinical remission.It remains to be answered whether MRI plays a role in complementing clinical disease activity in determining treatment targets in axial SpA.

| NOVEL MRI S EQUEN CE S
Over the years, different novel MRI sequences have been evaluated, and a few sequences are highlighted in this review.[47][48] Readout-segment echo-planar imaging (EPI) and intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI) have been evaluated to improve DWI image quality. 49,507][48] Volumetric interpolated breath-hold examination (VIBE) demonstrated better sensitivity at detecting erosive damage in the sacroiliac joint. 51These imaging sequences have been mainly evaluated in research settings, and further studies are needed to delineate their roles in clinical practices.

| Diffusion-weighted imaging
DWI is a novel MRI sequence based on the random Brownian motion of water molecules.DWI has been widely applied to the evaluation of intracranial diseases, such as cerebrovascular diseases and trauma.
More recently, the application of DWI has broadened to extra-cranial diseases, including musculoskeletal conditions.Several studies have evaluated the utility of DWI in the detection of sacroiliitis and axial SpA diagnosis.A study evaluated the diagnostic utility of DWI in a study of 305 patients with back pain. 45DWI and STIR demonstrated similar sensitivity in diagnosing axSpA in early disease.Despite its comparable sensitivity to STIR, the limited resolution of DWI rendered its poorer reliability.The addition of DWI to conventional MRI of the sacroiliac joint did not improve the overall diagnostic performance except for increased specificity and interobserver agreement. 46cently, an attempt to improve the image quality of DWI was made by using readout-segment EPI over the conventional singleshot EPI.In a study of 75 patients showed that readout-segment EPI had better image quality and diagnostic confidence compared with conventional single-short EPI. 49 to quantify the extent of inflammation, including the SPARCC MRI index, the ankylosing spondylitis spine MRI score for activity, or ASspiMRI-a, and the Berlin method. 52,53However, these scoring methods focused on the extent of inflammation without considering the intensity of inflammation in individual lesions.The SPARCC method only semi-quantitatively grades the degree of inflammation.
ADC is a measure of the magnitude of the diffusivity of water molecules within tissues and has been explored to provide information on the intensity of inflammation in SpA.In a study of 243 participants with SpA, the maximum ADC of active discovertebral lesions (defined on STIR images as hyperintense bone marrow signal contiguous with the vertebral endplates with or without involvement of the vertebral corner in any central sagittal section) were shown to be associated with back pain intensity, functional impairment (measured by the Bath Ankylosing Spondylitis Functional Index), and patient global assessment. 47ADC also correlates with clinical disease activity and spinal mobility (measured by ASDAS and BASMI, respectively). 48rategies to overcome the low-spatial resolution of DWI may improve the overall applicability of ADC in disease assessment in SpA.

| Whole-body MRI
Inflammation in SpA does not limit itself to the axial skeleton and may involve other musculoskeletal structures such as peripheral joints and entheses.WBMRI has been evaluated as an imaging tool to assess the inflammatory activities of the entire musculoskeletal system.5][56] Different indices have been proposed based on WBMRI inflammation.The TA B L E 1 Summary of selected studies on the application of artificial intelligence in axSpA image interpretation.

Modalities Cohorts Summary
Bressem et al. 60 Radiograph Deep learning algorithm was developed based on 1553 radiographs from the PROOF cohort (training n = 1324; validation n = 229) and tested in an independent GESPIC cohort (n = 458).All radiographs were centrally graded by trained and calibrated readers using the modified New York criteria.
The neural network achieved an excellent performance in the detection of definite radiographic sacroiliitis, with an AUC of 0.97 and 0.94 for the validation and test datasets, respectively.
Li et al. 61 Radiograph A dataset of 5389 pelvic radiographs was used for the development of a deep learning model for diagnosing AS (defined by the modified New York criteria) and subsequently tested on an external dataset of 583 images.

| Application of artificial intelligence (AI) in MRI interpretation
Despite their important roles in diagnosis and disease monitoring in SpA, interpretation of MRI is labor intensive, requiring specific skills that take time to acquire, and variability exists even among experienced specialists. 57AI has been increasingly applied in medical imaging interpretation and used in assisting physicians in identifying specific imaging features with speed and precision (Table 1).A small study by Faleiros et al. 58 Furthermore, the apparent diffusion coefficient (ADC) derived from readout-segment EPI had better discriminatory performance between different activity states of sacroiliitis compared with single-short EPI.However, ADC values from single-short EPI and readout-segment EPI vary across vendors, field strengths, and b-values.More studies are needed to confirm the role of readout-segment EPI in quantifying inflammation in sacroiliitis.Another DWI technique, IVIM-DWI, has also been explored in SpA.IVIM-DWI is a double exponential model based on the fitting of multiple b-values.IVIM-DWI has the ability to assess the diffusion of water molecules and perfusion separately. 50IVIM-DWI allows tissue characterization based on perfusion patterns to distinguish between physiologic and pathologic conditions.In a study of 56 participants with AS, Liu et al. 50demonstrated the superiority of IVIM-DWI over traditional DWI for the identification of different AS disease activity levels with a good correlation with the SPARCC score at the SI joints.The assessment of spinal inflammation also contributes to the overall disease activity assessment.Based on conventional MRI sequences, such as short tau inversion recovery (STIR) and fat saturation sequences, different scoring methods have been developed

5 |
included 56 sacroiliac joint MRI exams and compared three different machine learning models (Support Vector Machine, Multilayer Perceptron, and Instance-Based Algorithm) in the classification of MRI-active sacroiliitis.The study demonstrated the potential of machine learning methods to identify active sacroiliitis in SpA, with the multilayer perception method achieving the best results in that study.Based on an MRI of 326 participants with SpA and 63 participants with non-specific back pain, an algorithm was based on a convolutional neural network was developed for the diagnosis of inflammatory sacroiliitis.The sensitivity and specificity of the algorithm were comparable with the interpretation by a radiologist. 59The work serves as a proof-of-concept study for further applications of AI in MRI interpretation.Validation of the AI algorithm in independent cohorts is needed.Further studies await to evaluate the potential of AI in the detection of other MRI inflammatory and structural lesions or the quantification of inflammation.Despite the overall promises of AI in image interpretations, clinicians' judgment remains indispensable in diagnosis, taking into account of the clinical context and differential diagnoses.Several important aspects remain to be addressed, including ethical and legal responsibilities linked to the use of AI in medicine.CON CLUS I ON AND FUTURE DIREC TION MRI has invaluable roles in the diagnosis and assessment of disease activity in axial SpA.It is particularly useful before the initiation of biologic therapy.Future development involves novel MRI sequences and AI applications.
Patients with Axial Spondyloarthritis: Multicountry Registry of Clinical Characteristics; SPARCC, Spondyloarthritis Research Consortium of Canada; STIR, short tau inversion recovery.