The Value of Anteroposterior Pelvic Radiographs for Evaluating Pelvic Osteolysis

Abstract

We investigated the role of conventional radiographs and computed tomography scans for the routine followup of total hip arthroplasty patients. Among 92 total hip arthroplasties with a mean followup of 8.5 years, 94 acetabular lesions were detected among 63 hips using computed tomography and 42 of these hips had osteolysis diagnosed on radiograph. Using computed tomography as a gold standard, the sensitivity of anteroposterior pelvic radiographs for the detection of acetabular osteolysis was 67% and the specificity was 72%. Although smaller lesions were more frequently missed, osteolysis was diagnosed on radiograph in 20 of 22 total hip arthroplasties with lesion volumes of at least 10 mL. Because larger osteolytic lesions were generally detected on radiograph, two-dimensional and three-dimensional lesion sizes correlated. However, the limits of agreement for the volume estimates based on the radiograph area were −14.6 to 18.7 mL. Although radiographs can be useful to screen for clinically important pelvic osteolysis, computed tomography images are necessary to accurately measure lesion volumes.

Level of Evidence: Level III, diagnostic study. See Guidelines for Authors for a complete description of levels of evidence.

Conventional radiographs are routinely used to noninvasively evaluate patients undergoing total hip arthroplasty (THA) and remain the clinical standard for routine followup. Radiographic changes, such as the development of radiolucent lines or the appearance of osteolysis, are often discernable before a patient demonstrates clinical symptoms. Yet, because a radiograph is a two-dimensional (2D) image which represents the superposition of three-dimensional (3D) anatomic structures along the direction of the radiograph beam, subtle features, such as pelvic osteolysis, can be difficult to identify. Several studies have demonstrated radiographs tend to underestimate the incidence and extent of periacetabular osteolysis.^8,11,12 In contrast, computed tomography (CT) enables 3D image acquisition, allowing the diagnosis of osteolysis on multiple, independent slices. Ambiguity on one slice can be reduced by considering adjacent slices and reformatting the image along different slice planes (axial, sagittal, coronal, or oblique).

Computed tomography has been a valuable tool for understanding the 3D nature of periacetabular bone defects. Coupled with image analysis software, CT images enable 3D visualization, assessment of lesion location, measurement of defect volume, and identification of communication pathways with the joint space.⁵ Based on these attributes and previously published validation studies, we regard CT as the gold standard for the diagnosis and evaluation of periacetabular osteolysis.^3,6,10,11 However, CT images involve substantially increased costs for image acquisition and analysis in addition to increased radiation exposure compared to conventional radiographs.

Although conventional radiographs may not reveal smaller osteolytic lesions, larger lesions are often visible. Because larger radiographic lesions should correspond to larger CT lesions and smaller lesions will contribute comparatively little to the variance in osteolytic lesion volume, it may be possible to develop an empirical relationship between the 2D area measured on radiograph and the 3D volume of the defect when the full spectrum of clinically-observed osteolysis is considered. If the correlation is sufficiently strong, it may be possible to use conventional radiographic findings to estimate the volume of the defect. In such a situation, plain radiographs could prove a simpler and more cost effective method for evaluating osteolytic lesion volume. Alternatively, if a correlation exists but is not sufficiently strong, radiographs may be appropriate to use as a screening tool to identify those patients who would be candidates for CT scans.

As our primary hypothesis, we proposed the volume of clinically-relevant osteolytic lesions could be reasonably estimated based on their radiographic appearance using an anteroposterior (AP) pelvic radiograph. Reducing this generalized proposition to specific hypotheses that could be quantitatively evaluated, we proposed that, despite suboptimal sensitivity and specificity, a relationship would exist between 2D osteolytic lesion area measured on anteroposterior (AP) pelvic radiographs and the 3D volume based on CT scans. Furthermore, we hypothesized the limits of agreement¹ for the estimates of osteolysis volume based on radiographic area would be within 5 mL of the volume measured on CT. We finally proposed osteolysis lesion volume would not be related to radiographic loosening or patient perceptions of groin and buttock pain.

MATERIALS AND METHODS

We retrospectively reviewed our institutional database to identify all primary THAs performed with Duraloc® 100 cups (DePuy, Warsaw, IN) with a digital CT scan at minimum 5-year followup. We limited our study population to those cases with an AP pelvic radiograph taken within 3 months of the CT scan and an AP pelvic radiograph taken at nominal 6-weeks followup. Based on these criteria, 92 THAs (82 patients) comprised our study population. Although all of our CT images were saved as digital images, we used conventional film for our radiographs.

During the 1990s the Duraloc® 100 cup was used for the majority of the primary THAs we performed. This hemispheric, porous-coated titanium cup featured a single apical hole and was implanted by press fitting the component after under-reaming the acetabulum by 1 mm relative to the diameter of the cup. We encouraged all our patients to return for followup and routinely obtained an AP pelvic radiograph during our examinations. These radiographs were taken with the beam centered on the pubic symphysis while the patient's legs were internally rotated. For small to average-sized patients, radiographs were typically taken at 75 to 80 kV and 200 mA with an exposure time of 0.25 seconds. For larger patients, the exposure times were progressively increased. Our goal was to obtain a radiograph with adequate contrast to distinguish the contour of the femoral head inside the acetabular cup while retaining the ability to visualize the pelvic bone and sclerotic borders demarcating the margins of osteolytic lesions. More recently, when the radiographic findings have suggested the presence of pelvic osteolysis, we have often ordered a CT image to confirm and evaluate the 3D nature of the lesion. Computed tomography images also have been ordered when patients reported pain not readily explained. When a patient with bilateral THAs had only one symptomatic hip, we routinely obtained a scan of the entire pelvis and reconstructed images for both hip replacements. Computed tomography images were typically taken at 140 kV and 150 mA using a slice width of 1 mm.

The digital CT slice data were loaded onto a personal computer (Dell, Round Rock, TX) and analyzed by an experienced observer (NK, an orthopaedic surgeon). After automatically segmenting the prosthesis based on a Hounsfield threshold, the reviewer manually segmented the regions of osteolysis in each image slice using a previously validated software application (Musculoskeletal Imaging, VirtualScopics, Rochester, NY).^3,6 The boundaries of the lesions were routinely identified by the sclerotic margins delineating periacetabular regions devoid of bone. The software application automatically computed the total volume of the periacetabular bone defects based on the CT-image slice thickness and the segmented areas in each slice.

Using an AP pelvic radiograph taken within 3 months of the CT scan (mean, 11.3 ± 23.6 days before; range, 90 days before to 71 days after) and the nominal 6-week postoperative view, another independent reviewer (PP) used a soft grease pencil (Stabilo, Heroldsbery, Germany) to delineate all periacetabular bone defect margins. This reviewer, who was blinded to the results of the CT analyses, was an orthopaedic surgeon in the final month of a 1-year fellowship in adult joint reconstruction. The radiograph films were subsequently digitized with a Vidar Diagnostic Pro Plus Scanner (Vidar Corporation, Herndon, VA) at 150 dpi with 8-bit resolution using a 10 millisecond exposure and a Power5 translation table. These settings enabled the osteolytic lesion markings to be clearly visualized on the digitized images along with the borders of the femoral head and cup. An analyst (PD) used Martell's Hip Analysis Suite software (University of Chicago, Chicago, IL) to measure the area and location of each lesion with correction for magnification based on the femoral head diameter. Osteolysis confined to the region behind the apex of the cup was defined as a dome lesion. Osteolysis adjacent to the face of the cup was defined as a rim lesion. Hips with one large lesion encompassing both regions, or discrete lesions in both regions, were classified as having dome and rim lysis.

If the same bone defect was apparent on the nominal 6-week postoperative radiograph and the followup CT or radiograph, the defect was considered a preexisting osteoarthritic cyst and was not included in the cumulative osteolytic lesion size computed on CT or radiograph. Periacetabular bone defects without a communication pathway to the joint space on CT were also considered preexisting cysts and were not included in our CT definition of osteolysis. After reconstructing the 3D geometry of the implant and all periacetabular bone defects, another analyst (AS) rotated the 3D image to simulate the AP pelvic radiograph view. A third independent reviewer (RH) visually compared the osteolysis identified on CT with the findings on radiograph. Using the CT findings as a gold standard, the radiograph findings were classified as under-diagnosing, correctly diagnosing, or over-diagnosing the extent of osteolysis. The osteolysis was considered correctly diagnosed if the area on radiograph was within approximately 10% of the projected area on the 3D image rotated to simulate the AP pelvic radiograph view. An additional category was incorporated if the area was correctly diagnosed on radiograph, but the location of the lesion did not correspond to the CT image.

The stability of the cup was evaluated using the AP radiographs and the CT images. The cup was regarded as radiographically stable if the CT and followup radiographs did not demonstrate circumferential radiolucencies at the cup-bone interface and it had not migrated more than 3 mm or rotated more than 3° based on the immediate postoperative and followup radiographs. A standardized patient questionnaire was administered at the time the AP radiograph was taken and used to determine the incidence of groin or buttock pain and its relationship to acetabular osteolytic lesion volume.

Using the data from all 92 hips, we evaluated the sensitivity and specificity of AP pelvic radiographs for the diagnosis of acetabular osteolysis using the CT findings as a gold standard. We also evaluated the correlation between the total 2D osteolytic area based on the AP pelvic radiograph and the 3D volume derived from the CT image analysis using Pearson's correlation. We subsequently performed a regression analysis to determine the relationship between 2D lesion size and 3D volume. Based on the assumption the absence of osteolysis on radiograph should correspond to a lesion volume of zero, the least squares regression was forced through the origin by setting the intercept to zero. An estimate of the 3D volume then was computed for each THA by multiplying the 2D area measured on the AP pelvic radiograph by the slope of the linear regression. Limits of agreement¹ were calculated for the 3D volume estimates based on the measured radiograph area and the osteolytic volumes evaluated from CT. We considered a p value of less than 0.05 as the threshold for statistical significance. Statistical analyses were performed with the Statistical Package for the Social Sciences (SPSS, Chicago, IL).

RESULTS

The sensitivity of the AP pelvic radiographs for the detection of osteolysis was 67% (42 of 63), whereas the specificity was 72% (21 of 29). Using CT, 94 discrete acetabular osteolytic lesions were identified among 63 of the 92 hips at a mean followup of 8.5 ± 2.1 years (range, 5 to 13 years). The mean volume of osteolysis for these 63 hips was 12.1 ± 15.7 mL (range, 0.2-57.7 mL). In 29 hips, with a mean osteolytic volume of 5.2 ± 6.9 mL (range, 0.3-26.2 mL), the lysis was confined to the dome region on the CT. In seven hips, it was limited to the rim and had a mean volume of 5.3 ± 9.5 mL (range, 0.2-26.7 mL). In 27 hips, with a mean osteolytic volume of 21.3 ± 19 mL (range, 1.1-57.7 mL), the lysis occupied both the dome and rim regions. Using AP pelvic radiographs, we identified pelvic osteolysis among 50 hips. The mean area of the lysis on the radiograph was 3 ± 3.9 cm² (range, 0.1-20.6 cm²). Among the 42 hips with evidence of osteolysis on radiograph and CT, the mean size of the osteolysis was 3.2 ± 4.1 cm² (range, 0.1-20.6 cm²) on radiograph and 16.3 ± 17.5 mL (range, 0.3-57.7 mL) on CT. Twenty-one hips, with a mean volume of 3.6 ± 4.9 mL (range, 0.2-16.5 mL) on CT, had no osteolysis identified on radiograph. Of these cases, the volume was less than 1 mL in 12 hips, between 1 and 10 mL in seven hips, and greater than 10 mL in two hips. Osteolysis apparent only on CT had a smaller (p < 0.001) volume than lysis apparent on both radiograph and CT. Twenty-one hips had no evidence of osteolysis on radiograph or CT. Eight hips had osteolysis identified on radiograph, with a mean size of 1.6 ± 1.3 cm² (range, 0.1-3.2 cm²), but no lysis on CT.

Because the larger osteolytic lesions were generally detected on radiograph, 2D and 3D lesion sizes correlated (r = 0.80, p < 0.001) among all 92 hips (Fig 1). Although smaller lesions were often missed, osteolysis was diagnosed on radiographs in 20 of 22 THAs with osteolytic volumes of at least 10 mL. Based on a linear regression, the volume of pelvic osteolysis, in mL, could be best estimated by multiplying the area on radiograph, measured in square centimeters, by a factor of 3.9 mL/cm².

Fig 1: Each point on the graph represents a single THA. The squares designate cases where the radiograph area under-diagnosed the extent of osteolysis relative to the CT reconstruction oriented to simulate the AP pelvic radiograph view. Similarly, the circles represent cases where the osteolysis was correctly identified, whereas the diamonds designate hips where the size of the osteolysis was similar, but the location differed on radiographs and CT. The triangles designate cases where osteolysis was over-diagnosed on the radiographs. The heavy black dashed line represents the best-fit regression through the origin and has a slope of 3.9. The lighter dashed lines bound the regions within ± 10 mL of the best-fit regression. The numbers 2 through 6 correspond to the other figures included with this study.

For all 92 hips, the limits of agreement for the CT volume estimates derived from the radiograph area were −14.6 to 18.7 cc. While the 2D area on radiographs could provide an accurate estimate of the 3D volume in some cases (Fig 2), the radiograph findings could also over diagnose (Fig 3) or, more frequently, under diagnose (Fig 4) the extent of the osteolysis. Even when the radiograph findings closely approximated the CT findings oriented to simulate an AP pelvic radiograph, there remained the possibility the lesion volume could be overestimated (Fig 5) or underestimated (Fig 6) based on the radiograph area. Using the 2D area data to estimate the 3D volume for the 22 THAs with pelvic osteolysis of at least 10 mL, we found the volume estimates were within 10 mL of the actual CT volume in only 10 of 22 cases.

Fig 2A: B. (A) The region of osteolysis (bounded by the white arrows) on the radiograph was correctly identified based on the (B) CT findings. Multiplying the 5 cm² radiograph area by 3.9 yields a volume estimate of 19.5 mL. This closely approximates the actual CT volume of 20.2 mL at 11.3 years followup.

Fig 3A: B. (A) The region of osteolysis on the radiograph was over-diagnosed compared with the (B) CT findings. Multiplying the 8 cm² radiograph area by 3.9 yields a volume estimate of 31.2 mL. This overestimates the actual CT volume of 22.5 mL at 8.1-year followup.

Fig 4A: B. (A) The region of osteolysis on the radiograph was under-diagnosed compared with the (B) CT findings. Multiplying the 1.1 cm² radiograph area by 3.9 yields a volume estimate of 4.3 mL. This dramatically underestimates the actual CT volume of 49.4 mL at 7.1 year followup. The black arrows bound a 0.6 mL preexisting cyst identified on both radiograph and CT.

Fig 5A: B. (A) The region of osteolysis identified on the radiograph closely approximates the (B) CT findings. Multiplying the 20.6 cm² radiograph area by 3.9 yields a volume estimate of 80.3 mL. This substantially overestimates the actual CT volume of 54.5 mL at 11.3 years followup.

Fig 6A: B. (A) The region of osteolysis on the radiograph closely approximates the (B) CT findings. Multiplying the 5.9 cm² radiograph area by 3.9 yields a volume estimate of 23 mL. This substantially underestimates the actual CT volume of 42.6 mL at 10-year followup.

All 92 cups in this series were radiographically stable. Groin or buttock pain occurred in 11 of the hips. However, the presence of pain was not associated with (p = 0.76) volume of osteolysis: the volume of osteolysis among those patients with groin or buttock pain (9.5 ± 17 mL; range, 0-56.3 mL) was similar to that of patients without groin or buttock pain (8.1 ± 13.8 mL; range, 0-57.7 mL).

DISCUSSION

The osteolytic lesions we examined comprise a sampling of the clinical defects we have encountered. Despite the presence of large osteolytic lesions, all of the cups in this series were radiographically stable and there was no evidence of pelvic fractures. Although some authors have reported osteolysis to be symptomatic⁴ and others asymptomatic⁷; we found larger lesions were not more likely associated with groin or buttock pain. Based on several studies, we currently regard CT as the gold standard for the clinical evaluation of pelvic osteolysis.^3,6,10,11 In work by Leung et al⁶ using specimens retrieved postmortem from patients who had THA and naturally occurring acetabular bone defects documented by physical sectioning, only 39% of the lesions were detected on AP pelvic radiographs. Combining the AP pelvic radiograph with aniliac oblique view increased the detection rate to 52%. Although the mean defect volume was 6.6 ± 11.9 mL and eight of 23 lesions had a volume of less than 1 mL based on measurements derived from the physical sections, CT imaging showed 87% of the bone defects. When the defects were identified on CT, the mean difference between the volume measurements from CT and the physical sections was 0.3 ± 1.1 mL for CT scans obtained without the femoral head and 0.9 ± 3.6 mL in a subset of CT scans done including the femoral head.

Our study was limited to one particular cup design that featured a single central dome hole. Although this design may predispose the hip to osteolytic lesions that develop behind the dome hole, 34 of the 63 hips (54%) in this study with pelvic osteolysis had lesions that communicated with the rim. Consequently, we believe the spectrum of lesions observed in this study is broadly representative of those encountered with noncemented metal-backed modular cups. When evaluating osteolysis on radiographs, we considered only the AP pelvic view. Although further study would be required to assess the relationship between 2D lesion area on multiple radiographic views and defect volume on CT, including additional radiographic views would likely improve sensitivity but could compromise specificity. Lastly, we did not evaluate the intraobserver or interobserver variability associated with the radiographic and CT interpretations of osteolysis. However, because we failed to prove our hypothesis that radiographic findings can be used to reasonably estimate osteolytic defect volumes, we believe the importance of intraobserver and interobserver variability is diminished. Even with perfect agreement, we would still conclude radiographic findings could not be reliably used to estimate osteolytic volumes. Poor repeatability would only underscore the inability to reliably estimate the defect volumes based on the radiographic findings. In our experiment design, we intentionally selected two different reviewers to ensure the results from the CT and radiographic analyses were independent. In our experience, when the same reviewer reads both the CT and radiographic images, there is a stronger correlation between the findings, presumably because the reviewer may recall the results of one analysis when they are performing the second.

Several studies have reported plain radiographs tend to underestimate the incidence and extent of periacetabular osteolysis, particularly when a single view is used.^{2,6,8,9,11,12} We confirmed this general observation, but found a modestly higher sensitivity and a lower specificity than previously reported.^2,6,8,11 This reflects a trend at our institution toward more aggressively identifying potential osteolytic lesions on radiograph based on our experience that suspected radiograph lesions are typically present on CT.^2,11 Like other studies from our institution, we found osteolytic lesions with a volume of at least 10 mL were more readily identified on radiograph.^2,11 Because larger lesions are more likely to be clinically important, we believe routine radiographs can be used to identify most patients who are candidates for CT scans to accurately evaluate the osteolytic lesion volumes. Since the sensitivity of radiographs is less than 100%, we currently recommend a baseline CT scan at 5-year followup to screen for early evidence of osteolysis among patients who may be at risk for developing the condition. For all patients, we recommend a CT scan at 10-year followup to definitively evaluate the presence of osteolysis.

The most substantial difference between our study and prior work was the finding of a correlation between 2D and 3D osteolytic lesion sizes. Using a cadaver model and simulated defects with a mean volume of 9.8 mL, Claus et al² found no correlation between 2D area on radiographs and the 3D volume. We attribute our findings to the fact the cadaveric defects created by Claus et al² would not have featured the sclerotic borders typically characterizing clinical osteolysis. In our experience, the sclerotic border is an essential feature that enables demarcation of the lesion margins on CT and radiographs. In the absence of a sclerotic border, the margins of the lesions would be difficult to discriminate on radiograph. In a clinical outcome study, Puri et al⁸ suggested the volumetric bone loss measured on CT did not correspond to the radiograph findings. However, among the hips with osteolysis in that study, the mean volume was only 4.9 mL. One hip had a 35.1-mL lesion, whereas the remaining 24 had volumes ranging from 0.3 to 10.8 mL. If we restrict our study population to those hips with osteolytic volumes of 10 mL or less, we also find no relationship between 2D area and 3D volume (r = 0.21; p = 0.08). In view of this observation, correlation depends on the range of values considered. The strength of the correlation in our study resulted from the inclusion of large osteolytic lesions. Among populations with a reduced incidence of osteolysis or a smaller average lesion volume, a weaker correlation between the radiograph and CT lesion sizes would be expected.

Although many clinical studies report correlation results, Bland and Altman¹ have demonstrated a correlation does not imply a diagnostic test is clinically useful. To better evaluate clinical utility, Bland and Altman¹ proposed evaluating the limits of agreement. When we compared estimates of the 3D CT volume based on the 2D radiograph area with the actual CT volume, we found the limits of agreement ranged from −14.6 to 18.7 mL. Similarly, we found the actual volume measured on CT could be estimated to within 10 mL for only 10 of the 22 THAs with osteolytic volumes of at least 10 mL. A review of the radiograph and CT findings demonstrated several reasons for the poor agreement. Radiographic findings can overestimate or underestimate the extent of osteolysis (Figs 3, 4). Although this can result from suboptimal radiograph image quality, another possible reason is larger osteolytic lesions often have multiple lobes. The superposition of these complex geometries can result in multiple areas with different apparent degrees of bone loss. We have also found the sclerotic border is not always prominent on radiographs, making it more difficult to distinguish the margins of the osteolysis. Even when the area on radiograph closely approximates the 3D findings oriented to simulate the radiograph view, variations in the actual geometry of the lesion mean the projected shape may not accurately estimate the actual volume (Figs 5, 6). Portions of the lesion obscured by the implant component projection on the radiograph can also contribute to differences between the volume estimates derived from the radiograph findings and the actual CT volumes.

Our goal was to better define the role of conventional radiographs and CT scans for the routine followup of THA patients. Although we found radiographs valuable for screening, we conclude volume estimates based on radiographic findings are not reliable and CT images are necessary to accurately measure lesion volumes. In our practice, we continue to recommend routine followup radiographs annually for the first 3 years after surgery and every 2 to 3 years thereafter. When osteolysis is suspected on routine followup radiographs, we recommend obtaining a CT to confirm the diagnosis and evaluate the lesion volume.