Visual Identification of Vertebral Fractures in Osteoporosis Using Morphometric X-ray Absorptiometry†
The authors have no conflict of interest.
Visual identification of vertebral fractures from spinal radiographs (visual XR) makes use of the reader's expertise in ruling out non-fracture deformities or normal variants. Scan images of the spine acquired by DXA may be analyzed quantitatively (morphometric X-ray absorptiometry [quantitative MXA]) or visually (visual MXA). The aims of this study were to compare visual and quantitative MXA with visual XR for the identification of vertebral fractures. Spinal radiographs and MXA scans were acquired at baseline and 1 year in 70 women referred with osteoporosis. These were assessed visually by two expert readers (observer A, a radiologist; observer B, a physician with expertise in osteoporosis) for evidence of prevalent and incident vertebral fractures. Observer C (a radiographer with expertise in vertebral morphometry) performed visual and quantitative assessments of the MXA scans. Visual assessment of spinal radiographs by observer A was used as the gold standard for comparison of methods. Sensitivity for the identification of prevalent fractures by MXA was best for visual MXA by observer A (92%), whereas quantitative MXA had the lowest sensitivity (82%). Specificity was >90% for both visual and quantitative MXA. Kappa scores for agreement for identification of prevalent fractures between visual XR (observer A) and visual MXA (all three observers), and between visual XR and visual MXA performed by reader B were similar (κ = 0.85-0.87). Agreement with visual XR performed by observer A was slightly lower for quantitative MXA (κ = 0.77). Interobserver agreement between the two expert readers (observers A and B) was the same for both visual XR and visual MXA (κ = 0.86). Seven incident vertebral fractures were identified in four patients at follow-up. All four patients were identified by visual MXA, and three patients were identified by quantitative MXA. Observers A and B identified all seven incident fractures by visual MXA, and observer C missed one fracture that was also missed by quantitative MXA. An incident fracture of vertebra T6 was excluded from analysis by quantitative MXA because of poor image quality. We conclude that visual identification of vertebral fractures from MXA scans is superior to quantitative assessment. Used as a screening tool for conventional radiography, this approach could help reduce the radiation dose to the patient in the diagnosis and monitoring of osteoporosis.
Vertebral fractures may be identified by visual or quantitative assessment of spinal radiographs. The quantitative approach may also be applied to images of the spine acquired by DXA. This method is known as morphometric X-ray absorptiometry (MXA). The main advantage of this technique is that the radiation dose to the patient is substantially reduced compared with conventional radiography.(1)
There is evidence that quantitative analysis of MXA scans has good agreement with visual radiological assessment of spinal radiographs for the identification of vertebral deformities,(2–4) but agreement is generally better for quantitative assessment of spinal radiographs. In these method comparison studies, there has been a bias in favor of the quantitative assessment of spinal radiographs, because this and the gold standard approach used for comparison of methods (visual diagnosis by a radiologist) were both based on the same images (spinal radiographs).
Although there is no consensus on the optimum approach, visual assessment of vertebral fractures has the advantage that the expert reader is able to rule out nonfracture deformities or normal variants, which may be falsely identified as fractures by the quantitative approaches. Attention has recently focused on the use of MXA images of the spine for visual identification of vertebral fractures, and new applications (instant vertebral assessment [IVA]) have recently been developed (Hologic Inc., Bedford, MA, USA) to facilitate this approach.(5) Visual assessment of MXA scans has been compared with radiological evaluation of spinal radiographs in a population-based sample of women.(6) The results of that study were promising, but in that study, the MXA scan images were not evaluated by a radiologist, and agreement has not been tested for visual assessments of MXA scans and spinal radiographs made by the same reader. We have previously assessed quantitative but not visual MXA for the identification of vertebral fractures.(2, 7) The quantitative assessment was performed by a radiographer (LF), and a consensus of visual readings of spinal radiographs by expert readers was used as the gold standard. The expert readers did not assess the MXA scan images, and we have not previously evaluated interobserver agreement among expert readers for the reporting of spinal radiographs.
The aims of this study were to (1) compare quantitative and visual assessment of MXA scans with visual assessment of spinal radiographs for the identification of prevalent and incident vertebral fractures; (2) compare visual assessment of MXA scans and spinal radiographs made by the same observer for the identification of prevalent and incident vertebral fractures; and (3) test interobserver agreement between expert readers for visual assessment of MXA scans and spinal radiographs.
MATERIALS AND METHODS
We studied spinal radiographs and MXA scan images for 70 women referred with osteoporosis (bone mineral density [BMD] T score < −2.5 and/or vertebral fractures) to the Metabolic Bone Clinic, Northern General Hospital, Sheffield, UK. The women were selected consecutively if they fulfilled these criteria and if they were willing to participate in the study by undergoing MXA examination. The women were ages 49-87 years (70 ± 9 years). Mean BMD in these women measured by DXA at the lumbar spine using the Hologic QDR 4500-A densitometer (Hologic Inc.) was 0.727 ± 0.112 g/cm2 (mean T score, −2.9). The prevalence of vertebral fracture in this group (identified by qualitative radiological assessment of spinal radiographs) was approximately 98%. This population has been described in detail elsewhere.(2) Follow-up spinal radiographs and MXA scans were acquired at 1 year in a subset of 50 women, ages 49-87 years (mean, 67 ± 9 years). Of the remaining 20 women, 2 patients had been discharged from the clinic, 2 had died, and the remainder had undergone spinal radiography but not MXA at their follow-up visits (because of an administrative error). There were no significant differences between the women with fractures who were studied at baseline and those who were studied at follow-up. All patients gave informed consent, and the study was approved by the North Sheffield Research Ethics Committee.
Spinal radiography and MXA
Radiographic examinations were performed in the Diagnostic Imaging Department, and MXA scans were performed in the Osteoporosis Centre at the Northern General Hospital Trust, Sheffield, UK. The acquisition of spinal radiographs and MXA scans has been described in detail elsewhere.(2) Briefly, radiographs of the thoraco-lumbar spine and MXA scans of vertebrae T4-L4 were acquired following a standardized protocol. Spinal radiographs at baseline were acquired in the antero-posterior and lateral projections, and at follow-up in the lateral projection only. Single- and dual-energy (high definition) MXA scans were acquired at baseline and follow-up using the Hologic QDR 4500-A densitometer. The postero-anterior and lateral scans were both acquired with the patient in the supine position (with the C-arm of the scanner rotated through 90° for horizontal lateral scanning). The marking and quantitative analysis of MXA scan images (LF) was performed in previous studies, using our own locally derived reference data and the Eastell-Melton algorithm for identification of vertebral deformities.(2, 7–9) In quantitative analysis, both single- and dual-energy scans were viewed alongside each other to help determine the correct position for placement of vertebral markers, but for consistency, the markers were positioned on the single-energy scans only. Severity of deformity was classified by a reduction in vertebral height ratio ≤3 SD below the reference mean for a mild deformity or ≤4 SD below the reference mean for a moderate to severe deformity.
Visual assessment of spinal radiographs and scan images
The spinal radiographs were assessed independently by two expert observers (who were both medically qualified) for evidence of osteoporotic vertebral fracture. These were a skeletal radiologist (GJ; observer A) and a physician with expertise in osteoporosis (NFAP; observer B). Fracture assessment was performed qualitatively, and vertebral fractures were classified qualitatively by severity as mild, moderate, or severe, and by type as wedge, endplate, or compression. Follow-up radiographs were viewed alongside those acquired at baseline, and incident vertebral fractures were identified as fractures in vertebrae that were previously identified as normal at baseline.
The MXA scan images were assessed visually following the same method that had been used to assess spinal radiographs. Both single- and dual-energy scans were viewed for each patient as in the quantitative assessment. Because the densitometer used in this study did not incorporate the newer Hologic IVA software, the scan images were assessed in “view” mode so that the vertebral markers from the quantitative analysis were not displayed. The visual assessment of MXA images was performed independently by observers A and B. Visual MXA was also performed by a radiographer (LF) who has expertise in quantitative vertebral morphometry but no formal training in the visual identification of vertebral fractures (observer C). For the purposes of this study, observer C did not assess the spinal radiographs. This was because in clinical practice, a radiographer would not usually perform this role. In the UK, radiographers predominantly perform a technical role and are not medically qualified. They are responsible for acquiring, but not reporting on, medical images (although there has been some role extension in this area, mainly in the reporting of obstetric ultrasound images and barium enema examinations). A radiographer or technician could evaluate MXA images as a prescreen for spinal radiography, but because final reporting of these radiographs influences clinical management of the patient, a physician or radiologist would perform this evaluation.
The MXA scans were viewed on a visual display unit. The vertebral markers from quantitative analysis of the scan images were not displayed. For the identification of prevalent vertebral fractures at baseline, the single- and dual-energy scans were viewed side by side. For the identification of incident vertebral fractures, the recent scan was viewed alongside the baseline scan. The readers noted equivocal findings because of poor image quality and borderline changes (abnormal appearances or early changes that were too mild to be classed as fracture). The MXA scan images had also been analyzed quantitatively by Ferrar et al.(2, 7) in an earlier study.
Comparison of methods
Agreement for the identification of vertebral fractures was tested between visual MXA by all three observers and visual XR by observer A, and between quantitative MXA and visual XR by observer A. For the purposes of this analysis, visual XR by observer A was used as a surrogate gold standard. Agreement was also tested between visual MXA and visual XR performed by observer B. Interobserver agreement for visual MXA and visual XR was tested for observers A and B. Sensitivity and specificity were calculated for visual MXA and quantitative MXA. For these analyses, visual XR by observer A was used as the gold standard. Analyses were performed on a per-vertebra basis.
Inter-rater agreement between methods was tested using κ statistics. All analyses were performed on a per-vertebra basis.
Identification of prevalent vertebral fractures
The total numbers of prevalent vertebral fractures identified by each method and for each observer are shown in Table 1. In general, more fractures were identified from radiographs than from scan images, and observer A identified more fractures by visual XR and visual MXA than observers B and C. All observers identified slightly more moderate to severe compared with mild fractures, and the same was true for quantitative MXA.
Table Table 1. Identification of Prevalent Vertebral Fractures From Spinal Radiographs and MXA Scan Images
Approximately 12% (visual MXA) to 15% (quantitative MXA) of all vertebrae imaged by MXA were excluded from analysis because of poor image quality. A total of 133 vertebrae were excluded for quantitative MXA compared with 119, 105, and 103 vertebrae, respectively, for visual MXA performed by observers A, B, and C. This resulted in 41 missed fractures for quantitative MXA, and 21 (observers B and C) and 18 (observer A) missed fractures for visual MXA.
The numbers of false positive, false negative, and true identifications of prevalent vertebral fracture according to the different methods and observers are shown in Table 2. There were more false positives for quantitative MXA than for visual MXA, and the false positive rate was particularly low for visual MXA by observers B and C (3 and 11 false positives, respectively). The false negative rate for quantitative MXA and for visual MXA by observers B and C was similar (46, 47, and 41 false negatives, respectively), but was twice the rate for visual MXA by observer A. Approximately two-thirds of all false negatives for quantitative MXA and visual MXA were mild fractures. Vertebral level had little influence on the number of true fractures identified by visual MXA and quantitative MXA (data not shown).
Table Table 2. Sensitivity and Specificity for the Identification of Prevalent Vertebral Fractures
Agreement with visual identification of prevalent vertebral fractures by observer A was good for both visual MXA and quantitative MXA (particularly for the identification of moderate to severe fractures), but agreement was better for visual MXA than for quantitative MXA (Table 3). Agreement between the two expert readers (observers A and B) for visual MXA was good (κ = 0.86), and was similar to the agreement between the same two readers for visual assessment of spinal radiographs and the agreement between visual assessment of radiographs and MXA scans by observer B (Table 4).
Table Table 3. Inter-rater Agreement for the Identification of Prevalent Vertebral Fractures
Table Table 4. Inter-rater Agreement Between Expert Readers
Identification of incident vertebral fractures
Seven incident vertebral fractures in four patients were identified by visual XR (observers A and B) at 1-year follow-up. Observers A and B correctly identified all seven incident fractures by visual MXA. Observer C identified all four patients with incident fractures and six of seven incident fractures by visual MXA. A total of six of seven patients with incident fracture and five of seven incident fractures were identified by quantitative MXA. The false negatives were a mild endplate fracture of vertebra L4 (missed by both visual MXA performed by observer C and quantitative MXA) and a moderate wedge fracture of vertebra T6 that was missed because of exclusion from analysis by quantitative MXA. The number of false positives identified by visual MXA at follow-up was 3, 1, and 0 for observers A, B, and C, respectively. Thirteen false positives were identified by quantitative MXA.
The level of agreement for identification of prevalent vertebral fractures in this study was broadly similar for all visual assessments of spinal radiographs and MXA scans. As could be expected, the sensitivity for detection of prevalent fractures was slightly better for visual analysis of MXA scans and spinal radiographs performed by the same observer. The κ scores for agreement for visual MXA with visual assessment of spinal radiographs (κ = 0.85-0.87) and sensitivity for detection of prevalent vertebral fractures (85-92%) were slightly better than the values quoted in the previous study by Rea et al.(6) (κ = 0.79 for agreement with radiological assessment of spinal radiographs and sensitivity = 77%). In that study, the analyses of radiographs and MXA scan images were performed in two independent centers; therefore, the interobserver variability is likely to have been greater than in our study, in which the three observers are members of the same research group. Also, the radiological assessment of fractures was performed using the semiquantitative approach,(10) whereas in our study, all observers used a visual qualitative approach for analysis of both scans and radiographs. However, the optimum use of visual analysis of MXA scans may be in the osteoporosis clinic, where the physician may habitually assesses spinal radiographs and/or works closely with the local radiologist to identify vertebral fractures in patients with osteoporosis. In such circumstances, the concordance between visual assessments of MXA scans and spinal radiographs could reasonably be expected to be similar to that observed in our study.
The performance for quantitative assessment of MXA scans was not as good as for the visual analysis, and this was mainly because of the greater proportion of false positives identified by the quantitative approach. The numbers of missed fractures were similar for quantitative and visual assessment of MXA scans, which suggests that these were more likely to be caused by the relatively poor image quality for MXA in comparison with conventional radiography, rather than a weakness of the quantitative approach. The inferior image quality for MXA scans in comparison with conventional radiographs does mean that a number of vertebrae, particularly in the upper thoracic spine, cannot be evaluated. This is less of a concern if MXA is used as a prescreening tool for radiography than if MXA was used as a stand-alone technique. Also, the new IVA software that facilitates the visual assessment of MXA scans (Hologic Inc.) may offer improved image resolution in comparison with the earlier applications (QDR 2000 Plus and QDR 4500-A).
Because the prevalence of fracture in our study population was high, our analyses were largely based on the number of vertebral fractures rather than the number of individuals with fracture. Among our study population, although some fractures were missed by MXA, all patients with prevalent vertebral fractures were identified; therefore, for the identification of patients with fracture, visual MXA had 100% sensitivity and specificity. This means that if MXA had been used as a diagnostic prescreening tool at the first assessment, all patients with prevalent vertebral fracture (66 of 70 patients) would have been correctly referred for radiography to confirm the diagnosis.
Back pain can be indicative of further fracture in patients with established osteoporosis, but it is also a common complaint that can be attributed to various other causes. Approximately two-thirds of vertebral fractures are asymptomatic,(11, 12) and such fractures have recently been identified by IVA in 11-18% of patients with normal bone density.(13) In addition, vertebral fractures are difficult to assess clinically without recourse to radiography. The most useful application of MXA as a prescreening tool, therefore, is probably in the follow-up of patients with prevalent vertebral fractures who are known to be at greater risk of further vertebral fracture.(14–16) The side-by-side viewing of current and previous scans in MXA probably enhanced the correct identification of incident fractures in our study, and it is possible that the error rate may have been higher if the scans had been viewed separately. However, we chose this methodology to reflect the way fractures would be identified in clinical practice, whether by MXA or evaluation of spinal radiographs.
If visual analyses of MXA scans by an expert reader (observer A or B) had been used as a prescreen in our study population, all patients with incident fracture (seven patients) would have been correctly referred for radiography. This approach could represent a considerable saving in radiation dose to the patient. Patients could be scanned routinely at follow-up, or visual assessment of MXA could be targeted at patients who exhibit symptoms of incident vertebral fracture on attendance for follow-up. The system would be easy to implement by installing the appropriate software with which to view the scans in the osteoporosis clinic.
The results of our κ analysis of agreement between a radiologist (observer A) and a physician (observer B) for visual assessment of MXA scans suggest that the physician in the osteoporosis clinic is likely to be able to effectively implement this approach with a minimal amount of training.
In addition, concordance with the gold standard was similar for visual assessment of MXA scans by observer C to the agreement between the expert readers and the gold standard. Observer C is a diagnostic radiographer with no medical training. With appropriate training, radiographers could perform the initial visual assessments of MXA scans, referring any scans with borderline or uncertain appearances to an expert reader. This approach could save time for the busy clinician.
Our study had some limitations. The study population had a high prevalence of fracture, and in a population with fewer vertebral fractures, the agreement may be lower. There were only a small number of incident vertebral fractures at 1-year follow-up, and visual assessment of MXA scans should be evaluated over a longer timescale in the setting of the osteoporosis clinic. Also, this study was performed using an older version of the Hologic MXA software. The new IVA software incorporates features that may have further enhanced the visual assessment of vertebral fractures. However, with the Hologic Delphi software, spinal scans for IVA are acquired with the patient in the lateral decubitus position, which may increase the patient positioning error. Thus, we recommend that a comparison study be performed using the Delphi system.
We conclude that visual assessment of MXA scans agrees well with visual assessment of spinal radiographs and would be useful as a prescreening tool for radiography in the diagnosis and monitoring of osteoporotic vertebral fractures.
We acknowledge the staff of the Osteoporosis Centre, Northern General Hospital, UK, especially the radiographers for acquisition of MXA scans. We also thank the National Osteoporosis Society, UK, for their financial support of this research.