Textural features of 18F-fluorodeoxyglucose positron emission tomography scanning in diagnosing aortic prosthetic graft infection
- 1.2k Downloads
The clinical problem in suspected aortoiliac graft infection (AGI) is to obtain proof of infection. Although 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography scanning (PET) has been suggested to play a pivotal role, an evidence-based interpretation is lacking. The objective of this retrospective study was to examine the feasibility and utility of 18F-FDG uptake heterogeneity characterized by textural features to diagnose AGI.
Thirty patients with a history of aortic graft reconstruction who underwent 18F-FDG PET/CT scanning were included. Sixteen patients were suspected to have an AGI (group I). AGI was considered proven only in the case of a positive bacterial culture. Positive cultures were found in 10 of the 16 patients (group Ia), and in the other six patients, cultures remained negative (group Ib). A control group was formed of 14 patients undergoing 18F-FDG PET for other reasons (group II). PET images were assessed using conventional maximal standardized uptake value (SUVmax), tissue-to-background ratio (TBR), and visual grading scale (VGS). Additionally, 64 different 18F-FDG PET based textural features were applied to characterize 18F-FDG uptake heterogeneity. To select candidate predictors, univariable logistic regression analysis was performed (α = 0.16). The accuracy was satisfactory in case of an AUC > 0.8.
The feature selection process yielded the textural features named variance (AUC = 0.88), high grey level zone emphasis (AUC = 0.87), small zone low grey level emphasis (AUC = 0.80), and small zone high grey level emphasis (AUC = 0.81) most optimal for distinguishing between groups I and II. SUVmax, TBR, and VGS were also able to distinguish between these groups with AUCs of 0.87, 0.78, and 0.90, respectively. The textural feature named short run high grey level emphasis was able to distinguish group Ia from Ib (AUC = 0.83), while for the same task the TBR and VGS were not found to be predictive. SUVmax was found predictive in distinguishing these groups, but showed an unsatisfactory accuracy (AUC = 0.75).
Textural analysis to characterize 18F-FDG uptake heterogeneity is feasible and shows promising results in diagnosing AGI, but requires additional external validation and refinement before it can be implemented in the clinical decision-making process.
Keywords18F-FDG PET Aortic prosthetic graft infection Textural features
Aortoiliac prosthetic graft infection (AGI) is a severe complication after prosthetic graft placement, which is associated with high morbidity and mortality [1, 2, 3, 4, 5]. Positive bacterial cultures, either from percutaneous aspirated perigraft fluid or from surgically obtained material, are considered the gold standard for diagnosing AGI [1, 2, 3, 4, 5]. However, a perigraft abscess or fluid collection is not always present, and even if present it may not always be suitable for puncture. The diagnosis of AGI, therefore, remains an important challenge.
Non-invasive 18F-fluorodeoxyglucose positron emission tomography (18F-FDG PET) imaging may play an important role in increasing the diagnostic accuracy of infectious diseases with inherent elevated intracellular glucose metabolism [6, 7, 8, 9, 10, 11]. The PET images can either be evaluated semi-quantitatively using the maximal standardized uptake value (SUVmax) and the tissue-to-background ratio (TBR), and visually using the visual grading scale (VGS). For diagnostic assessments which are based on a combination of these parameters, our group reported a high sensitivity (up to 91%), but a low specificity (up to 64%) , which was also confirmed by others [6, 7, 11]. Additionally, there is no consensus with respect to the interpretation of these 18F-FDG PET findings. Recently, SUVmax >8 in the perigraft area was suggested as the cutoff value for proof of an infection of a thoracic prosthetic graft . However, this value must be interpreted with caution as this study was conducted in only nine patients and scans were not acquired according to European Association of Nuclear Medicine (EANM) recommendations . Moreover, our group found the above-mentioned quantitative measures when used as so-called cutoff values to determine infection to be of modest utility in the diagnosis of AGI . As a consequence, more accurate quantification methods are needed. Since a heterogeneous uptake is associated with infection, the distribution pattern of 18F-FDG activity may help identify AGI with a higher diagnostic precision . A powerful tool for quantifying such distribution is textural analysis, which may provide valuable information regarding biological heterogeneity. The concept of textural analysis is generally based on the spatial arrangement of voxels in a predefined volume of interest (VOI). Spatial heterogeneity can be depicted from different spatial interrelationships on 18F-FDG PET scans. Within the field of clinical oncology, textural analysis already has yielded promising results in predicting response by quantifying intra-tumoral heterogeneity [15, 16, 17, 18, 19, 20, 21, 22, 23].
In the current study, we introduced the principle of textural analysis into the field of infectious diseases and aimed to investigate feasibility and accuracy of textural features to diagnose AGI.
Design of the study
Patient characteristics and information about the initial operation, type of graft material, clinical symptoms and laboratory parameters at the time of 18F-FDG PET imaging, and definite type of treatment were collected from the medical records. Co-morbidities were defined as recommended by the Ad Hoc Committee on Reporting Standards . This study was approved by the institutional ethical review board (METc 2015/082). Patients’ data were analysed anonymously.
18F-FDG PET imaging and analysis
Non-gated PET/CT imaging was performed with a dedicated integrated PET/CT system (Biograph mCT PET/CT, Siemens, Knoxville, TN, USA). All patients fasted overnight with no restrictions on drinking water and with a minimum fasting time of 6 h prior to PET/CT. 18F-FDG was administered intravenously with a weight-based activity of 3 MBq/kg. Sixty minutes after tracer injection, patients were positioned on the camera table with the arms in upright position. PET images were acquired with 3 min per bed position. An initial low dose CT scan was performed to ensure that the region of interest was included in the field of view, where after an inspiration breath-hold low-dose CT for attenuation correction was performed with 100 kVp and 30 mAs. Image data were reconstructed using standard methods and images were standardized according to EANM guidelines .
An experienced nuclear medicine physician assessed the 18F-FDG PET images, including VGS, SUVmax and TBR. The five-point VGS was graded as follows: grade 0, 18F-FDG uptake similar to that in the background; grade I, low 18F-FDG uptake, comparable with inactive muscles and fat; grade II, moderate 18F-FDG uptake, clearly visible and higher than uptake by inactive muscles and fat; grade III, strong 18F-FDG uptake, but distinctly less than the physiologic urine bladder activity; and grade IV, very strong 18F-FDG uptake, comparable with the physiologic urinary activity of the bladder [10, 25]. A VOI was drawn around the area of the vascular prosthesis to calculate the SUVmax. SUVmax corresponded to the voxel with the highest 18F-FDG uptake. The TBR was defined as the SUVmax divided by the mean SUV of the caval vein (blood pool).
Volume of interest
Textural features extraction
Figure 2d and supplemental table 1 provide a full overview of all 64 analysed textural features. We extracted 19 first order textural features (based on the grey level distribution, but without spatial information of voxels). Texture can be characterized by replications of (small) texture elements. These texture elements consist of contiguous voxels with certain spatial and intensity properties. We obtained the distributions of three different texture elements, i.e. the grey level co-occurrence (or spatial dependence) matrix (GLCM) for pairwise arrangement of voxels , the grey level run-length matrix (GLRLM) for alignment of voxels with an identical intensity , and the grey level size-zone matrix (GLSZM) for characteristics of homogeneous zones . From these matrices, we extracted 46 s order textural features (which are thus based on spatial information of the grey levels). These textural features were extracted with a voxel-to-voxel distance offset of d = 1 and directional analysis was performed with a connectivity of 26 voxels (analysis in 13 angular directions). All extracted textural features were normalized to the range [0,1]. To determine the influence of noise, it was tested whether noise was equally distributed among the groups and the correlation between noise and each textural feature was computed. Therefore, a sphere of 3 cm in diameter was drawn in the liver; the coefficient of variation was determined as noise parameter.
Baseline characteristics are presented as mean ± standard deviation or percentages. To select candidate predictors to identify infection, univariable logistic regression analysis was performed. All potential predictors that met the Akaike Information Criterion (AIC) were considered significant . To discourage overfitting, the AIC is based on rewarding goodness of fit and penalizing complexity in the model. The AIC requires χ2 > 2 df, i.e. when considering a predictor with one degree of freedom df; this implies a significance level α = P(χ2 ≥ 2) = 0.16 . The accuracy of all candidate predictors was measured by the area under the receiver operating characteristic curve (AUC). Textural features were considered to have a good accuracy in case of an AUC > 0.8. Moreover, textural feature values may be subject to inter-observer variability in delineation of the prosthesis. Textural features were considered stable in case of a minimum acceptable excellent agreement indicated by an intra-class correlation coefficient (ICC) level of 0.75 . To obviate multicollinearity among all significant, accurate, and stable considered textural features, the pairwise Pearson correlation coefficient was evaluated. When the correlation of a pair of variables was >0.8, the variable with the lowest AUC was excluded from the set of features chosen for AGI characterization. Data were collected and analysed using IBM SPSS 20.0 software (IBM Corp, Armonk, NY, USA).
Baseline patient characteristics
Age; years, mean (range)
Bacteriology of infected prosthetic graft material in group I
Culture obtained from:
Perigraft fluid N (10)
Graft location and material at initial operation
Type of reconstruction
Visual grading scale
VGS for group I was 3.25 ± 1.06. VGS was 3.50 ± 0.71 for group Ia and 2.83 ± 1.47 for group Ib. The VGS for group II was 1.4. ± 0.94. The VGS was found to differ significantly between groups I and II (P < 0.01), with an AUC of 0.90. However, the VGS appeared not to differ significantly between groups Ia and Ib (P = 0.26), with an AUC of 0.64.
Maximum standardized uptake value
The SUVmax for group I was 7.01 ± 2.31 compared to 4.17 ± 1.86 for group II. The SUVmax for group Ia was 7.72 ± 2.22 compared to 5.83 ± 2.09 for group Ib. SUVmax was found to be predictive in distinguishing group I from II (P = 0.01) with an accuracy of AUC = 0.87. SUVmax was also able to distinguish group Ia from Ib (P = 0.13), but with an unsatisfactory accuracy of AUC = 0.75.
TBR for group I was 4.57 ± 2.14. TBR was 4.86 ± 2.15 for group Ia and 3.82 ± 2.14 for group Ib. The TBR for group II was 2.94 ± 1.54. The TBR was found to differ significantly between groups I and II (P = 0.06) with an AUC of 0.78, but appeared not to differ significantly between groups Ia and Ib (P = 0.35) with an AUC of 0.70.
Regression analysis results, accuracy, and robustness of the selected variables
I vs. II
Ia vs. Ib
Maximal standardized uptake value
Tissue to background ratio
Visual grading scale
First order textural features
GLRLM-based textural features
Short run high grey level emphasis*
GLSZM-based textural features
High grey level zone emphasis†
Small zone low grey level emphasis
Small zone high grey level emphasis
This study investigates the relationship between 18F-FDG uptake heterogeneity in the aortic prosthetic graft, as characterized by textural features, and AGI. This study shows that textural analysis of AGI is feasible and may increase the accuracy to diagnose AGI compared to conventional assessment.
In this study, several textural features were found to be robust for inter-observer variability in delineation of the prosthesis and seem to be suitable for AGI prediction. Short-run-high-grey-level-emphasis, which is highly dependent on the occurrence of short runs (and thus a heterogeneous 18F-FDG uptake) with high grey levels, was the only textural feature to distinguish proven (group Ia) from non-proven (group Ib) infection. The short-run-high-grey-level-emphasis demonstrated higher values for the studied infected prosthetic grafts compared to the uninfected prosthetic grafts. This finding, therefore, supports the hypothesis that a high and heterogeneous 18F-FDG uptake is associated with infected prosthetic grafts. The short-run-high-grey-level-emphasis was most efficient in identifying AGI within the suspected group, whereas for the same task the performances of SUVmax, TBR, and VGS measurements were all limited.
Diagnoses based on the standard parameters could significantly distinguish between patients being suspected (group I) and non-suspected (group II) of having AGI. VGS showed the highest accuracy of all studied parameters, indicating that these groups can sufficiently be distinguished without textural analysis. However, among these standard parameters, only the SUVmax was able to distinguish group Ia from Ib. Spacek et al. visually interpreted 18F-FDG uptake as “intense”, “inhomogeneous”, or “none” and found intense focal 18F-FDG uptake to be a significant predictor for AGI , which confirms that SUVmax results were found significant in the current study. However, as was also supported by our previous study , the accuracy of SUVmax was moderate and insufficient for changing clinical decision-making. Of interest, Keidar et al. visually assessed patterns of uptake for non-infected vascular grafts in patients undergoing a 18F-FDG PET for other reasons than suspected AGI . A diffuse homogeneous uptake was observed in 67 grafts (63%) and heterogeneous uptake was observed in 31 grafts (29%). Nine grafts (8%) demonstrated no 18F-FDG uptake and none of the grafts displayed focal 18F-FDG uptake. Keidar et al. hypothesized that diffuse 18F-FDG PET uptake in non-infected grafts is a result of a local sterile inflammatory process around the prosthesis due to a foreign body-related reaction, and also related to the type of implanted material. Moreover, Berger et al. found the mentioned standard parameters largely to overlap in infected and uninfected central vascular grafts , which confirms our findings that conventional parameters are not sufficient in distinguishing these groups.
The current study demonstrated the association of high and heterogeneous 18F-FDG uptake with AGI; however, due to the relatively small patient cohort it seems not applicable for clinical decision making yet. Nevertheless, this finding is of utmost importance, since it warrants studies with larger, prospective patient cohorts with the construction of multivariable prediction models. Such studies should focus on the textural features, which were found robust for inter-observer variability in the current study. One of the most important limitations in testing the utility of textural analysis in the diagnosis of AGI is the fact that 80% of the patients received antibiotic therapy at the time of scanning, which may have yielded false-negative cultures. However, this bias is the result of the clinical reality, since the omission of antibiotic therapy could adversely affect patients and this issue is related to each study predicting AGI. Moreover, routine microbiologic techniques can sometimes fail to isolate the microorganism from perigraft material [35, 36]. Sonication techniques have been described to identify indolent gram-positive microorganisms by using ultrasound energy to agitate particles from the graft sample for microbiology [35, 36]. However, we did not apply these sonication techniques in this retrospective study because they were not available in our hospital. In addition, the used definition for VOI leads to analysis on the whole prosthetic graft volume, which possibly results in an underestimation of the predictive value of the textural features. However, as we mentioned earlier, we chose for this definition since analysis of the most-diseased segment of the aortic graft would require a semi-objective identification of the 18F-FDG-avid area  and since group II does not contain such an area.
Textural analysis to characterize 18F-FDG uptake heterogeneity is feasible and shows promising results in diagnosing AGI and can encourage further research to facilitate implementation of automated textural analysis algorithms into clinical practice. Further research regarding the construction, refinement, and validation of prediction models in larger prospective cohorts is required before it can be implemented in the clinical decision-making process.
18F-FDG PET, 18F-fluorodeoxyglucose positron emission tomography; AGI, aortic graft infection; AIC, Akaike information criterion; AUC, area under the receiver operating characteristic curve; CRP, C-reactive protein; CT, computed tomography; EANM, European Association of Nuclear Medicine; EVAR, endovascular aortic repair; GLCM, grey level co-occurrence matrix; GLRLM, grey level run-length matrix; GLSZM, grey level size-zone matrix; ICC, intra-class correlation coefficient; PTFE, polytetrafluoroethylene; SUVmax, maximal standardized uptake value; TBR, tissue-to-background ratio; VGS, visual grading scale; and VOI, volume of interest
Conception and design: BRS, RJB, RB, CJZ, and RHJAS
Provision of study materials or patients: BRS, RJB, CJZ, and RHJAS
Collection and assembly of data: BRS, RJB, CJZ, and RHJAS
Data analysis and interpretation: BRS, RJB, RB, AWJMG, MMPJR, CJZ, and RHJAS
Manuscript writing: BRS, RJB, RB, AWJMG, MMPJR, CJZ, and RHJAS
Final approval of manuscript: BRS, RJB, RB, AWJMG, MMPJR, CJZ, and RHJAS
Compliance with ethical standards
No external funding was received.
Conflict of interests
Informed consent was not required (retrospective study).
- 13.Boellaard R, O’Doherty MJ, Weber WA, et al. FDG PET and PET/CT: EANM procedure guidelines for tumour PET imaging: version 1.0. Eur J Nucl Med Mol Imaging. 2010;37:18–200.Google Scholar
- 25.Saleem BR, Pol RA, Slart RH, Reijnen MM, Zeebregts CJ. 18F-Fluorodeoxyglucose positron emission tomography/CT scanning in diagnosing vascular prosthetic graft infection. Biomed Res Int. 2014;471971:1–8.Google Scholar
- 26.Doane DP. Aesthetic frequency classifications. Am Stat. 1976;30:181–3.Google Scholar
- 29.Thibault G, Fertil B, Navarro C, Pereira S, Cau P, Levy N, et al. Textural indexes and gray level size zone matrix. Application to Cell Nuclei Classification. Pattern Recognit Inf Process. 2009:140–145.Google Scholar
- 31.Steyerberg EW. Clinical prediction models. A practical approach to development, validation, and updating. New York: Springer; 2009.Google Scholar
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.