OBJECTIVE. Sarcoidosis lesions revealed on MRI in the axial skeleton and long bones resemble osseous metastases, which can lead to a potentially significant misdiagnosis. We hypothesized that osseous sarcoidosis lesions could be differentiated from osseous metastases on MRI and sought to propose and evaluate features distinguishing these entities.
MATERIALS AND METHODS. MR images obtained at 1.5 T of 34 subjects (22 with osseous metastatic disease, 12 with osseous sarcoidosis) with 79 single or multiple bone lesions (40 metastatic, 39 sarcoidal) were reviewed independently by two blinded, experienced musculoskeletal radiologists. Fluid-sensitive and T1-weighted images were viewed separately. Proposed discriminating features were peri- or intralesional fat, specified border characteristics, and the presence of an extraosseous soft-tissue mass. An additional feature for spinal lesions was posterior element involvement. On the basis of these criteria, the readers provided a binary diagnosis and confidence score.
RESULTS. The overall sensitivity for both readers was 46.3% and specificity, 97.4%. T1-weighted images were associated with higher sensitivity than T2-weighted images (59.0% vs 34.1%, respectively; p = 0.025) and with comparable specificity (97.6% vs 97.2%, p = 0.91). Diagnostic accuracy was higher using the discriminators of a mass or posterior element involvement for metastasis (83.3%) than border characteristics (68.0%) or lesion fat (65.0%) for sarcoidosis; the latter two features provided near 100% specificity but poor sensitivity (14.3% and 0%, respectively). Readers reported higher confidence diagnosing osseous sarcoidosis lesions than metastatic lesions, with a trend for higher confidence with T1-weighted images (p = 0.088).
CONCLUSION. Osseous sarcoidosis lesions cannot be reliably distinguished from metastatic lesions on routine MRI studies by readers experienced in evaluating these lesions.
Sarcoidosis is an inflammatory disorder of unknown cause. Although most commonly involving the lungs and lymph nodes, sarcoidosis can involve any organ system including the osseous skeleton . The diagnosis of sarcoidosis is made on clinical grounds and confirmed by the finding of noncaseating granulomas at tissue biopsy, although this finding cannot exclude other causes of granulomatous disease .
Sarcoidosis lesions have been classically described in the small bones of the hands and feet where they are seldom misinterpreted as metastases. However, on MRI in skeletal sites typical for metastases such as the long bones, pelvis, and vertebrae, sarcoidosis lesions appear as multiple round or irregularly shaped intramedullary lesions that are isointense or mildly hyperintense to muscle on T1-weighted images, are of variable signal intensity on fluid-sensitive sequences, and that enhance nonspecifically after IV administration of gadolinium chelate [3, 4]. This MRI appearance closely resembles that of osseous metastases [5, 6] (Figs. 1A and 1B). In cases with classic hilar sarcoidosis involvement, a presumptive diagnosis of sarcoidal bone lesions can be made on the basis of the clinical and imaging features. However, when osseous involvement is the presenting finding or is an incidental finding, the diagnosis can be considerably more problematic. With the expanded uses of MRI, osseous sarcoidosis is detected with greater frequency. To our knowledge, the literature has not reliably distinguished between the MRI features of osseous sarcoidosis and the MRI features of osseous metastases. We hypothesized that these lesions could be differentiated from one another on MRI and sought distinguishing imaging features.
Materials and Methods
The institutional review board waived the requirement for informed consent for this retrospective study. MR images of 22 subjects with metastatic disease (11 men and 11 women) and 12 patients with sarcoidosis (six men and six women) were included. The age range of the subjects with metastatic lesions was 42-88 years (average, 63.5 years), and the age range of the subjects with sarcoidosis lesions was 29-63 years (average, 50.3 years). All images were obtained of individuals either with biopsy-proven neoplasm metastatic to bone or with bone sarcoidosis lesions, except images obtained of one subject with pulmonary sarcoidosis in whom bone biopsy was nondiagnostic. This subject’s bone lesions had resolved without treatment on 6-month follow-up MRI and were presumed to be sarcoidosis on the basis of clinical findings. The metastatic lesions were from primary neoplasm of the breast (n = 7), prostate (n = 4), lung (n = 3), kidney (n = 2), and thyroid (n = 2); and from melanoma, alveolar soft part sarcoma, carcinoid tumor, and rectal tumor (1 each).
Images were selected from MRI studies performed from 2000 to 2008 with routine musculo-skeletal protocols at 1.5 T, randomly ordered in an image file composed of 39 images of sarcoidosis bone lesions and 40 images of metastatic bone lesions. The anatomic sites imaged were 51 spines (36 metastases, 15 sarcoidosis), 12 pelves (four metastases, eight sarcoidosis lesions) eight shoulders or humeri (sarcoidosis), seven femurs (sarcoidosis), and one sternum (sarcoidosis). To limit bias in selection of slice, spine images were obtained at approximately sagittal midline. All other images were selected through the slice providing the greatest diameter of the lesion or lesions.
The image set included 39 T1-weighted images and 40 fluid-sensitive images (T2-weighted with or without fat saturation or inversion recovery) that were randomly ordered in the file. The images were viewed individually, not side-by-side, at the same sitting. Images of bone lesions of the hands and feet were excluded because these sites were considered more characteristic of sarcoidosis and an uncommon site for metastases. Because contrast administration was not performed in many cases, we decided to exclude enhanced images from the sample.
On the basis of our experience evaluating MRI of osseous sarcoidosis, we proposed three potential discriminating features: the presence of intra- or perilesional fat (Fig. 2), lesion border characteristics (Figs. 3, 4, 5A, and 5B), and the presence of an associated soft-issue mass including posterior element involvement for spinal lesions (Figs. 6A and 6B). The discriminator border characteristic was evaluated as an overall category and was also evaluated as the following three descriptors: sharply defined, brushlike, and poorly defined. Peri- or intralesional fat was proposed as a distinguishing feature because we had observed this finding on MRI in cases of bone sarcoidosis and because this feature is not commonly seen in metastatic lesions that displace or replace marrow fat. A brushlike border interdigitating with surrounding fatty marrow was a proposed distinguishing feature because we observed it on MRI in sarcoidosis cases. An extraosseous mass and, for spinal lesions, posterior element involvement were selected as described criteria for malignancy .
The images were evaluated independently by two musculoskeletal radiologists, with 18 and 22 years’ experience, respectively, who were cognizant that sarcoidosis lesions resemble metastatic lesions on MRI. The readers were informed that they were to evaluate 79 MR images of lesions that were either bone metastases or bone sarcoidosis lesions, but they were blinded to additional clinical information and were unaware of the relative percentages of each lesion type. For each image, the reader provided a binary diagnosis (sarcoidosis or metastatic lesions) and a 3-point ordinal assessment of their confidence in their diagnosis (2 = definitely the binary diagnostic choice, 1 = probably the binary diagnostic choice, 0 = unsure). The reviewers were asked to note which of the proposed differentiators they had used for diagnosis. The readers were also given the option to write in a different reason for their choice than the proposed criteria. This option was included to assay for differentiating features occurring to the readers that might have been overlooked in the provided criteria. Write-in reasons and cases in which readers provided no rationale for their decisions were compiled in a fourth grouping labeled “Other.” The pulse sequence (T1-weighted or fluid-sensitive T2-weighted) used for each image was noted.
Images of subjects with biopsy-proven osseous metastases were defined as reference standard positive. The sensitivity, specificity, and overall accuracy associated with the binary diagnostic choice based on each stated imaging criterion and the pulse sequences used were assessed using logistic regression for correlated data. Specifically, generalized estimating equations (GEEs) based on a binary logistic regression model were used to model concordance between the imaging-based and biopsy-proven diagnoses as a function of pulse sequence (T1 vs T2) and reader identification (to control for systematic differences between readers) while accounting for the correlation between results provided for the same patient. An exact McNemar test was performed to compare readers in terms of sensitivity, specificity, and their propensity for each diagnosis (metastasis, sarcoidosis).
Interreader agreement in terms of diagnosis was assessed using kappa coefficients. An exact Wilcoxon matched-pair signed rank test was used to compare readers with respect to confidence scores. An exact Mann-Whitney test was used to compare sequences and diagnosis groups (positive vs negative findings on reference standard) in terms of diagnostic confidence scores represented for each subject and as an average over readers.
The overall reader sensitivity was 46.3% (37/80) and specificity, 97.4%. The overall positive predictive value was 94.9% (37/39), and the negative predictive value was 63.9% (76/119).
Reader 1 chose the diagnosis of sarcoidosis in 81% (64/79) of the readings, and reader 2 chose the diagnosis of sarcoidosis in 70% (55/79) of the readings. Reader 1 diagnosed metastasis correctly in 35% (14/40) of the readings and diagnosed sarcoidosis correctly in 97.4% (38/39) of the readings. Reader 2 diagnosed metastasis correctly in 58% (23/40) of the readings and diagnosed sarcoidosis correctly in 97.4% (38/39) of the readings.
The two readers provided concordant diagnoses 75.9% (60/79) of the time, yielding a kappa coefficient of 0.364, which represents fair agreement. Agreement between the readers occurred for 95% (37/39) of the sarcoidosis cases but for only 57.5% (23/40) of the metastases. A positive diagnosis was provided more frequently by reader 2 (30.4% positive calls, 24/79) than by reader 1 (19%, 15/79). Although this difference was not quite significant (p = 0.064, McNemar test), it did result in reader 2 showing significantly higher sensitivity (23/40 = 57.5%, McNemar test) than did reader 1 (14/40 = 35%). There was no difference (p = 1.0, McNemar test) between readers in terms of specificity.
Impact of Pulse Sequence on Image Interpretation
The sensitivity, specificity, and positive and negative predictive values based on the pulse sequence used are summarized in Table 1. Using GEE, we found that the T1-weighted sequence was associated with significantly higher sensitivity than the fluid-sensitive or T2 sequence (59.0% vs 34.1%, p = 0.0253) whereas there was no difference between sequences in terms of specificity (97.6% vs 97.2%, p = 0.911); however, that difference between sequences was significant only for reader 2. For reader 2, sensitivity was significantly higher (p = 0.0028) for T1-weighted images (17/21 = 81.0%) than for T2-weighted images (6/19 = 31.6%), whereas for reader 1, the sensitivity of T1 (6/18= 33.3%) was not significantly different (p = 0.8416) than that of T2 (8/22 = 36.4%).
Reason for Diagnosis
The sensitivity, specificity, and overall accuracy associated with diagnoses based on each provided criterion are summarized in Table 2. The GEE analysis showed that diagnostic accuracy was significantly higher when the stated reason for the diagnosis was mass or posterior element involvement than when the noted reason was lesional or perilesional fat (p = 0.0366), border characteristics (p = 0.0478), or “other” (p = 0.0334). There were no other significant differences among reasons in terms of accuracy. The discriminators lesion fat and border characteristics were useful for the exclusion of metastatic disease because they led to perfect specificity but poor sensitivity. Only mass or posterior element involvement was associated with a sensitivity of greater than 60%. The reason termed “other” included “write-in” reasons for diagnostic choices and all cases in which no explicit reason for the diagnosis was noted. Additional reasons written in by the readers were multiplicity (n = 3), segmental appearance (n = 5), and diffuse disease (n = 2). No reason for diagnostic choice was provided for 28 of 158 readings (17.7%).
The readers’ confidence scores correlated with pulse sequence are summarized in Table 3. Each reader’s confidence scores correlated with the proposed discriminators are summarized in Table 4. There is no significant difference between readers (p = 0.698) in the level of confidence. The readers reported significantly higher confidence (p < 0.001) when evaluating reference standard-verified sarcoidosis cases (mean ± SD, 1.14 ± 0.55) than when evaluating verified metastases (0.78 ± 0.66). When the readers were evaluating reference standard-verified sarcoidosis cases, there was a trend for higher reader confidence when a T1-weighted sequence was used, but the difference between sequences did not quite achieve statistical significance (p = 0.088). There was no difference in the confidence levels associated with the two sequences when readers evaluated metastases (p = 0.846) or when averaged over all patients (p = 0.249). The readers reported significantly higher confidence scores for reference standard-verified sarcoidosis cases than for subjects with metastases when the reason stated for diagnosis was border characteristics (p = 0.0004), lesional fat (p = 0.0095), or poor margination (p = 0.0009).
The best known pattern of sarcoidosis bone involvement is latticelike osteolysis identified on radiographs of the hands and feet [8, 9]. On radiography and CT, sarcoidosis bone lesions can be mixed, lytic, or sclerotic but also may be undetectable, particularly in the axial skeleton and long bones. In a retrospective evaluation of sarcoidosis patients with musculoskeletal symptoms, 20 MRI studies revealing bone lesions were correlated with radiographs. Fairly equivalent radiographic and MRI findings were noted in 55% of cases, with less conspicuous or inconspicuous radiographic findings in 45% . In clinical practice, radiographic correlation for bone lesions revealed on MR images should be performed if radiographic images are available.
Sarcoidosis bone lesions show variable uptake on bone scintigraphy . FDG uptake on PET can be seen in skeletal sarcoidosis . Multifocal skeletal sarcoidosis may present as a false-positive for osseous metastases on PET, as described in case reports [13, 14].
In the literature, case reports and reviews describe the MRI appearance of osseous sarcoidosis as closely resembling that of osseous metastases [3-6, 15-25]. The diagnosis of sarcoidosis has usually been established clinically before bone lesions are detected at MRI. Uncommonly, sarcoidosis bone lesions resembling bone metastases on MRI may be the initial presentation of undiagnosed disease [3, 5, 6] or may be revealed as an incidental finding in an asymptomatic subject imaged for routine indications . When interpreting an MRI study revealing osseous sarcoidosis lesions, the radiologist may not be aware of a clinical history of sarcoidosis or may not consider that diagnosis for multiple intramedullary lesions. A diagnosis of osseous metastases in a sarcoidosis patient with multiple axial or long bone lesions on MRI may potentiate a search for a primary lesion, distressing the patient and resulting in a clinical dilemma.
By prevalence, sarcoidosis bone lesions resembling metastases are expected to be encountered much less frequently than bone metastases on routine MRI studies. Skeletal metastasis is the most common malignancy of bone in adults and is the third most common site of metastatic involvement after lung and liver . The incidence and prevalence data for sarcoidosis range widely among studies, from 0.03 to 640 per 100,000 . In a review of the published data on the epidemiology of sarcoidosis, Thomeer and colleagues  ascribe the wide range of reported prevalences to selection bias, noting that assessment of incidence and prevalence is further complicated by the large percentage of asymptomatic individuals presenting with incidental findings at imaging. The estimated prevalence of osseous involvement in subjects with sarcoidosis derived from retrospective radiographic series ranges from 1% to 13%, averaging 5% . Because long bone and axial skeleton sarcoidosis lesions revealed on MRI may be radiographically occult, studies based on radiographic series probably underestimate the prevalence of bone sarcoidosis.
Our study shows that sarcoidosis and metastatic bone lesions cannot be accurately differentiated by experienced musculoskeletal radiologists utilizing proposed discriminators and educated “hunch” (i.e., no provided reason). Specificity was high but overall sensitivity was only 46.3%. The discriminators lesional or perilesional fat and border characteristics were useful for the exclusion of metastatic disease because they led to perfect specificity but provided poor sensitivity. Lesion border characteristics vary widely in both sarcoidal and metastatic bone lesions and overlap considerably. The presence of intralesional fat has been described as a feature excluding malignancy . In the case of sarcoidosis bone lesions, intralesional fat may reflect lesion involution. Mass or posterior element involvement was posited to militate for metastatic disease and was the most sensitive criterion among the criteria evaluated. Cases of vertebral sarcoidosis with a paravertebral mass or posterior element involvement are considered rare but have been reported in the literature [23-25].
To achieve adequate statistical power for an assessment of sensitivity and specificity, we constructed our image sample to have a near-even ratio of metastatic-to-sarcoidosis lesions. Because estimates of positive and negative predictive values depend on the prevalence of the condition being diagnosed and because the prevalence of sarcoidosis in our sample is higher than would be expected in a random sample, our sample provides biased estimates of predictive values. The readers were unaware of the near-even ratio of sarcoidosis-to-metastatic lesions in the image file and together chose the diagnosis of sarcoidosis in 75% of readings. This result suggests that either the readers “adjusted” to unstated oversampling of sarcoidosis cases or the readers were biased for that diagnosis when informed of that clinical possibility.
The T1-weighted sequence was associated with significantly higher sensitivity for the detection of metastatic lesions than the fluid-sensitive T2-weighted sequence. This finding is expected because hypercellular bone lesions such as metastases show long T1 relaxation times, whereas T2 relaxation times vary by tumor morphology . Conventional T2-weighted images usually show bone sarcoidosis as low to intermediate in signal intensity, whereas fluid-sensitive fat-saturated sequences typically show moderately hyperintense lesions.
Limitations of our study include that it is retrospective and is thus prone to selection bias. Sarcoidosis images were drawn from a small file and metastatic images were drawn from a larger file, which may have introduced selection bias. Most images of metastases were of spine, whereas the sarcoidosis file was composed of more varied anatomic sites. Readers may have performed differently if all skeletal sites were available for both sets of lesions. Lesion size varied for both cohorts but was not measured to assess for discrepancies. Showing T1-weighted and fluid-sensitive images side-by-side rather than randomly ordered in the reading file would have more closely paralleled routine assessment of intramedullary lesions on MRI and might have allowed a more accurate assessment of lesion fat. The younger average age of the sarcoidosis cohort relative to the metastasis cohort might have affected the conspicuity of intramedullary lesions on MRI.
We decided not to include enhanced images in the image file because images of many of the sarcoidosis and metastases cases were obtained without contrast administration, limiting the number of enhanced examples available for evaluation. Often cases that incidentally reveal sarcoidosis lesions have been imaged without contrast material for routine indications (e.g., back pain). The administration of IV gadolinium for MRI of bone metastases varies by institution and many centers do not routinely administer contrast material for nonprimary bone lesions. In our clinical experience, when gadolinium contrast is administered, sarcoidosis bone lesions enhance in a nonspecific manner that does not improve differentiation from bone metastases on MRI. We are not aware of any references to a characteristic or pathognomonic appearance on enhanced MR images of bone sarcoidosis or to references on the utility of contrast material in bone sarcoidosis in the literature.
Sarcoidosis may affect multiple organ systems; on an MRI examination revealing bone lesions, the surrounding soft tissues may show findings that are diagnostically relevant such as splenomegaly, hepatic lesions, cardiac involvement, regional lymphadenopathy, and neurosarcoidosis. Contrast enhancement patterns may aid in the detection and in the differential diagnosis in some cases—for example, with sarcoidosis muscle nodules and leptomeningeal enhancement . This possibility argues for contrast administration in suspected bone sarcoidosis cases.
Sarcoidosis bone involvement should be considered in the differential diagnosis of multiple intramedullary lesions detected on MRI studies of patients with suspected or proven sarcoidosis. This consideration tempers the potential misdiagnosis of metastatic disease and allows the radiologist and clinician to discuss further diagnostic workup. Detection of otherwise occult sarcoidal skeletal lesions at MRI can alter the clinical assessment of granulomatous load and may influence treatment. If multifocal bone lesions are revealed on MRI in a patient without a known diagnosis of primary neoplasm or sarcoidosis, after clinical correlation a chest CT could be suggested to assess for characteristic lymphadenopathy and parenchymal changes.
The importance of biopsy in assessing multifocal bone lesions in patients with sarcoidosis has been emphasized in several case studies [5, 6, 13, 18, 32]. It should be noted that sarcoidosis bone lesions revealed on MRI may be occult on CT, making CT-guided biopsy difficult. In cases with the diagnosis of sarcoidosis and question of a primary tumor (e.g., equivocal breast lesion, nonspecific lung nodule), awareness that sarcoidosis lymphadenopathy, hepatic involvement, and bone lesions may mimic metastatic disease should spur meticulous investigation [33, 34]. In subjects with established dual diagnosis of sarcoidosis and primary tumor, the morphology of bone lesions on MRI is not accurate to distinguish between these lesion types.
In summary, in cases of multifocal bone lesions on MRI and an established diagnosis of sarcoidosis, a diagnosis of bone sarcoidosis can be made presumptively; however, differentiation from metastasis cannot be achieved by morphologic criteria. Discussion between the radiologist and clinician on a per-case basis is recommended to determine the advisability of follow-up imaging or biopsy.
We thank David Naidich for his helpful suggestions. We wish to express our gratefulness to the late Alvin Teirstein for his inspiration and guidance.
Siltzbach LE, James DG, Neville E, Turiaf J, Battesti JP, Sharma OP. Course and prognosis of sarcoidosis around the world. Am J Med 1974; 57:847–852
Milman N, Lund JO, Graudal N, Enevoldsen H, Evald T, Nørgård P. Diagnostic value of routine radioisotope bone scanning in a series of 63 patients with pulmonary sarcoidosis. Sarcoidosis Vasc Diffuse Lung Dis 2000; 17:67–70
Baldini S, Pupi A, Di Lollo S, Marchionni N, Shraim R, Bosi A. PET positivity with bone marrow biopsy revealing sarcoidosis in a patient in whom bone marrow metastases had been suspected. Br J Haematol 2008; 143:306
Thomeer M, Demedts M, Wuyts W. Epidemiology of sarcoidosis. In: Drent M, Costabel U, eds. Sarcoidosis: European Respiratory Society Monograph. Sheffield, UK, European Respiratory Society, 2005; 32:13–22
Simpfendorfer CS, Ilaslan H, Davies AM, James SL, Obuchowski NA, Sundaram M. Does the presence of focal normal marrow fat signal within a tumor on MRI exclude malignancy? An analysis of 184 histologically proven tumors of the pelvic and appendicular skeleton. Skeletal Radiol 2008; 37:797–804