Detection of Osteomyelitis in the Diabetic Foot by Imaging Techniques: A Systematic Review and Meta-analysis Comparing MRI, White Blood Cell Scintigraphy, and FDG-PET

Most persons with longstanding diabetes develop peripheral neuropathy that, together with peripheral vascular disease (and microvascular dysfunction), often leads to foot complications. Patients with diabetes with these complications have an ∼25% lifetime risk of developing a foot complication (1,2), and these now appear to be the most common diabetes-related reason for hospitalization. At presentation, >50% of diabetic foot wounds are clinically infected (3). Most are initially soft-tissue diabetic foot infections (DFIs), but these often spread contiguously to underlying bone, resulting in diabetic foot osteomyelitis (DFO). DFO is now the most frequent cause of nontraumatic lower-extremity amputations that are associated with a 5-year mortality of ∼50% (4). Therefore, prompt identification and optimal treatment of DFO are important to help avoid poor outcomes (5).

Infection in the diabetic foot is defined by the presence of classic signs of inflammation and sometimes so-called “secondary” signs (3). Identifying DFO, which occurs in ∼20% of mild infections and >50% of severe cases (3,6,7), can be more difficult. This is related to the fact that: 1) DFO can occur in association with uninfected as well as infected ulcers; 2) radiographic changes in bone may be nonspecific and delayed for a few weeks after infection; and 3) patients with diabetes are also at risk for developing neuro-osteoarthopathy of the foot (Charcot foot [8,9]). The approach to treatment depends on proper diagnosis, as DFO usually requires antibiotic and surgical treatment, whereas Charcot disease requires proper offloading, sometimes with later surgical correction. Thus, diagnosing DFO requires a systematic approach that includes clinical, imaging, microbiological, and histopathological methods.

The most widely accepted criterion standard for diagnosing DFO is the presence of characteristic findings on histopathological examination and growth on culture of an aseptically obtained specimen bone (10). Bone biopsy, however, is an invasive procedure, and histology and culture are relatively expensive and time-consuming. Thus, it is important to determine which of the other available diagnostic tests for DFO might be appropriate in selected patients. Clinical examinations, such as the probe-to-bone test (11), inflammatory markers (especially the erythrocyte sedimentation rate) (12), and plain X-rays are nearly always the first steps in diagnosing DFO. In some cases, however, they fail to provide diagnostic results. In these situations, more advanced imaging techniques are often needed.

Imaging offers a complementary and less invasive, although often expensive, approach to diagnosing DFO, with a wide panel of modalities including: MRI, scintigraphy with ^99mTc-hexamethylpropyleneamineoxime (HMPAO) or ¹¹¹In-oxine–labeled white blood cells (^99mTc-HMPAO–WBCs or ¹¹¹In-oxine–WBCs) with single-photon emission computed tomography (SPECT/computed tomography [CT]), [¹⁸F]fluorodeoxyglucose positron emission tomography (¹⁸F-FDG–PET/CT), or ^99mTc-antigranulocyte antibody scintigraphy (13–18). Choosing the most appropriate advanced imaging modality must be based on not only the patient’s clinical presentation but also the equipment and expertise available at the treating center. Key factors include any recent or ongoing antibiotic therapy, the presence of neuropathic disease of the foot, the financial costs of various tests, the waiting time before imaging can be performed, any possible contraindications to the tests, and patient preference and likely adherence.

Of course, another key issue is which of the tests is most diagnostically useful. To determine the performance characteristics of the currently available advanced imaging tests for diagnosing DFO, we performed a systematic review and meta-analysis of the literature. The results of this review should help clinicians and organizations in preparing guidelines for the multimodality approach required for diagnosing DFO.

We conducted this systematic review in accordance with methods outlined in the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy (19) and have presented it following the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) guidelines (20).

We only included studies that: provided original data; were designed to provide information on diagnosis; were conducted in patients known or suspected to have both diabetes (type 1 or 2, regardless of method of glycemic treatment) and a foot infection; and were published in English. We considered studies for inclusion with either a prospective or retrospective design and blinded or nonblinded. Because of the risk of introducing bias, we excluded case-control studies, case reports, case series, and animal studies. We also excluded: reviews; articles on topics not germane to our study question (e.g., Charcot neuro-osteoarthropathy, magnetic resonance angiography for arterial disease, orthopedic implants, inflammatory markers, therapeutic studies, and studies on non–foot infections); articles focused on other index tests or using other less specific radiopharmaceuticals (⁶⁷Ga-citrate, labeled diphosphonates, labeled IgG, labeled ciprofloxacin, or labeled ubiquicidin); and articles that used currently outdated methodologies or protocols.

We searched the databases of Medline and Embase for studies published through August 2016. We used a combination of Medical Subject Headings terms and free-text words to define: our population of interest (persons with diabetes); the pathologic process of interest (osteomyelitis or infections of bone); and the specified imaging techniques used (MRI, ¹¹¹In-oxine–WBC SPECT/CT, ^99mTc-HMPAO–WBC SPECT/CT, or ¹⁸F-FDG–PET/CT). No studies with antigranulocyte antibodies could be included, because of the very limited application in diabetic foot diagnostics. Table 1 shows the extended query and complete search terms.

Three reviewers (C.L., M.T., and R.H.J.A.S.) independently screened all retrieved studies obtained based on their title and abstract. In a subsequent secondary screening, we evaluated the full text of the selected articles for eligibility. We jointly discussed, and resolved by consensus, any discrepancy among the reviewers that arose in study screening and selection. We also conducted a search of the references included in the retrieved articles seeking any additional potentially relevant articles.

Our preferred reference standard for the diagnosis of osteomyelitis (against which we compared the diagnostic performance of the target imaging techniques) was evaluation of a specimen of affected bone (collected by surgical or percutaneous biopsy) by histopathological review and/or culture. We only included studies that used this standard in our pooled estimation of diagnostic performance metrics (sensitivity and specificity). However, as it is common in clinical practice to use follow-up (history and physical examination, blood tests, and plain X-rays) to diagnose osteomyelitis, we also included studies that used this approach for just the calculation of positive and negative predictive values.

Two reviewers (M.T. and L.E.J.O.) independently extracted data from each study and evaluated them for quality using the QUADAS-2 method (19). They jointly discussed and resolved any discrepancies by consensus. The domains of interest in standardized data extraction were: population characteristics; imaging methods; reference standard for diagnosing osteomyelitis; descriptive and quantitative results (we generated a two-by-two contingency table for each imaging modality); and frequencies for final diagnoses (using RevMan v.5 for extraction).

For quality assessment, QUADAS-2 considers four main domains: risk of bias in patient selection (low, high, or unclear); index test; reference test; and, study flow and timing. We assessed the risk of bias and applicability concerns per imaging modality per study, then overall as follows: 1) low, if there was a low risk of bias in all key domains; 2) unclear, if we could not assess the risk of bias in one or more key domains; and 3) high, if the risk of bias was high for one or more key domains (20).

We used the STATA program (version 12.1; StataCorp), SPSS v.21 for Windows (IBM), and RevMan 5 for statistical analyses, setting statistical significance at P < 0.05. Using a hierarchical random effects model for binary data, we calculated a summary estimate of the diagnostic odds ratio (DOR) and 95% CIs per imaging technique. The odds ratio is a measure of effect size, describing the strength of association or nonindependence between two binary data values. The likelihood ratio (LR) we applied is used for assessing the value of performing the diagnostic imaging tests. The sensitivity and specificity of the test is used to determine whether a test result usefully changes the probability that a condition (such as the state of the diabetic foot, infected or not) exists.

We generated the summary receiver operating characteristic (sROC) curves and documented the sensitivity, specificity, and positive and negative LRs for every index test studied (i.e., MRI, ¹¹¹In-oxine–WBC SPECT/CT, ^99mTc-HMPAO–WBC SPECT/CT, or ¹⁸F-FDG–PET/CT). We assessed statistical heterogeneity among included studies using the I² statistic (21,22), which expresses the percentage of the variability that might be because of heterogeneity rather than sampling error.

Our search of PubMed and Embase identified 3,894 and 2,755 articles, respectively (Fig. 1). After screening each article based on its title and abstract, we assessed the full article for 38 and included 29 studies (23–51) in our meta-analysis. Among these studies, 13 were focused on MRI, 9 on ¹¹¹In-oxine–WBC, 10 on ^99mTc-HMPAO–WBC, and 6 on ¹⁸F-FDG–PET/CT. Two of the included studies were posters for which we could extract the data, but the information was insufficient to allow us to perform a bias assessment. For one of these posters, we obtained additional information from an author and used this to include the study in our meta-analysis.

Most of the included studies provided information on the population characteristics and key methodological aspects. Most studies were prospective, but some were retrospective, and some enrolled patients did not undergo all tests. Many included studies used the dual reference standard of clinical follow-up for test-negative patients and bone biopsy with subsequent pathological examination for test-positive patients. Many studies lacked data regarding the use of antibiotic therapy, anti-inflammatory drugs, or control of serum glucose levels. Only one reviewer assessed imaging results in most studies; in those with more reviewers, the authors rarely provided information regarding interobserver agreement.

In some studies, investigators used different protocols for the several imaging modalities. In most studies, the imaging protocols used did not conform to current standard acquisition procedures, and they were often combined with a bone scan (especially for those using ¹¹¹In-oxine–WBC), causing preselection of patients. Most of the WBC studies included in our review used only planar images acquired with fixed, single times and often failed to explain the acquisition protocol. The heterogeneity in imaging protocols of acquisition and interpretation is evident for WBC scan using both ^99mTc-HMPAO and ¹¹¹In-oxine.

Studies with PET mainly consisted of visual assessment of ¹⁸F-FDG uptake without any semiquantitative analysis (e.g., measuring the standardized uptake value). For studies using MRI, most scanners were of low magnetic field strength, and, although the articles were often outdated, they used correct protocols and sequences for identifying infections.

Figures 2 and 3 show the results of our assessment of methodological quality, based on the QUADAS-2 checklist (patient selection, index test, reference standard, flow, and timing) and our assessment on the risk of bias. Because we could not assess the two included posters adequately, we assessed them as unknown (Fig. 2, yellow) for all corresponding domains.

For each test, we have combined the sROC curves, and their corresponding findings, in Figs. 4 and 5, respectively. Per imaging modality, the findings are as described below.

This pooled analysis included 6 studies comprising 254 patients. The performance characteristics were: sensitivity, 89% (95% CI 68, 97); specificity, 92% (85, 96); DOR, 95 (18, 504); positive LR, 11 (4.7, 25.0); and negative LR, 0.11 (0.03, 0.4).

This pooled analysis included 9 studies comprising 206 patients. The performance characteristics were: sensitivity, 92% (72, 98); specificity, 75% (66, 82); DOR, 34 (6.9, 165.7); positive LR, 3.6 (1.9, 6.7); and negative LR, 0.1 (0.03, 0.4).

This pooled analysis included 10 studies comprising 406 patients. The performance characteristics were: sensitivity, 91% (95% CI 86, 94); specificity, 92% (78, 98); DOR, 118 (30, 459); positive LR, 12 (3.7, 36.3); and negative LR, 0.1 (0.06, 0.16).

This pooled analysis included 13 studies comprising 421 patients. The performance characteristics were: sensitivity, 93% (95% CI 82, 97); specificity, 75% (63, 84); DOR, 37 (11.3, 121.3); positive LR, 3.66 (2.1, 6.4); and negative LR, 0.10 (0.04, 0.26).

In summary, ¹⁸F-FDG–PET/CT and ^99mTc-HMPAO–WBC scans each had the highest specificity (92%), followed by MRI and ¹¹¹In-oxine–WBC scan (both 75%). The sensitivity for all of the imaging modalities were similar, with MRI at 93%; ¹¹¹In-oxine–WBC scans, 92%; ^99mTc-HMPAO–WBC scans, 91%; and ¹⁸F-FDG–PET/CT, 89%.

This meta-analysis documents the independent performance characteristics of the four imaging modalities most commonly used in the diagnosis of osteomyelitis of the foot in patients with diabetes (i.e., MRI, ¹¹¹In-oxine–WBC SPECT/CT, ^99mTc-HMPAO–WBC SPECT/CT, and ¹⁸F-FDG–PET/CT). To the best of our knowledge, this is the first meta-analysis and systematic review that included all of these imaging modalities, allowing us to assess their comparative diagnostic values. The most relevant finding of our analysis was the higher specificity compared with other imaging techniques for both ¹⁸F-FDG–PET/CT and ^99mTc-HMPAO–WBC scintigraphy (with either planar or SPECT/CT acquisitions). By contrast, the sensitivity was very similar for WBC-scan, ¹⁸F-FDG–PET/CT, and MRI.

The sensitivity and specificity of radiolabeled WBC reported in the articles we analyzed ranged from 75 (29) to 100% (30–32) and from 67 (32) to 100% (33), respectively. Factors that could have influenced the variations in results include: the number of patients imaged; the choice of imaging radiopharmaceutical; the acquisition protocol of the images; and the interpretation criteria followed for scan analysis (change of uptake with time and qualitative vs. semiquantitative analysis). Only two of the articles on ^99mTc-WBC (33,34) used a methodology confirmed and approved by the European Association of Nuclear Medicine (EANM) (52). In four articles (35,37,45,50), the protocols of acquisition and/or the interpretative criteria were not explained in the text. Although we are unable to make a direct comparison among these articles, the two conducted according to EANM procedural guidelines reported a sensitivity of 86 and 100% and a specificity of 100% (33,34), whereas the remaining four studies showed a lower average sensitivity and specificity (89 and 67%, respectively). If a currently approved protocol of acquisition and interpretation of images had been used, it is possible that the diagnostic accuracy would have been higher. We were unable to undertake a similar analysis for the studies of ¹¹¹In-labeled WBC, as all were performed with outdated imaging protocols.

Most of the studies included in our review used only planar images acquired with fixed, single times and often failed to explain the acquisition protocol. The heterogeneity in imaging protocols of acquisition and interpretation is evident for WBC scan using both ^99mTc- HMPAO and ¹¹¹In-oxine.

In only a few studies did the authors evaluate the value of semiquantitative analysis. Nawaz et al. (24) performed an analysis in a cohort of 110 subjects studied with both ¹⁸F-FDG–PET/CT and MRI. They used only visual assessment of ¹⁸F-FDG uptake (without any semiquantitative analysis) and did not perform CT coregistration, which might account for the relatively low sensitivity of the technique compared with MRI (81 and 91%, respectively). Conversely, Kagna et al. (25) demonstrated the value of performing a maximum standardized uptake value evaluation and CT coregistration to precisely evaluate the extension of infection into bone and soft tissue. This study was limited, however, by the fact that microbiological confirmation of infection was only performed in two cases (Fig. 2, yellow).

Familiari et al. (33), in accordance with EANM recommendations, used both qualitative and target/background ratio analysis and acquired images at three time points (30 min, 3 h, and 20 h postinjection), with time corrected for technetium decay. The performance characteristics for osteomyelitis (confirmed by histopathology or bone culture) using specific interpretation criteria were: sensitivity, 86%; specificity, 100%; positive predictive value; 100%, negative predictive value, 86%; and diagnostic accuracy, 92%. Similarly, Unal et al. (32) performed a target/background ratio for early and late images and found high sensitivity (94%) and specificity (100%). Other groups investigated the role of SPECT or SPECT/CT with different protocols of acquisition and, consequently, different results (34,35).

Compared with a systematic review of ¹⁸F-FDG–PET/CT for diagnosing DFO published 3 years ago (53), we found a higher sensitivity with a similar specificity; this appears mainly related to their using a per-patient analysis, whereas we used a per-study analysis. Nawaz et al. (24) concluded that ¹⁸F-FDG–PET/CT, although less sensitive when compared with MRI (81 vs. 91%), had higher specificity (93 vs. 78%) and diagnostic accuracy (90 vs. 81%). These results are comparable with the findings of our systematic review.

In contrast, our results differed from those in a meta-analysis by Kapoor et al. (54) of the diagnostic performance of MRI compared with WBC scintigraphy, bone scanning, and plain radiography. They concluded that MRI was superior to WBC scans, using the studies of Croll et al. (42) and Levine et al. (43). We also included these two studies, but we found a higher specificity for WBC scan labeled with ^99mTc-HMPAO than for MRI, probably because of the very small number of WBC scans they included against MRI (only 2 out of the 17 included articles of the patients received WBC). Also, studies of Kapoor et al. (54) are dated (>20 years old), using outdated WBC techniques (for instance, no late imaging after 24 h and lacking additional [hybrid low-dose] CT), resulting in moderate accuracy of WBC for this indication. In the future, we suggest exploring the appealing approach of hybrid imaging (55) in defining an optimal approach to imaging DFIs.

A final issue concerns the convenience of the different imaging modalities to the affected patient. In general, all modalities require that the patient be injected with a contrast agent or a radiopharmaceutical, followed by imaging with the specified scanner (PET, SPECT, or MRI). An advantage of MRI is the lack of exposure of the patient to a dose of radiation. Disadvantages of MRI include the fact that some patients experience claustrophobia in the (long) gantry and that this procedure is contraindicated in some patients who have an implanted device (e.g., implantable cardioverter defibrillator or cardiac pacemakers). Furthermore, the specificity of MRI is decreased by metal scatter in patients who have metallic hardware (e.g., screws or plates) in situ at the suspected site of infection. A key disadvantage of PET and SPECT are that they impart a radiation dose to the patient. Furthermore, these procedures are in general more expensive than MRI. The cost of each procedure is an important concern, not only for the patient but also for the health care system. Cost will, however, vary considerably over time, by the specific procedures used, and depending on the health care system and insurance issues germane to the patient under treatment. Furthermore, there is a difference between the actual cost and price (or charge) for any diagnostic test. For example, in many centers in the U.S., the amount charged for imaging tests is much higher than their costs, with the profits often used to offset losses related to other aspects of care.

Unfortunately, because of the limited number of published articles, we could not analyze the value of using antibodies against granulocytes (either whole IgG or F[ab] fragment), despite the fact that they are now routinely used in some nuclear medicine centers (16–18). In fact, the number of patients included in all of the available studies was relatively small, ranging from 6 to 110, with most enrolling <20. This leads to wide CIs, making comparisons between different techniques difficult and subgroup analyses or meta-regression impossible. Additionally, direct comparisons of all available imaging techniques within the same patient groups were not available, as none of the studies compared all four of the different techniques in the same study.

Many studies used the dual reference standard for osteomyelitis of clinical follow-up for test-negative patients and bone biopsy with subsequent pathological examination for test-positive patients. Despite the potential for introducing bias, we included all of these studies, as excluding them would significantly reduce the number of articles available, precluding performing a comparison among tests. We also believe that this approach is more representative of what is widely done in clinical practice.

As mentioned earlier, a key problem in diagnosing DFO is differentiating it from neuro-osteoarthropathy (Charcot foot) (56). Although some studies included Charcot foot as a separate diagnosis from soft tissue infection and osteomyelitis, we elected to leave this issue for a future review. Another factor is that the published articles often failed to provide information on the patients’ use of antibiotics or anti-inflammatory drugs and on their control of glucose levels. Thus, we cannot define the optimum time for imaging following antibiotic therapy or the possible effects of antibiotic or anti-inflammatory treatment, or glycemic control, on WBC or ¹⁸F-FDG–PET/CT scans. However, three prospective studies (57–59) have concluded that WBC scintigraphy performed under (or just after) antibiotic treatment retains a high sensitivity and specificity, perhaps even greater than for MRI for detecting residual disease (57). A retrospective study of 297 patients with suspected osteomyelitis or soft tissue infection by Glaudemans et al. (60), although not focused on the diabetic foot, suggested that there were no significant differences in performance results of WBC scintigraphy results between patients who were and those who were not receiving antibiotic therapy. We are unable to assess the possible effect of hyperglycemia on ¹⁸F-FDG–PET/CT findings, as glucose levels were often missing in the retrieved articles. In one study, however, hyperglycemia in a fasting state did not appear to significantly influence the quality of ¹⁸F-FDG–PET/CT imaging (61). Despite all of these limitations, including some biases, in general, there will be a preference for ^99mTc-HMPAO–labeled WBC scintigraphy and ¹⁸F-FDG–PET/CT, which are both techniques that proved their value in many other infectious diseases and offer the best highest specificity for diagnosing DFO. Sensitivity results were comparable among all imaging modalities.

Our systematic review and meta-analysis suggest that ^99mTc-HMPAO–labeled WBC scintigraphy and ¹⁸F-FDG–PET/CT offer the highest specificity for diagnosing DFO while demonstrating comparable sensitivity to the other imaging techniques we reviewed (MRI and WBC scintigraphy with ¹¹¹In-oxine). In view of the continued lack of consensus on this issue, we believe there is a need for a standardization of diagnostic methods and an evidence-based sequential approach. Selecting the most appropriate imaging test in any clinical situation depends upon the particular circumstances of the patient, the expertise, and equipment available at the treating site and the costs of the procedures. The goal is certainly to use the most cost-effective imaging method to allow accurate diagnosis and prompt treatment of this common, complex, and costly problem. This will require further prospective studies with larger numbers of patients and with a direct comparison between the different radiological and nuclear medicine techniques.

Acknowledgments. The authors thank N. Smidt for assistance and general advice for the review, S. van der Werf for helping conduct the literature searches, and Y. Takwoingi and M. Leeflang (all from the University Medical Center Groningen) for assisting with the data analyses and statistical evaluations.

Funding. R.H.J.A.S. is the recipient of a research grant from the University Medical Center Groningen Healthy Aging Pilot Fund.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. C.L. analyzed data and wrote the manuscript. M.T. analyzed data and reviewed the manuscript. A.W.J.M.G. designed the study and wrote and reviewed the manuscript. L.E.J.O. analyzed data and edited the manuscript. P.A.E. reviewed the manuscript. P.C.J. designed the study and wrote and reviewed the manuscript. B.A.L. reviewed and edited the manuscript. M.J.I. analyzed the data and reviewed the manuscript. A.S. reviewed the manuscript. R.H.J.A.S. designed the study and wrote, edited, and reviewed the manuscript. R.H.J.A.S. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.