Figures
Abstract
Diagnosing bone infection in its acute early stage is of utmost clinical importance as the failure to do so results in a therapeutically recalcitrant chronic infection that can only be resolved with extensive surgical intervention, the end result often being a structurally unstable defect requiring reconstructive procedures. [18F]-FDG-PET has been extensively investigated for this purpose, but the results have been mixed in that, while highly sensitive, its specificity with respect to distinguishing between acute infection and sterile inflammatory processes, including normal recuperative post-surgical healing, is limited. This study investigated the possibility that alternative means of acquiring and analyzing FDG-PET data could be used to overcome this lack of specificity without an unacceptable loss of sensitivity. This was done in the context of an experimental rabbit model of post-surgical osteomyelitis with the objective of distinguishing between acute infection and sterile post-surgical inflammation. Imaging was done 7 and 14 days after surgery with continuous data acquisition for a 90-minute period after administration of tracer. Results were evaluated based on both single and dual time point data analysis. The results suggest that the diagnostic utility of FDG-PET is likely limited to well-defined clinical circumstances. We conclude that, in the complicated clinical context of acute post-surgical or post-traumatic infection, the diagnostic utility accuracy of FDG-PET is severely limited based on its focus on the increased glucose utilization that is generally characteristic of inflammatory processes.
Citation: Brown TLY, Spencer HJ, Beenken KE, Alpe TL, Bartel TB, Bellamy W, et al. (2012) Evaluation of Dynamic [18F]-FDG-PET Imaging for the Detection of Acute Post-Surgical Bone Infection. PLoS ONE 7(7): e41863. https://doi.org/10.1371/journal.pone.0041863
Editor: Vance G. Fowler, Duke University Medical Center, United States of America
Received: May 25, 2012; Accepted: June 26, 2012; Published: July 30, 2012
Copyright: © Brown et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by grant AI069087 (MSS) from the National Institute of Allergy and Infectious Diseases. Support was also obtained from resources provided through the Clinical and Translational Sciences Award (RR0298884) to the University of Arkansas for Medical Sciences. Professional and financial affiliations for MS: Study funding from Cubist Pharmaceuticals and served on Unique Properties Advisory Board, Cubist Pharmaceuticals. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have received funding from a commercial source (Cubist Pharmaceuticals). This does not alter the authors‘ adherence to all the PLoS ONE policies on sharing data and materials.
Introduction
Osteomyelitis is a complicated infection that often requires surgical debridement as well as appropriate, long-term antimicrobial therapy. This is particularly true in chronic infections [1], which are characterized by a compromised blood supply and the formation of necrotic bone [2], [3]. Thus, the definitive detection of infection in its early, acute phase would be ideal. A number of noninvasive imaging modalities have been investigated in this regard [4], [5]. Plain radiographs are capable of detecting infection only after it has progressed to reveal the osteolytic lucency characteristic of chronic infection [6]. Magnetic resonance imaging (MRI) demonstrates higher sensitivity but its ability to detect infection in the setting of post-operative inflammation is limited, particularly in the context of metallic implants. Techniques based on in vitro radiolabeling and injection of a patient’s autologous leukocytes are capable of in vivo tracking to sites of infection but are hindered by the necessity of blood handling and its associated risks. Three-phase bone scans are exquisitely sensitive for the detection of bone remodeling resulting from multiple etiologies but are limited with respect to distinguishing between infection and other sources of inflammation [4], [6].
Positron emission tomography utilizing 2-deoxy-2-[18F]-fluorodeoxyglucose (FDG-PET) has also been investigated as an imaging modality for the diagnosis of osteomyelitis in the post-operative setting [7]–[10]. Some studies have concluded that FDG-PET could be used to differentiate recuperative remodeling from infection as early as 14–21 days post-surgery [8]–[10]. While these results are promising, the transition to a chronic state can occur in as few as 10 days post-infection [3]. This suggests that FDG-PET as it is generally applied in the clinical setting provides definitive diagnostic results only as the infection nears or progresses into the chronic stage. Indeed, based on a clinical study of 21 patients with confirmed osteomyelitis, Kalicke et al. [11] concluded that “in early post-operative follow-up it was impossible to differentiate between post-surgical reactive changes and further infection using FDG-PET”.
Because glucose consumption is increased in metabolically active cells, the uptake of FDG as a glucose analog is increased in rapidly dividing tumor cells and activated inflammatory cells. Thus, the problem with FDG-PET lies in its inability to distinguish between infection and sterile causes of increased glucose uptake. Recent studies have suggested that it might be possible to increase FDG-PET specificity by employing dual-time point imaging [12], [13], but to date this has not been evaluated in a well-defined experimental setting. The goal herein was to use an established rabbit model to comprehensively compare alternative methods for the acquisition and analysis of [18F]-FDG-PET data with a specific emphasis on distinguishing between infection and normal recuperative post-surgical inflammation during the therapeutically critical acute phase of infection.
Materials and Methods
Rabbit Osteomyelitis Model
All animal procedures were reviewed and approved by the Institutional Animal Care and Use Committee of the University of Arkansas for Medical Sciences. The experiments were carried out in a series of 3 trials, each of which consisted of 12 rabbits (6 infected, 6 uninfected). Thus, a total of 36 male New Zealand white rabbits underwent surgery as previously described [14]. In brief, a 1-cm mid-radial segment was surgically removed, injected with either 10 µl of sterile phosphate-buffered saline (PBS) or an equal volume of PBS containing 106 colony-forming units (cfu) of the Staphylococcus aureus osteomyelitis isolate UAMS-1, and replaced in its original orientation prior to wound closure. All animals were imaged as described below before being humanely euthanized 14 days post-surgery, at which time the surgical limb was harvested for bacteriological and histological analysis as previously described [14].
Quantitative histological and radiological analyses were done as previously described [14] with the results reported as the average ± the standard deviation for the infected (Inf) and uninfected (Cont) groups. Histological results were obtained only at the completion of the experiment while X-ray studies were performed at both 7 and 14 days.
Representative X-rays of the surgical site are shown for infected and uninfected rabbits as a function of time after surgery. Summed FDG-PET images below each X-ray illustrate uptake in both the surgical (R, surgical site illustrated by bracket) and non-surgical limbs (L) in an infected and an uninfected animal. “Regions of interest” (ROIs) used for measurement of SUV in each limb are illustrated by the circles.
Uptake in the surgical (top) vs non-surgical limb (bottom) is shown for the infected and uninfected control groups as a function of time after the administration of FDG. Results for each group obtained 7 and 14 days after surgery are shown as the average SUVmax (solid line) ± one standard deviation from the mean (shaded regions). Results represent the averages ± standard deviations obtained after combining all experimental animals in the 1st and 2nd trials (see text).
Imaging Studies
Radiographs of the surgical limb were taken on post-operative days 7 and 14 using a closed-system AXR small animal radiograph. X-rays were scored by an orthopedic surgeon who was blinded to the infection status of each rabbit and not involved in the analysis of any other data including that obtained by FDG-PET. Scoring was based on evidence of (a) periosteal elevation, (b) sequestration, (c) architectural deformation, and (d) deformation of soft tissue as previously described [14]. Each parameter was scored on a five-point scale (0–4), with 4 representing the most severe evidence of disease.
The ratio of SUVmax in the surgical vs. non-surgical limbs (SUR) was plotted as a function of time after the administration of FDG. Results for the infected and uninfected control groups 7 and 14 days after surgery are shown as the average SUVmax (solid line) ± one standard deviation from the mean (shaded regions). Results represent the averages ± standard deviations obtained from the 3rd experimental trial. Arrows along the axes indicate the timepoint (bottom) and SUR (left) at which the greatest discrimination between the two groups was observed.
Positron emission tomography with [18F]-FDG was performed immediately after radiography. Under anesthesia maintained by inhaled isofluorane, approximately 37–74 MBq (1–2 mCi, range 0.9–2.3 mCi) [18F]-FDG was injected intravenously into an ear vein. Dynamic acquisition was performed for 90 minutes using a small animal Concorde Focus 220 microPET system equipped with a 14×14 array of lutetium oxyorthosilicate crystals. This allowed for a 19 cm (transaxial) ×7.6 cm (axial) field-of-view and intrinsic spatial resolution of approximately 1.3 mm at the center of the field.
FDG-PET data was analyzed by a nuclear medicine physician who was also blinded to the infection status of each rabbit and unaware of radiographic scoring results. Raw PET data was reconstructed using filtered backprojection and binned into 2-minute increments beginning at time 0. Transverse, coronal, and sagittal reconstructed images were interpreted using AMIDE medical imaging software. Volumes-of-interest encompassing the surgical site in the right radius were drawn on reconstructed transverse images, and maximum standardized uptake values corrected for rabbit weight (in kg, range 2.6–3.4 kg) (SUVbw,max; hereafter referred to as SUV) were calculated at each timepoint. The ratio (SUR) of SUV in the surgical limb to SUV in the non-surgical limb was also calculated for each animal at each data acquisition timepoint, t, based on the formula SURt = [(SUV right, t - SUVleft, t)/SUV left, t]. Additionally, the change in SUV at the surgical site (ΔSUVt1:t2) was calculated based on the formula ΔSUVt1:t2 = (SUV right, t2– SUV right, t1)/SUV t1 where t1 and t2 are early and late timepoints, respectively, within the 90-minute data acquisition period. For simplicity in presenting results, we define SURt = ΔSUV0:t.
Left: Results illustrate the number of 1000 randomly-generated datasets in which the indicated timepoint after the administration of FDG exhibited maximum discrimination between the infected and uninfected experimental groups. Right: Results illustrate the proportion of “cut-off” ratios (surgical vs. non-surgical limb) that exhibited optimal discrimination at the alternative timepoints shown. Note that >850 of 1000 “iterations” found the optimum timepoint to be 11 minutes after administration of FDG (left) and that the optimal SUR cutoff in the majority of these was ∼1.27.
Left: Results illustrate the number of 1000 randomly-generated datasets in which the indicated timepoint after the administration of FDG exhibited maximum discrimination between the infected and uninfected experimental groups. Right: Results illustrate the proportion of “cut-off” ratios (surgical vs. non-surgical limb) that exhibited optimal discrimination at the alternative timepoints shown. Note that the number of “iterations” in which the greatest discrimination was observed were comparable when the assay was done 57 or 85 minutes after the administration of FDG and that the optimal SUR cutoff differed between these two timepoints.
Classifiers of infection status over 1000 iterations are plotted, with frequency of classifier identification as the best in any given iteration indicated by size of the dot for that classifier. On day 7, the 5 best classifiers all occurred when t1 = 0; that is, at a single timepoint SUR, with the single best classifier being <15 min after the administration of FDG. On day 14, all but one (δSUV at t1 = 15 min and t2 = 40 min) of the 5 best classifiers of infection status were a single timepoint SUR, with the single best classifier being >85 min after the administration of FDG.
Statistical Analyses
Data from the first two trials was examined individually to evaluate the variance within and between experimental trials. Because the variance between trials was small by comparison to the variance within each trial, data from these experiments was combined for further analysis. Mixed effect models were used to examine SUV and SUR trends over scan time. Separate analyses were performed for data obtained at day 7 and day 14. The fixed effects terms of the models included a factor representing infection status, linear and quadratic scan time terms, and terms representing the interaction between infection status and scan time, both linear and quadratic. Means and 95% confidence intervals for each experimental group were calculated for each scan time. An α-level of 0.05 was used to determine the statistical significance of the interaction and infection status effects.
Classification and regression tree (CART) methodology was used to generate simple prediction models based on SUR values. The SUR measures were recorded every 2 minutes for 90 minutes, producing 45 measurements per rabbit per week of the experiment. Data from trials 1 and 2 were used to create prediction models used to evaluate data obtained from rabbits during trial 3.
To assess how generalizable the simple models were, data from all three trials was combined and 12 rabbits randomly selected and held back as a “test” set. Data from the remaining rabbits were used as a “training” set on which the CART methodology was applied to define a simple model (i.e. the best scan time and SUR ‘cut point’ for distinguishing between infected and uninfected rabbits). This was repeated 1000 times to determine which scan times were most often selected as well as the associated SUR value identified as the cut-point.
The CART analysis described above was also expanded to include ΔSURt1:t2 measures as potential predictors to determine whether analysis based on multiple timepoints offered a diagnostic advantage. The process was again repeated 1000 times and, after each iteration, the ΔSURt1:t2 resulting in the “best” split was captured.
Results
Due to anesthesia complications during scanning, data was obtained for the entire 90 min scan period at day 7 for only 21 of 24 rabbits and at day 14 for only 20 of 24 rabbits included in trials 1 and 2. These failures included both infected and uninfected rabbits, with a complete dataset available for at least 10 rabbits from each group on each scan day.
Bacteriological and histological analysis confirmed the infection status of all rabbits based on the isolation of the specific strain used to initiate the infection and the presence of Gram-positive intra-osseous cocci respectively (data not shown). Additionally, overall histopathology scores were significantly higher in the infected vs. the uninfected experimental groups (Fig. 1). Radiographs revealed little difference between the infected and control groups 7 days after surgery, but the two groups diverged thereafter, with the overall difference reaching statistical significance (p<0.001) by day 14 (Fig. 1).
Representative summed FDG-PET images obtained 7 and 14 days after surgery are illustrated in Fig. 2 in the context of the comparative X-rays. FDG uptake at the surgical site increased over the entire 90-min imaging period in both the infected and uninfected experimental groups, but the average SUV at the surgical site in the infected rabbits was consistently higher than the average SUV in the uninfected controls throughout this imaging period at both 7 and 14 days (Fig. 3). In contrast, the average SUV was comparable in both experimental groups in the non-operative limb. During the early imaging period (post-injection time <20 min), SUV was higher at the operative site in infected rabbits at 7 days than at 14 days, but was essentially equivalent after 90 minutes.
Differences between the infected and uninfected groups, and differences within the infected group on day 7 vs. day 14, were also apparent when the analysis was done based on SUR rather than absolute SUV (Fig. 4). In general, the overall patterns confirmed that uptake was delayed at day 14 by comparison to day 7. The greatest discrimination between the two groups on day 7 occurred 11 min after administration of FDG, while on day 14 the greatest discrimination was observed at 85 min (Fig. 4). The SUR “cut point” that offered the greatest discrimination between the infected and uninfected groups also differed, with 1.27 being optimal at day 7 and 1.45 being optimal at day 14.
Based on these results, we carried out a 3rd independent experiment in which the infection status of the animal was predicted using two models. Model 1 defined an infected animal as one having an SUR ≥1.27 based on data obtained 11 min after administration of FDG and Model 2 defined an infected animal as one having an SUR ≥1.45 based on data obtained 85 min after administration of FDG. All diagnostic parameters were considerably higher when Model 1 was applied to data obtained 7 days after surgery and, conversely, when Model 2 was applied to data obtained 14 days after surgery. When Model 1 was used to analyze data obtained 14 days after surgery, or when Model 2 was used to analyze data obtained 7 days after surgery, all diagnostic parameters were considerably lower (Table 1).
CART analysis done with data from day 7 rabbits included in all 3 trials showed that Model 1 parameters (cut-point ≥1.27 at 11 min) offered the greatest discrimination in >88% of 1000 random iterations (Fig. 5). The results obtained when Model 2 was applied to data derived at day 14 were more ambiguous in that a significant proportion of the 1000 iterations suggested alternative cut-points derived at alternative times after the administration of [18F]-FDG (Fig. 6). However, all of these alternative parameters were chosen less frequently than those employed in Model 2, and this suggests that the alternative time points for analysis are unlikely to result in an improved diagnostic outcome.
To determine whether dual-time point analysis would allow for better discrimination between infected and uninfected animals, one-step CART analysis was extended to include comparisons with continuous values of ΔSUV. To simplify the analysis, data was examined at 5 min intervals throughout the 90-minute scan period and was pooled, and the 5 parameters yielding the best classification of infection status were identified for day 7 and day 14. On day 7 post-surgery, the best classifiers were all single time point SURs (t1 = 0) (Fig. 7) rather than any ΔSUV. On day 14, 4 of the 5 best classifiers were single timepoint SURs (t1 = 0) (Fig. 7). The single timepoint most often found to offer to the greatest discrimination differed on day 7 vs. day 14, with 10 min being the single best option for the former and 85 min being the single best option for the latter (Fig. 7).
We next analyzed the data from the 3rd experiment using an ‘either/or’ approach. Specifically, data obtained from all 12 rabbits on both scan days was combined into a single dataset (n = 24). This dataset was then analyzed based on defining an infected rabbit as one in which either the Model 1 or Model 2 parameters were met and, conversely, defining an uninfected rabbit as one in which neither of these parameters were met. Using this approach, all diagnostic parameters were found to be 75% (Table 1).
Discussion
All forms of bone infection present a therapeutic challenge, but this challenge is greatly magnified once the infection has progressed from an acute to a chronic state. Therefore, it is important to develop non-invasive diagnostic techniques capable of detecting infection in its earliest stages. The most definitive diagnostic method is microbiological culture, but in the specific context of bone and implant-associated infection this requires access to the site, and this can only be achieved using invasive procedures. Because such procedures themselves offer an opportunity for infection, it is imperative that they be employed judiciously. This accounts for the need to develop non-invasive diagnostic imaging modalities capable of both detecting infection in its early stages when it is more therapeutically amenable and distinguishing between infection and sterile forms of inflammation. In a previous study using a rabbit model to examine the diagnostic utility of static FDG-PET in the clinical context of post-surgical infection, we found that FDG-PET did offer promising diagnostic utility at least by comparison to plain-film X-ray, but only after the infection had progressed to a chronic stage as defined in the specific context of this rabbit model [9], [14]; prior to that time, it was not possible to reliably distinguish between infection and recuperative post-surgical inflammation.
In this report we attempted to build on this previous work to determine whether the diagnostic utility of FDG-PET during the critical acute phase of infection could be increased by employing alternative methods of data acquisition and analysis including dynamic imaging based on changes in SUV over time. We found that FDG-PET could distinguish between infection and normal post-operative healing with >90% accuracy as early as 7 days after surgery, which is potentially significant in that this is within the time frame of acute infection when it may still be possible to resolve the infection without additional surgical intervention [2], [3]. However, achieving this was dependent on the specific analysis method employed, with the critical issue being assessment of uptake within 10–12 min after FDG administration.
This is in contrast to the optimum method defined by our experiments done 14 days after surgery, which demonstrated that the best discrimination is obtained using data acquired ∼85 min after the administration of FDG. In this context it is important to note that X-rays also provided a reliable indication of infection status on day 14. To the extent that our previous studies using this same model demonstrated that X-rays provided reliable diagnostic information only after the infection had progressed to a chronic stage [9], [14], this suggests that the infected rabbits included in this study had developed changes characteristic of chronic infection by day 14. This relatively rapid time frame presumably reflects our use of the high infectious dose, an approach we employed to ensure that the infection was established in all experimental animals. Thus, when taken together, the observation that the optimum imaging time was significantly delayed by comparison to studies done 7 days after infection reflects delayed uptake owing to secondary changes related to chronic infection.
These results demonstrate that the diagnostic utility of FDG-PET is highly dependent on the status of the underlying infection and the specific method by which it is applied. This could be useful information in certain clinical settings, most notably in patients with findings concerning for infection in the early post-operative or post-traumatic period where the infection is likely to be in its acute phase and in which the timing of inoculation is known. This also raises the possibility that FDG-PET could be used not only to detect infection but also to stage the infection as an important consideration in planning the therapeutic approach. However, when we applied our optimized methods independently of the post-surgical interval, the diagnostic accuracy of FDG-PET was significantly reduced, thus confirming that FDG-PET has limited diagnostic utility when the length of time since infection onset is unknown. Additionally, while dynamic FDG-PET is now clinically feasible owing to improvements in instrumentation, our data does not support the hypothesis that dual-time point imaging offers a significant advantage over single time point imaging in osteomyelitis of unknown duration.
Finally, while certain of these results appear promising under well-defined clinical conditions, it must be noted that these results are based on a well-controlled experimental system with age-matched animals maintained on the same diet, and it is certain that this represents a best-case scenario by comparison to the diversity that would be encountered in individual patients. Indeed, given this diversity, it would be difficult to envision that infections would progress in a consistent timeframe in all patients. Thus, we conclude that the diagnostic utility of FDG-PET is likely to be limited beyond very well-defined clinical circumstances irrespective of the manner in which it is applied. This suggests that imaging tracers specific for bacteria, rather than host physiological features like inflammation, are likely to be of greater diagnostic value.
Author Contributions
Conceived and designed the experiments: TLYB KEB TBB RAS SGM MSS. Performed the experiments: KEB TLA RAS SGM MSS. Analyzed the data: TYLB HJS KEB TLA TBB WB JMG RAS SGM MSS. Wrote the paper: TYLB MSS.
References
- 1. Cierny G (2011) Surgical treatment of osteomyelitis. Plast Reconstr Surg. 127: 190S–204S.
- 2. Parsons B, Strauss E (2004) Surgical management of chronic osteomyelitis. Am J Surg. 188: 57–66.
- 3. Lew DP, Waldvogel FA (2004) Osteomyelitis. Lancet. 364: 369–379.
- 4. Gotthardt M, Bleeker-Rovers CP, Boerman OC, Oyen WJ (2010) Imaging of inflammation by PET, conventional scintigraphy, and other imaging techniques. J Nucl Med. 51: 1937–1949.
- 5. van der Bruggen W, Bleeker-Rovers CP, Boerman OC, Gotthardt M, Oyen WJ (2010) PET and SPECT in osteomyelitis and prosthetic bone and joint infections: a systematic review. Semin Nucl Med.. 40: 3–15.
- 6. Vijayanathan S, Butt S, Gnanasegaran G, Groves AM (2009) Advantages and limitations of imaging the musculoskeletal system by conventional radiological, radionuclide, and hybrid modalities. Semin Nucl Med. 39: 357–368.
- 7. Guhlmann A, Brecht-Krauss D, Suger G, Glatting G, Kotzerke J, Kinzl L, Reske SN (1998) Fluorine-18-FDG-PET and technetium-99m antigranulocyte antibody scintigraphy in chronic osteomyelitis. J Nucl Med. 39: 2145–2152.
- 8. Koort JK, Mäkinen TJ, Knuuti J, Jalava J, Aro HT (2004) Comparative 18F-FDG PET of Experimental Staphylococcus aureus osteomyelitis and normal bone healing. J Nucl Med. 45: 1406–1411.
- 9. Jones-Jackson L, Walker R, Purnell G, McLaren SG, Skinner RA, Thomas JR, Suva LJ, Anaissie E, Miceli M, Nelson CL, Ferris EJ, Smeltzer MS (2005) Early detection of bone infection and differentiation from post-surgical Inflammation using 2-deoxy-2[18F]-fluoro-D-glucose positron emission tomography (FDG-PET) in an animal model. J Orthop Res. 23: 1484–1489.
- 10. Mäkinen TJ, Lankinen P, Pöyhönen T, Jalava J, Aro HT, et al. (2005) Comparison of 18F-FDG and 68Ga-PET imaging in the assessment of experimental osteomyelitis due to Staphylococcus aureus. Eur J Nucl Med Mol Imaging. 32: 1259–1268.
- 11. Kälicke T, Schmitz A, Risse JH, Arens S, Keller E, Hansis M, Schmitt O, Biersack HJ, Grünwald F (2000) Fluorine-18 fluorodeoxyglucose PET in infectious bone disease: Results of histologically confirmed cases. Eur J Nucl Med. 27: 524–528.
- 12. Visioni A, Kim J (2011) Positron emission tomography for benign and malignant disease. Surg Clin North Am. 91: 249–266.
- 13. Sahlmann CO, Siefker U, Lehmann K, Meller J (2004) Dual time point 2-[18F]fluoro-2′-deoxyglucose positron emission tomography in chronic bacterial osteomyelitis. Nucl Med Commun. 25: 819–823.
- 14. Smeltzer MS, Thomas JR, Hickmon SG, Skinner RA, Nelson CL, Griffith D, Parr TR Jr, Evans RP (1997) Characterization of a rabbit model of staphylococcal osteomyelitis. J Orthop Res. 15: 414–427.