{"paper_id":"d68bb555-512b-42b1-9402-cd05ac2114d5","body_text":"In the human body, the main function of white adipose tissue is to contribute to\nenergy homeostasis by absorbing and storing lipids, as well as by preventing ectopic\nlipid deposition. White adipose tissue deposits are found mainly in the subcutaneous\ncompartments of the upper and lower body, as well as in the visceral\ncompartment ( 1 ) . In\nrecent decades, evidence has been mounting that the quantity of visceral adipose\ntissue (VAT) is linked to a number of metabolic dysfunctions, such as insulin\nresistance, hyperinsulinemia, dyslipidemia, and hypertension ( 2 ) . In addition, calculating the\nvariation in the quantity of VAT over time can be a useful way of evaluating\noutcomes in patients who have undergone bariatric surgery ( 3 ) , have dietary\nrestrictions ( 4 ) ,\nparticipate in weight loss programs, or follow specific physical exercise\nregimens ( 5 ) .\nVarious methods have been proposed to calculate the amount of fat tissue  in\nvivo . In clinical practice, some anthropometric indices have been\nproposed for quick, reliable evaluation ( 6 ) , such as waist circumference, hip circumference,\nwaist-to-hip ratio, skinfold thickness, and body mass index (BMI), although none of\nthose are able to differentiate the distribution among the compartments or to\ndistinguish between fat content and muscle mass. Although ultrasound has proven to\nbe an accurate means of evaluating the thickness of subcutaneous\nfat ( 7 ) , its\nperformance continues to be suboptimal for the quantification of\nVAT ( 8 ) . To address\nthese issues, some authors have used other tools, such as dual-energy X-ray\nabsorptiometry ( 9 ) \nand body impedance analysis ( 10 ) , which provide data on lean and fat tissue. However,\nquantitative evaluation of body fat distribution is still difficult to perform.\nComputed tomography (CT) and magnetic resonance imaging (MRI) have both been used as\ntools to investigate the distribution of subcutaneous adipose tissue (SCAT) and\nVAT ( 11 - 13 ) . Although each method shows\nadvantages and disadvantages for that purpose, they both can accurately quantify VAT\nand SCAT ( 14 ) , thus\nquantifying total adipose tissue. CT is considered the most well-established imaging\nmethod for abdominal fat quantification, because adipose tissue has always the same\n(low) density. On MRI scans, the signal intensity of fat is high in T1- and\nT2-weighted sequences, although the numerical value varies depending on several\nfactors ( 15 ) . In the\nuse of CT and MRI, one option is to evaluate the amount of fat contained in a single\nimage (slice) acquired at the level of the umbilicus, which has been reported to\ncorrelate well with the total VAT ( 16 ) . The use of that strategy results in considerably less\nradiation exposure during CT and in a markedly shorter duration of MRI\nexaminations ( 16 ) .\nHowever, some limitations of single-slice analysis have also been reported, mainly\nthe fact that VAT can undergo great variations due to bowel movement or variable\nfilling of the intestine ( 17 ) .\nSeveral types of software have been used in the analysis of images obtained from CT\nand MRI scans. Some such software is developed in-house, and the results are\ntherefore not reproducible, because the software is not publicly\navailable ( 18 ) .\nOther studies have employed specific plugins that can be used with freeware (e.g.,\nImageJ; NIH, Bethesda, MD, USA), although such plugins are very difficult to use in\nclinical practice ( 19 ) .\nOsiriX (Pixmeo, Geneva, Switzerland) is image processing software, dedicated to\nmedical imaging, that is widely used in many radiological applications. The basic\nversion of OsiriX is available for free online. The region growing algorithm of the\nsoftware can be used in order to calculate the area of different compartments of the\nbody and has previously been used in abdominal imaging ( 20 ) .\nThe objective of this study was to test the feasibility of using OsiriX to calculate\nthe amount of VAT in patients who have undergone CT and MRI of the abdomen. We also\ncalculated the intraobserver and interobserver reproducibility.\n\nThis was a retrospective study designed to quantify VAT in patients who underwent\nabdominal CT and abdominal MRI at our institution between 2010 and 2015. The\nstudy was approved by the local institutional review board, and the requirement\nfor written informed consent was waived. The image archive and communication\nsystem of our hospital were screened to identify patients who had undergone\nabdominal CT and abdominal MRI, for any reason, with no more than three months\nbetween the two examinations. The exact interval between the CT and MRI\nexaminations was noted. Examinations of the upper abdomen, lower abdomen, or\nentire abdomen, with or without contrast, were included in the evaluation. For\nboth imaging methods, a slice acquired at the level of the umbilicus was\nconsidered. We included CT examinations with a non-contrast acquisition and MRI\nexaminations with at least one non-contrast, non-fat-saturated T1- or\nT2-weighted sequence.\nMRI examinations were performed in one of two 1.5 T MRI scanners (Symphony or\nAera; Siemens Healthineers, Erlangen, Germany) equipped with phased-array\nabdominal coils. Depending on the clinical problem to be investigated, different\nacquisition sequences were used. However, all of the cases included the\nacquisition of at least one T1-weighted sequence (breath-hold acquisition; echo\ntime = 4.76 ms; repetition time = 280 ms; number of excitations = 1; matrix, 256\n× 256; and slice thickness = 4 mm) or one T2-weighted sequence (breath\nhold acquisition; echo time = 199 ms; repetition time = 4000 ms; number of\nexcitations = 1; matrix, 256 × 256; and slice thickness = 4 mm). When it\nwas available, we selected the slice acquired at the level of the umbilicus in a\nnon-contrast T1-weighted sequence, a non-contrast T2-weighted sequence, or\nboth.\nCT examinations were performed in multidetector scanners, either a 16-slice\nscanner (Somatom Emotion; Siemens Healthineers) or a 64-slice scanner (Somatom\nDefinition; Siemens Healthineers). The technical parameters of CT acquisition\nwere adjusted according to the clinical problem under investigation and patient\nbody size. The slice acquired at the level of the umbilicus was selected in the\nnon-contrast acquisition. The slice thickness was 5 mm.\nFor CT and MRI, the images were obtained from the archive at our hospital. Those\nimages were uploaded to a separate workstation on which the OsiriX software was\ninstalled.\nOn CT scans, abdominal fat has a hypodense appearance, whereas it has a high\nsignal intensity on T1- and T2-weighted MRI scans. The individual CT and MRI\nscans were anonymized and analyzed in random order by two readers, working\nindependently-a radiology resident and a radiologist, both with experience in\nabdominal imaging (more than two years and more than ten years, respectively).\nPrior to that analysis, both readers had a training session in which a series of\nfive CT scans and five MRI scans, not included in the study, were evaluated in\nconsensus in order to optimize the segmentation technique.\nOn each image, a region of interest (ROI) was manually drawn over the abdominal\nwall to delineate the interface between the abdominal wall and the abdominal\nfat. No extreme precision is needed in this phase, because the difference in\ndensity/intensity between the abdominal wall and the abdominal fat is high on CT\nand MRI. The region growing (segmentation) algorithm was selected, thus allowing\nthe segmentation ROIs to be drawn with a semi-automated method. The cursor is\nplaced on a portion of the abdominal fat, and the software automatically creates\na ROI that includes all pixels with gray levels similar to those selected. The\nthreshold (range of gray levels to be included in the evaluation) can be\nmodified by the operator, who uses a slider to improve the\ncalibration ( 5 , 21 ) .\nAt the end of the procedure, the software provides the size of the area included\nin the region growing algorithm that was considered for statistical analysis.\nThe procedure is depicted in  Figures 1  and\n 2 .\nFigure 1 A: Segmentation performed on a CT slice acquired at the level of the\numbilicus. The first step was to draw an ROI passing through the\nabdominal wall, separating the SCAT from the VAT. B: Once the ROI\nwas defined, a point was selected within the VAT area (green cross,\nwhite arrow) and an interval of pixels to be taken into account\n(black arrowhead) was chosen by the reader, in order to include all\nof the VAT within the ROI in the segmentation process. The software\nthen calculated the segmented area (green area) and assigned it a\nvalue (white arrowhead). BL, bowel loop(s).\nA: Segmentation performed on a CT slice acquired at the level of the\numbilicus. The first step was to draw an ROI passing through the\nabdominal wall, separating the SCAT from the VAT. B: Once the ROI\nwas defined, a point was selected within the VAT area (green cross,\nwhite arrow) and an interval of pixels to be taken into account\n(black arrowhead) was chosen by the reader, in order to include all\nof the VAT within the ROI in the segmentation process. The software\nthen calculated the segmented area (green area) and assigned it a\nvalue (white arrowhead). BL, bowel loop(s).\nFigure 2 A: Segmentation performed on an MRI slice acquired at the level of\nthe umbilicus. In this case, a true fast imaging T2-weighted\nsequence was selected. An ROI was drawn to separate the VAT from the\nSCAT. B: The area within the ROI (white arrow) was calculated by\nchoosing an interval of pixels to be taken into account (black\narrowhead). A point within the VAT (green cross) was then selected\nin order to segment the image. To obtain the VAT area (white\narrowhead), the reader has to choose an interval of pixels in order\nto cover the entire area of adipose tissue surrounding the bowel\nloops (green area). BL, bowel loop(s).\nA: Segmentation performed on an MRI slice acquired at the level of\nthe umbilicus. In this case, a true fast imaging T2-weighted\nsequence was selected. An ROI was drawn to separate the VAT from the\nSCAT. B: The area within the ROI (white arrow) was calculated by\nchoosing an interval of pixels to be taken into account (black\narrowhead). A point within the VAT (green cross) was then selected\nin order to segment the image. To obtain the VAT area (white\narrowhead), the reader has to choose an interval of pixels in order\nto cover the entire area of adipose tissue surrounding the bowel\nloops (green area). BL, bowel loop(s).\nThe threshold and fat area data are expressed as mean ± standard\ndeviation. For the thresholds, coefficients of variation (CVs) were also\ncalculated.\nTo evaluate intraobserver reproducibility, the more experienced operator repeated\nthe evaluation, using the same method reported above, after two months.\nIntraobserver and interobserver reproducibility were assessed with the\nBland-Altman method. Values of  p  < 0.05 were considered\nstatistically significant.\nIn the literature, CT is considered a reliable method to measure VAT. Therefore,\nCT was used as the reference in our study. The accuracy of MRI was estimated as\nthe inverse consistency error between CT-determined VAT quantity and that\nmeasured on T1- and T2-weighted MRI scans.\n\nDuring the study period, 4137 patients underwent abdominal CT and 1977 patients\nunderwent abdominal MRI. Among those, there were 31 (14 males and 17 females) who\nunderwent both types of examination. The mean age of the 31 patients was 57 ±\n15 years (range, 34-92 years), and the mean interval between the two examinations\nwas 28 ± 12 days. The CT and MRI examinations were performed for a variety of\nreasons: hepatic lesion (n = 4); pancreatic lesion (n = 5); tumor of the\ngenitourinary tract (n = 7); liver metastases (n = 5); Crohn's disease (n = 5);\ndiverticulitis (n = 1); endometriosis (n = 1); aortic aneurysm (n = 1);\ngastrointestinal stromal tumor (n = 1); and small bowel lymphoma (n = 1). CT scans\nwere available for all 31 patients; T1-weighted MRI scans were available for 26\npatients; T2-weighted MRI scans were available for 23 patients; and T1- and\nT2-weighted MRI scans were both available for 20 patients.\nAs previously described, the threshold (range of gray levels to be included in\nthe evaluation) was adjusted manually. The mean threshold values used for CT\nscans, T1-weighted MRI scans, and T2-weighted MRI scans in the first evaluation\nmade by the more experienced reader were 145 ± 40 (CV = 27.6%), 475\n± 220 (CV = 46.3%), and 367 ± 159 (CV = 43.3%), respectively.\nThe mean area of abdominal fat on CT scans, T1-weighted MRI scans, and\nT2-weighted MRI scans, as calculated by the more experienced reader, was 145\n± 63 cm 2 , 130 ± 71 cm 2 , and 130 ± 68\ncm 2 , respectively. The mean area of abdominal fat on CT scans,\nT1-weighted MRI scans, and T2-weighted MRI scans, as calculated by the less\nexperienced reader, was 143 ± 68 cm 2 , 124 ± 67\ncm 2 , and 130 ± 63 cm 2 , respectively.\nThe intraobserver reproducibility was 90% for CT (bias = 0.14 cm 2 ;\n p  = 0.912), 92% for T1-weighted MRI (bias = −3.4\ncm 2 ;  p  = 0.035), and 90% for T2-weighted MRI\n(bias = −0.30 cm 2 ;  p  = 0.887). The interobserver\nreproducibility was 82% for CT (bias = 1.52 cm 2 ;  p  =\n0.488), 86% for T1-weighted MRI (bias = −4.36 cm 2 ;  p \n= 0.006), and 88% for T2-weighted MRI (bias = −0.52 cm 2 ;\n p  = 0.735). The reproducibility between T1- and T2-weighted\nMRI was 87% (bias = −0.11 cm 2 ;  p  = 0.957).\nIn comparison with that of CT, the accuracy of T1- and T2-weighted MRI was 89%\nand 92%, respectively.\n\nThe main finding of the present study was that OsiriX can be used in order to\nquantify VAT on CT scans, T1-weighted MRI scans, and T2-weighted MRI scans. Overall,\nthe accuracy of MRI, in comparison with that of CT, was high, as were intraobserver\nand interobserver reproducibility, although the quantification of VAT seems to be\nless reproducible on T1-weighted images.\nCT and MRI have both been used in order to quantify VAT ( 22 , 23 ) , although CT has certainly been used more frequently.\nMultidetector CT has the advantage of performing quick scans with very high\nresolution. However, the high dose of ionizing radiation administered to patients is\na major limitation of CT. Conversely, MRI has the great advantage of not using\nionizing radiation as well as allowing for precise tissue characterization. However,\ncertain contraindications (e.g., non-MRI-compatible implants and claustrophobia), as\nwell as the high cost and long examination times, can limit the use of MRI in\nclinical practice ( 5 ) .\nRegarding the evaluation of VAT, one major advantage of CT over MRI is that fat\nalways shows very low attenuation, with little variability among individuals. That\nallows a relatively narrow threshold to be used when applying a region growing\nalgorithm, as confirmed by our data, given that we found a CV of approximately 25%.\nThat is also why we used CT as the reference to calculate the accuracy of MRI, which\nwas found to be high (approximately 90% for T1- and T2-weighted MRI). One\nexplanation for that finding is that CT and MRI were performed at different time\npoints. Therefore, bowel movement and differences in rectal filling may have\naffected the quantity of VAT in the selected slice. However, the signal intensity of\nfat is high on T1- and T2-weighted MRI scans, although that intensity is highly\nvariable, not only among the different types of sequences employed but also among\nindividual patients. That is consistent with our findings, given that the CV\nexceeded 40% for T1- and T2-weighted MRI scans, which implies that a fully automated\nsystem for VAT segmentation and quantification using MRI may be difficult to\nconstruct.\nPrevious studies have quantified VAT on the basis of images of the abdomen as a\nwhole ( 23 )  or a\nsingle slice acquired at the level of the umbilicus ( 17 ) . Although analysis of the entire abdomen\ncertainly has the advantage of greater accuracy, it is extremely time consuming and\nhardly applicable in clinical practice. Various authors have demonstrated that VAT\nquantification using a single slice acquired at the level of any one of several\nanatomic landmarks correlates strongly with total VAT. In a study comparing\ndual-energy X-ray absorptiometry evaluation of whole-body fat and CT evaluation of\nSCAT at the level of the interspace of the fourth and fifth lumbar vertebrae\n(L4-L5), Smith et al. ( 24 )  found that the two approaches correlated strongly,\nespecially among men. Abate et al. ( 16 )  found that, although the most reliable single-slice\nevaluation was achieved with a slice acquired at the L2-L3 level, the addition of a\nslice acquired at the L1-L2 level and another acquired at the L3-L4 level can\nincrease the predictability from 85% to 90%. Even if the use of three slices is\npossible, it takes a considerable amount of time to perform the segmentation and the\napproach should therefore be used only in cases in which greater precision is\nneeded. A slice acquired at the level of the umbilicus has been used because it is\ncommonly included in abdominal examinations performed for any reason. In one study\nusing that approach, Schwenzer et al. ( 17 )  found that VAT measured at the level of the umbilicus\ncorrelated strongly with total VAT, especially among women.\nA wide variety of software has been used for VAT quantification, all such software\nrequiring continuous adjustments by the operator. However, most studies on the topic\nhave provided very few technical details, which limits the reproducibility of the\nresults. Addeman et al. ( 15 )  compared a new automated software known as AdipoQuant\nwith the free software ImageJ, the latter having already been used for this\npurpose ( 19 ) . The\nauthors found that the two programs provided almost identical VAT values, with\nexcellent agreement. However, the processing time per slice was only 2 s for\nAdipoQuant, compared with 8 min for ImageJ. OsiriX has previously been used for\nadipose tissue quantification. In a study involving 62 obese patients, O'Leary et\nal. ( 25 )  used OsiriX\nto quantify VAT, as a means of determining the risk of acute pancreatitis. Those\nauthors found that the quantity of VAT correlated positively with the risk of\npancreatitis, although they did not report exactly how segmentation was performed.\nKinsella et al. ( 26 ) \nalso used OsiriX to evaluate changes in fat distribution in patients who underwent\nrenal transplantation, identifying a significant correlation between VAT and BMI,\nalthough they also provided no details about the segmentation technique. Lee et\nal. ( 27 )  assessed\nthe evaluation of a single CT slice acquired at the level of the umbilicus, as\nanalyzed with the Rapidia software (3DMED, Seoul, Korea), in comparison with the use\nof bioelectrical impedance analysis, in terms of the quantification of VAT. The\nauthors found that the VAT area was smaller when evaluated by bioelectrical\nimpedance analysis than when evaluated by single-slice CT, with a tendency to\nincrease in parallel with increases in BMI. Yu et al. ( 28 )  also used Rapidia to evaluate\nthe correlation between VAT and liver fibrosis among patients with nonalcoholic\nfatty liver disease, finding that the VAT area was significantly larger in the\npatients with fibrosis than in those without.\nIn addition to the commercial software and freeware available, in-house systems of\nVAT quantification have been developed and described by various authors. The main\nlimitations of such studies are that they provide few technical details on the\nsoftware build and that the software is not publicly available, thus precluding any\ntesting of the reproducibility of the results. Maurovich-Horvat et\nal. ( 29 )  performed a\nsemi-automated evaluation of CT-based fat quantification in obese population with\nin-house software, finding excellent intraobserver and interobserver\nreproducibility. Yoshizumi et al. ( 30 )  evaluated VAT in a CT slice acquired at the level of the\numbilicus using an in-house algorithm: the attenuation range of CT values for fat\ntissue was calculated, and a related histogram was constructed, considering the mean\nattenuation plus or minus two standard deviations. Those authors also found that\nintraobserver and interobserver reproducibility were high. In our study,\nintraobserver reproducibility was ≥ 90%, whereas interobserver\nreproducibility was lower, although still relatively high (82-88%). We found that\nstatistical significance was achieved only for the T1-weighted MRI scans. This\nsomewhat unexpected finding might be explained by the fact that the hyperintense\nfluid within the bowel-which has a signal intensity similar to fat-could somehow\nhave affected the segmentation.\nOsiriX offers several advantages over other software in the evaluation of adipose\ntissue. First, because it is freeware, there is a greater likelihood that data will\nbe comparable across studies. Second, because it allows rapid data analysis, it can\nbe applied to large populations as well as to several examinations of the same\npatient in order to analyze changes over time. In addition, we have demonstrated\nthat the VAT segmentation performed with OsiriX has high intraobserver and\ninterobserver reproducibility, which underscores the applicability of this\nmethod.\nThe clinical relevance of the present study mainly resides in the fact that we have\nshown that it is possible to use MRI as a reliable means of quantifying VAT. The\nmain advantage of that approach is the absence of ionizing radiation, which implies\nthat evaluations can be repeated as needed over time. In addition, MRI is\nparticularly useful in specific cohorts of patients (e.g., those with hepatic\nlesions, pancreatic lesions, or Crohn's disease). Therefore, concurrent\nquantification of VAT with no need for a separate examination may represent a\nfurther advantage, given that the presence of a high quantity of VAT has been\nimplicated in predisposition to several diseases, as well as in a poor response to\nseveral treatments ( 2 , 3 ) .\nOur study has several limitations. First, it was a prospective evaluation of\nretrospective data, the CT and MRI examinations having been performed at different\ntime points. Although it seems reasonable to assume that the quantity of VAT would\nnot have changed significantly over a period of three months, such changes could\nhave occurred, which would have affected our evaluation and might explain, at least\nin part, the differences observed. In addition, we included patients with a wide\nrange of diseases, evaluated with different MRI protocols-some including only the\nupper abdomen, some including only the lower abdomen, and some including both.\nHowever, that limitation is mitigated by the fact that we always evaluated the same\nslice acquired at the level of the umbilicus, in T1- or T2- weighted sequences, in\neach patient. Furthermore, the sample size was relatively small. Nevertheless, it\nwas possible to obtain high levels of accuracy and reproducibility.\nIn conclusion, OsiriX can be used in order to quantify VAT on CT and MRI scans\n(T1-weighted or T2-weighted). We found that MRI showed high accuracy, in comparison\nwith that of CT, as well as high intraobserver and interobserver reproducibility.\nHowever, accuracy, intraobserver reproducibility, and interobserver reproducibility\nwere higher for T2-weighted MRI scans, which might therefore be more suitable for\nVAT quantification.","source_license":"CC-BY-4.0","license_restricted":false}