Introduction

Radiographic assessment of the sacroiliac joints (SIJs) and spine is the primary imaging modality used in the clinical assessment of patients with spondyloarthritis (SpA) and, in spite of the remarkable advances related to new modalities such as magnetic resonance imaging (MRI), remains the cornerstone of diagnosis, evaluation, and classification of SpA. It also constitutes the principal validated endpoint for the assessment of disease-modifying therapies and prognostic factors for structural damage. Interest in radiography for SpA has been enhanced by the development, validation, and application of radiographic scoring methods in clinical trials and research. At the same time there has been growing acceptance of the constraints posed by an over-reliance on radiography for the assessment of patients in clinical practice and research. Areas of controversy and recent interest in clinical practice include the optimal approach to imaging of the SIJ, the diagnostic utility of radiography, particularly in the context of the new Assessments in Spondyloarthritis International Society (ASAS) classification criteria and in relation to other imaging modalities, and the role of radiography in follow-up assessment and disease management. The use of radiography in clinical trials and research has generated substantial controversy around the development, validation, and implementation of scoring methods for quantitative assessment of structural damage. The aims of this review are to outline the key arguments that form the basis for each area of controversy, present conclusions considered justified by the weight of evidence, provide recommendations on the use of radiography in clinical practice and research based on these conclusions, and to suggest areas of unmet need and future directions.

Radiography of the SIJ – what constitutes the optimal approach to diagnostic evaluation?

Radiography depicts abnormalities of cancellous or cortical bone and so will demonstrate focal or diffuse changes of osteoporosis, sclerosis, erosion or ankylosis. In the SIJs, the earliest abnormality is a loss of distinctness of the subchondral bone followed by more obvious erosion and widening of the joint space. Progressive sclerosis and joint space ankylosis represent later features of disease. Changes in bone marrow or cartilage are indirectly visualised. For example, inflammation within bone marrow is implied when erosion of the calcified bony matrix is visible and cannot be detected in the absence of bone changes. Limited histopathological evaluation of SIJ in early disease has shown that a prominent feature is subchondral inflammation eroding through the overlying cartilage . This is consistent with the classic depiction on radiography of the serrated postage stamp appearance of early erosions in the lower iliac portion of the joint. Early involvement of the iliac portion is consistent with both histopathological and MRI evaluation in patients with early SpA .

A major challenge to early diagnosis using radiography is the complex anatomy of the joint. The cartilaginous portion, which is the region primarily affected in early disease, is convex antero-inferiorly in the anterior one-third of the joint. In addition, the joint cavity is oblique with the posterior aspect located medially and the anterior located laterally. This creates difficulty in the interpretation of early subchondral changes using the conventional antero-posterior (A–P) view. Overlying bowel gas and soft tissues further confound detection of abnormalities. Erosion may therefore be quite severe in the SIJ before being detected and this has led to the use of various imaging approaches in clinical practice to optimise radiographic assessment of the SIJ.

Oblique views of individual SIJ are aimed at superimposing the anterior and posterior joint margins and are taken at an angle of 25–30° to the A–P plane. In the Ferguson view, the patient is in the same position as for the A–P pelvis but the tube is angled 30–35° cephalic and is centred to the midportion of the pelvis. With this projection, the symphysis pubis overlaps the sacrum and this has been proposed as an approach that more directly visualises the joint. But does any one of these views offer superior diagnostic accuracy? Moreover, some patients have questionable abnormalities reported on the routine A–P view and are then sent for additional radiographic assessment aimed at specific evaluation of the SIJ which results in additional radiation exposure. Is this practice justified by the improvement in diagnostic accuracy?

The data addressing these questions are relatively limited. One report described the assessment of 43 consecutive patients referred with clinical suspicion of inflammatory back pain . Each patient had four views of the SIJ taken: a postero-anterior (PA) view of the pelvis (PA pelvis), an angled view (PA pelvis with the tube angled 30° caudally) and oblique views of each SI joint. The PA view of the pelvis is sometimes used in preference to the A–P view to take advantage of radiographic beam divergence, producing an image more closely parallel to the surface of the SIJ in the PA projection. Films were graded as positive, equivocal, or negative for sacroiliitis by four readers. Computed tomography (CT) was conducted on 35 unselected patients as the gold-standard evaluation. The impact of a series of views upon radiographic diagnosis was established by comparing each observer’s reading of each individual view with his reading of the combined series of views in that patient. The accuracy of different views was determined by comparison of individual or combined series readings with CT diagnosis. The interpretation of individual views and the combined series of views differed in 25–30% of radiographs. The percent equivocal findings for sacroiliitis were 27%, 20%, and 22% for the P–A, angled P–A, and oblique views, respectively, compared to 8% for the combined series. Sensitivity for sacroiliitis was 32%, 34%, and 30% for the P–A, angled P–A, and oblique views, respectively, compared to 47% for the combined series. Using CT as gold-standard reference, the oblique view was most often interpreted incorrectly (19%), usually due to false-negative findings, as compared to the P–A (12%) and angled P–A (16%) views. However, the combined series was also interpreted incorrectly in 19% when compared to CT. Reading the combined series did correctly resolve diagnostic uncertainty in most of those patients considered equivocal according to the single view. The main conclusions of this study were that the individual views had comparable diagnostic accuracy and that obtaining additional views when findings were equivocal on a single view enhanced diagnostic accuracy, although this enhancement in accuracy was only observed in 10–15% of patients and was not statistically significant.

A second report assessed 100 consecutive patients who had routine A–P radiographs of the pelvis of which six were judged as abnormal and five as equivocal . Review of the additional specific sacroiliac radiographs confirmed the six abnormal cases, and resulted in one equivocal case being judged abnormal. In no case did the additional SIJ radiograph result in a normal diagnosis being changed to abnormal. In a third report, there was no significant benefit to the addition of oblique views in a study of 29 patients whose radiographs were assessed by four experienced readers . Only five patients (eight SI joints) had a significant change in New York radiographic criteria grade. In three of the five patients, this change was judged as misleading as determined by blind evaluation of the clinical record and HLA-B27 status. Thus, in most cases oblique films did not change the grading and when they did they were misleading as often as helpful.

In a study of 445 patients with SpA, oblique sacroiliac-joint views and the A–P pelvic view were compared for radiographic severity of sacroiliitis according to the modified New York classification (mNY) criteria in patients recruited to a placebo-controlled trial of salazopyrin for SpA . Percent agreement between the A–P view and oblique SIJ views was 86.4% and 89.7% for left and right SIJ, respectively. As expected, agreement was higher with increasing grade of sacroiliitis. Agreement was about 60% for radiographs considered minimally abnormal and about 90% for those considered unequivocally abnormal. SIJ views tended to be read more severely but there were no instances where definite sacroiliitis on A–P view was read as normal on the oblique SIJ and vice versa. A limitation of this study was that only 39 (8.8%) and 34 (7.7%) had normal right and left SIJ radiographs, respectively, so that the sensitivity of the single reader was enhanced and so the setting was not comparable to clinical practice where radiography is conducted to help distinguish between SpA and other forms of back pain. A gold-standard assessment such as CT was also not available.

In conclusion, there is little evidence to support the use of specific SIJ views over the conventional A–P of the pelvis. Moreover, there is minimal additional diagnostic utility to the use of specific SIJ views where the A–P of the pelvis is equivocal for sacroiliitis which comes at the expense of additional exposure to radiation. In the setting of early SpA where an A–P pelvic radiograph is reported as equivocal for SpA, MRI of the pelvis would be a more useful option than additional radiography. It is unlikely that further standardisation and training of readers to assess radiographs of the SIJ will change these conclusions. An extensive program of calibration and training of 23 radiologists and 100 rheumatologists over a 6-month period failed to improve the sensitivity and specificity of radiography for the detection of sacroiliitis .

Radiography of the SIJ – what constitutes the optimal approach to diagnostic evaluation?

Radiography depicts abnormalities of cancellous or cortical bone and so will demonstrate focal or diffuse changes of osteoporosis, sclerosis, erosion or ankylosis. In the SIJs, the earliest abnormality is a loss of distinctness of the subchondral bone followed by more obvious erosion and widening of the joint space. Progressive sclerosis and joint space ankylosis represent later features of disease. Changes in bone marrow or cartilage are indirectly visualised. For example, inflammation within bone marrow is implied when erosion of the calcified bony matrix is visible and cannot be detected in the absence of bone changes. Limited histopathological evaluation of SIJ in early disease has shown that a prominent feature is subchondral inflammation eroding through the overlying cartilage . This is consistent with the classic depiction on radiography of the serrated postage stamp appearance of early erosions in the lower iliac portion of the joint. Early involvement of the iliac portion is consistent with both histopathological and MRI evaluation in patients with early SpA .

A major challenge to early diagnosis using radiography is the complex anatomy of the joint. The cartilaginous portion, which is the region primarily affected in early disease, is convex antero-inferiorly in the anterior one-third of the joint. In addition, the joint cavity is oblique with the posterior aspect located medially and the anterior located laterally. This creates difficulty in the interpretation of early subchondral changes using the conventional antero-posterior (A–P) view. Overlying bowel gas and soft tissues further confound detection of abnormalities. Erosion may therefore be quite severe in the SIJ before being detected and this has led to the use of various imaging approaches in clinical practice to optimise radiographic assessment of the SIJ.

Oblique views of individual SIJ are aimed at superimposing the anterior and posterior joint margins and are taken at an angle of 25–30° to the A–P plane. In the Ferguson view, the patient is in the same position as for the A–P pelvis but the tube is angled 30–35° cephalic and is centred to the midportion of the pelvis. With this projection, the symphysis pubis overlaps the sacrum and this has been proposed as an approach that more directly visualises the joint. But does any one of these views offer superior diagnostic accuracy? Moreover, some patients have questionable abnormalities reported on the routine A–P view and are then sent for additional radiographic assessment aimed at specific evaluation of the SIJ which results in additional radiation exposure. Is this practice justified by the improvement in diagnostic accuracy?

The data addressing these questions are relatively limited. One report described the assessment of 43 consecutive patients referred with clinical suspicion of inflammatory back pain . Each patient had four views of the SIJ taken: a postero-anterior (PA) view of the pelvis (PA pelvis), an angled view (PA pelvis with the tube angled 30° caudally) and oblique views of each SI joint. The PA view of the pelvis is sometimes used in preference to the A–P view to take advantage of radiographic beam divergence, producing an image more closely parallel to the surface of the SIJ in the PA projection. Films were graded as positive, equivocal, or negative for sacroiliitis by four readers. Computed tomography (CT) was conducted on 35 unselected patients as the gold-standard evaluation. The impact of a series of views upon radiographic diagnosis was established by comparing each observer’s reading of each individual view with his reading of the combined series of views in that patient. The accuracy of different views was determined by comparison of individual or combined series readings with CT diagnosis. The interpretation of individual views and the combined series of views differed in 25–30% of radiographs. The percent equivocal findings for sacroiliitis were 27%, 20%, and 22% for the P–A, angled P–A, and oblique views, respectively, compared to 8% for the combined series. Sensitivity for sacroiliitis was 32%, 34%, and 30% for the P–A, angled P–A, and oblique views, respectively, compared to 47% for the combined series. Using CT as gold-standard reference, the oblique view was most often interpreted incorrectly (19%), usually due to false-negative findings, as compared to the P–A (12%) and angled P–A (16%) views. However, the combined series was also interpreted incorrectly in 19% when compared to CT. Reading the combined series did correctly resolve diagnostic uncertainty in most of those patients considered equivocal according to the single view. The main conclusions of this study were that the individual views had comparable diagnostic accuracy and that obtaining additional views when findings were equivocal on a single view enhanced diagnostic accuracy, although this enhancement in accuracy was only observed in 10–15% of patients and was not statistically significant.

A second report assessed 100 consecutive patients who had routine A–P radiographs of the pelvis of which six were judged as abnormal and five as equivocal . Review of the additional specific sacroiliac radiographs confirmed the six abnormal cases, and resulted in one equivocal case being judged abnormal. In no case did the additional SIJ radiograph result in a normal diagnosis being changed to abnormal. In a third report, there was no significant benefit to the addition of oblique views in a study of 29 patients whose radiographs were assessed by four experienced readers . Only five patients (eight SI joints) had a significant change in New York radiographic criteria grade. In three of the five patients, this change was judged as misleading as determined by blind evaluation of the clinical record and HLA-B27 status. Thus, in most cases oblique films did not change the grading and when they did they were misleading as often as helpful.

In a study of 445 patients with SpA, oblique sacroiliac-joint views and the A–P pelvic view were compared for radiographic severity of sacroiliitis according to the modified New York classification (mNY) criteria in patients recruited to a placebo-controlled trial of salazopyrin for SpA . Percent agreement between the A–P view and oblique SIJ views was 86.4% and 89.7% for left and right SIJ, respectively. As expected, agreement was higher with increasing grade of sacroiliitis. Agreement was about 60% for radiographs considered minimally abnormal and about 90% for those considered unequivocally abnormal. SIJ views tended to be read more severely but there were no instances where definite sacroiliitis on A–P view was read as normal on the oblique SIJ and vice versa. A limitation of this study was that only 39 (8.8%) and 34 (7.7%) had normal right and left SIJ radiographs, respectively, so that the sensitivity of the single reader was enhanced and so the setting was not comparable to clinical practice where radiography is conducted to help distinguish between SpA and other forms of back pain. A gold-standard assessment such as CT was also not available.

In conclusion, there is little evidence to support the use of specific SIJ views over the conventional A–P of the pelvis. Moreover, there is minimal additional diagnostic utility to the use of specific SIJ views where the A–P of the pelvis is equivocal for sacroiliitis which comes at the expense of additional exposure to radiation. In the setting of early SpA where an A–P pelvic radiograph is reported as equivocal for SpA, MRI of the pelvis would be a more useful option than additional radiography. It is unlikely that further standardisation and training of readers to assess radiographs of the SIJ will change these conclusions. An extensive program of calibration and training of 23 radiologists and 100 rheumatologists over a 6-month period failed to improve the sensitivity and specificity of radiography for the detection of sacroiliitis .

Diagnostic utility – how useful is radiography in early SpA?

The mNY criteria for ankylosing spondylitis represent a major advance in the field for defining this disease and rely substantially on the finding of radiographic sacroiliitis . This was also considered an important advance in helping clinicians diagnose AS because of the relative absence of objective manifestations of disease such as physical findings and laboratory abnormalities early in the disease course. But in recent years they have been deemed inadequate because prospective studies have shown that evolution of unequivocal radiographic sacroiliitis as defined by the mNY criteria may occur over many years and during this time patients may experience substantial symptomatology comparable to patients with established AS . Patients were often considered as having undifferentiated SpA but this was not considered to be a useful approach as it correctly implied a poorly defined category of patients. With advances in diagnostic imaging and the introduction of highly effective new therapies, the lengthy diagnostic delay of 7–9 years was no longer considered acceptable and this culminated in the development and preliminary validation of the new ASAS classification criteria for SpA aimed at capturing patients early in their disease course prior to the evolution of radiographic sacroiliitis and those with established SpA . According to these criteria a patient may be classified as having axial SpA by one of two approaches: (1) an ‘imaging arm’ that requires either radiographic sacroiliitis or a positive MRI for sacroiliitis with one clinical feature of SpA or (2) a ‘clinical arm’ that requires B27 positivity and two clinical features of SpA but does not require evidence of SpA on imaging. The gold-standard for the estimation of sensitivity and specificity in the study that developed and validated these criteria was the rheumatologist expert opinion of unselected patients with back pain less than 45 years of age referred with a suspicion of SpA . Because these criteria are increasingly being used to establish a diagnosis of SpA, it is important to review the implications of their implementation for the diagnostic utility of radiographic sacroiliitis, particularly in relation to other imaging modalities.

The validation study of 649 consecutive patients less than 45 years of age referred to rheumatologist members of ASAS for diagnostic evaluation of chronic low back pain reported radiographic sacroiliitis by mNY criteria in 29.7% of 391 patients diagnosed with axial SpA according to rheumatologist expert opinion . Mean symptom duration in these patients was 6.1 years. Devenir des Spondylarthropathies Indifferenciees Recentes (DESIR) is a prospective longitudinal cohort in France involving 25 rheumatology centres and 708 patients . Consecutive patients aged less than 50 years with inflammatory back pain (IBP) for greater than 3 months but less than 3 years and symptoms suggestive of SpA according to the rheumatologists’ assessment were included in the DESIR cohort. Patients had to fulfil the Calin or Berlin IBP criteria . There were 181 patients who fulfilled the mNY criteria at baseline and 475 who fulfilled the axial SpA criteria leading to a similar proportion of 27.6% of patients with radiographic sacroiliitis in patients diagnosed with axial SpA. In a cohort from Germany, 350 patients less than 45 years of age with chronic back pain and any one of IBP, B27 positivity, and sacroiliitis detected by imaging were referred for diagnostic evaluation . Mean symptom duration before diagnosis of SpA was 7.7 years and radiographic sacroiliitis was evident in 50.3% of 159 patients diagnosed with axial SpA. In a second study from Germany, 318 patients were referred by a similar screening strategy and 82 (61.7%) of 133 patients diagnosed with axial SpA had radiographic sacroiliitis . When the screening strategy required at least 2 of IBP, B27 positivity, sacroiliitis detected by imaging, family history of AS, and good response to non-steroidal anti-inflammatory drug (NSAID), a total of 242 patients were referred of whom 55 (61.8%) had radiographic sacroiliitis of 89 patients diagnosed with axial SpA. Mean symptom duration was 8–9 years for patients referred by either strategy.

The sensitivity of radiography for sacroiliitis in early SpA therefore seems to range from 30% to 60% with the higher estimate being evident in patient cohorts pre-screened using some characteristics of SpA and/or from cohorts with longer symptom duration. Nevertheless, radiographic sacroiliitis may be evident in 20–30% of patients with symptom duration of only 2–3 years . Several prospective studies have assessed development of radiographic sacroiliitis in patients presenting with features of SpA. One study of 88 patients with IBP but with radiographically normal SIJ showed that only 36% had developed radiographic sacroiliitis after 5 years and only 59% after 10 years . A prospective study of 68 patients with undifferentiated SpA according to European Spondyloarthropathy Study Group (ESSG) criteria reported the development of radiographic sacroiliitis in 10% of patients after 2 years . A more recent study of patients with symptom duration less than 5 years reported radiographic sacroiliitis in 11.6% of patients after 2 years . This was 24% in patients who had elevated C-reactive protein (CRP).

These data have important implications for the investigation of patients with axial SpA who have had radiographs prior to presentation that were considered negative for sacroiliitis. It does not appear worthwhile to repeat radiographic examination where earlier exams were conducted within the previous 5 years, and especially if there is access to MRI, unless the patient presents with an elevated CRP. While not the primary focus of this review it is worthwhile noting that several reports have described abnormalities on MRI in patients with non-radiographic SpA with sensitivity of 54–95% and specificity of 83–100%, although only small numbers of controls with nonspecific back pain were included and some studies assessed dynamic imaging with gadolinium enhancement, which is not a procedure used in clinical practice . The validation study that led to the development of the ASAS axial SpA criteria showed that active inflammation in the SIJ was reported on MRI by the site radiologist in 64.7% of patients less than 45 years of age presenting with back pain and subsequently diagnosed as having SpA as compared to 2.6% of those considered as non-SpA . Active inflammation was evident in 61.6% of patients with radiographic sacroiliitis that was grade 1 or less. Moreover, the inclusion of MRI improved the sensitivity of the Amor (69.3% → 82.9%) and ESSG criteria (72.4% → 85.1%) without impacting specificity. A recent report that assessed MRI sequences commonly used in clinical practice and used age- and sex-matched healthy controls as well as controls with non-specific back pain showed that sensitivity was 51% and specificity 97% in patients with non-radiographic SpA . A study of patients suspected of having SpA but without radiographic sacroiliitis that combined MRI and histopathological assessment of SIJ after needle biopsy showed that specificity of MRI for sacroiliitis was 100% while sensitivity was only 38% . But this was a 10-year follow-up study and since then there have been substantial refinements in MRI scanning technology. Several prospective studies have also demonstrated the predictive validity of MRI for the development of radiographic sacroiliitis . Although CT has also been shown to have superior diagnostic utility to plain radiography , it seems unlikely that it will supplant MRI unless it is shown to have superior diagnostic utility in view of the exposure to radiation. Comparative studies between plain radiography and CT have however highlighted the difficulty in reliably establishing a diagnosis of sacroiliitis with uncertainty being threefold greater using plain imaging and false-positive diagnoses ranging from 20% to 35% .

What is the role of radiography in follow-up assessment and disease management?

Once radiographic sacroiliitis is evident on pelvic radiography, there is no purpose to further evaluation of the SIJ. But it has been proposed by ASAS that spinal radiography be considered a core outcome domain for the assessment of disease-controlling anti-rheumatic therapy in patients with SpA . Hip radiography was also placed on the research agenda as a possible outcome domain for disease-modifying therapies. Plain radiography can show a variety of features in the spine. In spinal vertebrae the earliest feature is the loss of the cortex at the corner of the vertebral body giving the appearance of an erosion. Bone remodelling and new bone formation lead to the radiographic appearance of squaring and sclerosis at the vertebral corner. Further new bone formation from the vertebral corner across the disc space to the adjacent vertebral corner or syndesmophyte may ultimately lead to complete ankylosis. This occurs not only at the vertebral rim but also in the interior of the disc. Spondylodiscitis is radiographically evident as disruption and loss of the vertebral endplate. Facet joint abnormalities consist of erosions, loss of joint space, and ankylosis. They are not readily visible in the thoracic spine because of overlapping structures. They are visible on lateral radiographs of the cervical spine and it has been shown that structural changes of joints space narrowing and fusion can be reliably detected . Assessment of facet joints in the lumbar spine requires oblique views. Systematic prospective radiographic studies are limited but clinical observations indicate that abnormalities typically originate in the lumbar spine and ascend cranially although the cervical spine may be preferentially affected in some patients. Typical features occur relatively late in the course of the disease and are in general not contributory to the diagnosis. In a cross-sectional cohort of patients with AS and a mean disease duration of almost 12 years, more than 60% of patients had features attributable to AS on their spinal radiographs, but only a minority had syndesmophytes extending over multiple vertebrae . A ‘bamboo spine’, which reflects an end-stage of spinal AS, was observed in less than 5%. Radiographic abnormalities are associated with impaired spinal mobility . The relationship is not linear and the association increases with increasing level of ankylosis.

The pace of radiographic progression in terms of new bone formation as detected by syndesmophytes and ankylosis is variable but only about 20% of patients show progression over 2 years . Moreover, until recently there was minimal evidence that any therapeutic approach had disease-modifying potential. Consequently, there has been little interest in the radiographic assessment of the spine for follow-up and disease management. But there is now evidence from two recent studies that continuous use of NSAIDs may prevent the development of new bone in patients at risk of progression because of elevated acute phase reactants or the presence of baseline syndesmophytes . This may particularly impact the approach to treatment of patients who may be symptomatically well controlled on minimal or discontinuous NSAID therapy but who show evidence of radiographic progression. Since NSAID therapy may be associated with gastrointestinal, cardiovascular, and renal morbidity, appropriate patient selection for long-term continuous therapy is essential. Consequently, there may now be justifiable grounds to monitor progression with spinal radiography, particularly if the patient has elevated acute phase reactants and evidence of new bone formation on the spinal radiograph. But this should not be done within a time frame that is shorter than 2 years since this is the minimum period that must elapse before radiographic progression can be reliably detected .