Pelvic MRI and CT images are interchangeable for measuring peripouch fat

A total of 27 pouch patients with inflammatory bowel diseases, who underwent pelvic MRI-DIXON and CT scan within one year, were included. Peripouch fat areas were measured at the middle height level of pouch (AreaM) and the highest level of pouch (AreaH). Our results demonstrated that measurements of perianal fat thickness, AreaM and AreaH based on MRI image were accurate and reproducible (correlation efficiency(r): intraobserver: 0.984–0.991; interobserver: 0.969–0.971; all P < 0.001). Bland-Altman analysis showed that more than 92.593% (25/27) of dots fell within the limits of agreement. We also identified strong agreements between CT and MRI image in measuring perianal fat thickness(r = 0.823, P < 0.001), AreaM (r = 0.773, P < 0.001) and AreaH (r = 0.862, P < 0.001). Interchangeable calculating formula to normalize measurements between CT and MRI images were created: Thickness_CT = 0.610 × Thickness_MRI + 0.853; AreaM_CT = 0.865 × AreaM_MRI + 1.392; AreaH_CT = 0.508 × AreaH_MRI + 15.001. In conclusion, pelvic MRI image is a feasible and reproducible method for quantifying peripouch fat. Pelvic MRI and CT images are interchangeable in retrospective measurements of peripouch fat, which will foster future investigation of the role of mesentery fat in colorectal diseases.


Correlation between measurements of pelvic MRI-DIXON-F and CT images. The measurements
of MRI-DIXON-F images were significantly higher than CT images, in terms of perianal fat thickness, AreaM and AreaH (Table 1). However, scatter plots demonstrated that there were highly significant correlations between measurements obtained from CT and MRI-DIXON-F images, in terms of peripouch fat thickness (r = 0.823, P < 0.001, Fig. 4a), AreaM (r = 0.773, P < 0.001, Fig. 4b) and AreaH (r = 0.862, P < 0.001, Fig. 4c). To achieve normalization between two different imaging modalities, regression analyses were also performed to get regression formulas. The formula facilitated the mutual conversion of measurements obtained from pelvic MRI-DIXON-F  images and pelvic CT images. Therefore, the equivocal CT/MRI measurements could be calculated using the following formula: Thickness_CT = 0.610 × Thickness_MRI + 0.853; AreaM_CT = 0.865 × AreaM_MRI + 1.392; AreaH_CT = 0.508 × AreaH_MRI + 15.001, when the MRI/CT measurements were known.

Discussion
We analyzed a total of 27 patients who had both pelvic MRI-DIXON-F and CT images within one year. The median time interval and interquartile range (IQR) between the pelvic CT scan and pelvic MRI-DIXON-F scan was 20 (−90 to 70) days. For pelvic MRI-DIXON-F image, scatter plots and Bland-Altman graph showed that both intraobserver and interobserver discrepancies were extremely small in measuring perianal fat thickness, AreaM and AreaH, and more than 92.593% (25/27) of dots fell within the limit of agreements. In addition, our study demonstrated that measurements based on MRI-DIXON-F images were significantly higher than CT images, in terms of perianal fat thickness, AreaM and AreaH. Nonetheless, there were significantly correlations between measurements obtained from MRI-DIXON-F and CT images. Finally, mathematical formulas for mutual transition between measurements of MRI-DIXON-F and CT images were constructed. Measurements of abdominal visceral fat using MRI and CT scans were first compared by Seidell in a prospective study including 7 healthy male volunteers 25 . Strong correlations had been reported between MRI and CT images when measuring abdominal visceral fat 25 . Thereafter, several studies had verified the reliability of prospective and retrospective measurements of abdominal visceral fat using CT and MRI [18][19][20][21] . The reported results in abdominal visceral fat were consistent with what we found in peripouch fat. In our study, measurements based  on MRI-DIXON were significantly higher than CT images, in terms of perianal fat thickness, AreaM and AreaH. These results were contrary to Waduud's study, which found that MRI measurements of total abdominal area and abdominal waist circumference were lower than the counterpart CT measurements 21 . But in Waduud's study, CT scan was performed in the supine position while MRI scan was performed in the prone position, and the differences were attributed to different patient positions. But, in our study, both CT and MRI scans were performed in the supine position, so patient position was not a major cause. Potretzke 28 reported that using CT image to recognize fat could lead to overestimation of abdominal fat due to the incorrect inclusion of colonic content. In our opinions, there are two possible reasons for the higher measurements in MRI-DIXON. Firstly, the MRI-DIXON images are more efficient in depicting peripouch fat than CT images. MRI-DIXON is specific in displaying fat tissue, which is superior to MRI-T1 in depicting water-fat borders and image details 29,30 . In CT scan, only Hounsfield units are used to differentiate fat from adjacent tissue. In the case of IPAA, pouch body is located at the narrow pelvic cavity, and some fat tissue can't be shown on CT image due to its partial volume effect. Secondly, MRI images might have more motion artefact. Compared with CT scan, MRI scan takes a longer acquisition time, usually greater than a breath-hold, which may lead to more motion artefact. In this study, the fat tissue was differentiated from adjacent non-fat tissue by setting automatic threshold when analyzing MRI-DIXON images. Using this technique, only manual circling of peripouch fat (regions of interest) was required. This approach has the advantage of minimizing interobserver and intraobserver differences, and it is much more subjective and time-saving compared with individualized threshold 21 . We only used a single slice technique rather than a volume-based technique for assessment of peripouch fat. Borkan's study confirmed that there was no additional advantage of volumetric analysis compared with using multiple levels images in measuring abdominal fat 31 . We initially planned to analyze the peripouch fat area at three levels: the lowest level of pouch, the middle level of the pouch body, and the highest level of pouch. Since the pouch is very close to the ani elevator at the lowest level of pouch, the area of peripouch fat is too small to be analyzed. Therefore, we elected to measure the perianal fat thickness.
In our Pouch Center, we have routinely used pelvic MRI to assess fistula, abscess and anastomotic leak of the pouch, with its higher resolution than CT. However, CT imaging has the advantage of being quicker to acquire, which avoids potential interference of bowel movement on region of interest measures 18,24,32 . CT can also be used in patients when MRI is contraindicated, for instance, patients with intrauterine device or other implanted metallic objects in pelvic cavity. In addition, CT is more cost effective than MRI, leading to its more frequent use in clinical practice. However, CT scan is associated with a significant dose of radiation, which restricts its application in certain patients, such as pregnant women. MRI does not involve radiation exposure, which allows for frequent repeat imaging and it could be applied in at-risk patients, such as pregnant women 18 . Furthermore, MRI is superior to CT in measuring fat, as fat has a typical short longitudinal relaxation time which makes it easy to be differentiated from other adjacent tissues in MRI image 21 . Since both CT and MRI scans have their own advantage, disadvantage, indication and contraindication, respectively, so it is very important to explore whether they are interchangeable. The MRI-DIXON is a T1 weighted gradient echo sequence, which displays fat and water separately 29,30,33 . MRI-DIXON has four images (in phase, opposed phase, water phase and fat phase) which could be gained in a single scan, and the fat phase (shorted as MRI-DIXON-F) is specialized in exhibiting fat tissue 29 . MRI-DIXON-F image is superior to standard MRI-T1 image in depicting water-fat borders and image details 30 , and it is reported to be associated with a high interobserver reliability in measuring fat area around biceps 33 .
This is the first study to compare measurements of peripouch fat using retrospective CT and MRI-DIXON images. The CT and MRI-DIXON scans used in this study were all performed without pre-defined research criteria, so it exactly represented current clinical practice. We demonstrated that the intraobserver and interobserver variability for measurements based on MRI images were extremely small in all three parameters. It validated that MRI image was a robust and reproducible method for measuring peripouch fat. Strong positive linear correlations had been identified and mathematical formula had been developed to normalize measurements between imaging modalities.
The measurements were easily gotten from retrospective CT and MRI images, so it provided an opportunity for researchers and clinicians to determine peripouch fat change without any additional cost. Our clinical practice and primary research results of another study indicated that, the peripouch fat area had something to do with chronic pouchitis and pouch failure. So the dynamic monitoring of peripouch fat might be helpful in the prediction of chronic pouchitis and pouch failure. Since both CT imaging and MRI imaging have their own advantages, so they are mutually supplemental in the routine follow up and complication evaluation in IBD patients. The conclusion of this study would be practically used to facilitate depicting the dynamic alteration of peripouch fat, and researching on the underlying mechanism between peripouch fat and pouch complication.
There were some limitations in this study. Firstly, the small sample size was a major limitation. The 1,863 pouch patients in our database were not followed up routinely by pelvic CT or MRI scan. For most patients, CT or MRI scan was enforced to evaluate pouch complications, such as anastomatic stricture, perianal abscess and pouch fistula. Furthermore, MRI-DIXON was not routine series for pelvic MRI scan, and only approximate 50% of pelvic MRI scans had MRI-DIXON series in our clinic. Therefore, selection bias was inevitable. Secondly, timing gap between CT and MRI scan was also a potential confounding factor. The CT and MRI images included in our analyses were taken at different time points. During these periods, weight loss or gain may happen, which might interfere with the reliability of comparison analysis. Waduud's study verified that the timing of imaging did not affect the linear relationship of measurements on CT and MRI 21 . Since one year was only a relatively short period as compared with the long survival period of pouch, and our study still showed that there were strong positive correlations between measurements of CT and MRI. This suggested CT and MRI images performed up to a year apart could be reliably analyzed and compared as demonstrated in our study. Thirdly, some MRI images were incomplete and didn't include whole area of pelvic, so it was impossible for us to calculate subcutaneous fat area. Instead, perianal fat thickness was applied. Finally, most pelvic CT or MRI images didn't reach the height of umbilical level or the third lumbar vertebrae level, which was often used in measuring the abdominal fat distribution. Therefore, the total abdominal fat distribution and total abdominal visceral fat were not analyzed in this study. Despite the above mentioned limitations, strong relationships were still demonstrated between MRI-DIXON-F and CT measurements in this study.
In conclusion, our study demonstrates that measurements of peripouch fat based on pelvic MRI-DIXON-F image is feasible and highly reproducible for pouch patients with underlying IBD, since both intraobserver and interobserver differences are extremely small. Pelvic MRI images and CT images are interchangeable in SCIeNtIFIC RePoRTS | 7: 12443 | DOI:10.1038/s41598-017-12732-6 retrospective measurements of peripouch fat, which will provide new revenue to the prospective as well retrospective assessment of fat and mesentery in other gastrointestinal disorders.

Methods and Materials
Patients. This study was performed using pelvic MRI-DIXON-F and CT images from our Center for Ileal Pouch Disorders. Patients were identified from a prospectively maintained Institutional Review Board approved Pouch Database between 2002 and 2016 in Cleveland Clinic. Demographics, comorbidities, pouch complications and pouch outcomes were all prospectively maintained in the pouch database.
Inclusion and Exclusion Criteria. The inclusion criteria were patients who had (1) underlying UC, Crohn disease (CD) and indeterminate colitis; (2) one or more pelvic MRI-DIXON scans in our electronic medical images systems between pouch construction date and pouch failure date; (3) one or more pelvic CT scans in our electronic medical systems between pouch construction inception date and pouch failure date; and (4) the interval between CT scan and MRI-DIXON scan was less than one year.
Exclusion criteria were patients with (1) pelvic MRI images but no DIXON series; (2) radiation or chemotherapy between CT and MRI; and (3) Fig. 5c). CT images were reset using the CT Hu (−30 to −190) in workstation 18,34 . Both the pelvic MRI-DIXON-F images (Fig. 5d-f) and CT images (Fig. 5g-i) with rulers were downloaded from the digital imaging system at the above mentioned 3 levels, and stored in a specific file using the patients' medical record number. Each MRI image was analyzed by two individual raters to identify interobserver difference. Both raters, at the time of measuring, were blinded to clinical data and the measurements of the other rater. Images were analyzed for two times by the same rater to identify intraobserver differences.
The thickness of perianal subcutaneous fat was calculated at Level-1 (Fig. 5d,g) after setting scales in the guide of the attached ruler. The peripouch fat areas were calculated at Level-2 (middle peripouch fat area, AreaM) and Level-3 (highest peripouch fat area, AreaH). Fat thickness and fat areas were calculated using the ImageJ software (http://imagej.nih.gov/ij/). For all patients, perianal fat thickness, AreaM and AreaH were measured in both pelvic MRI-DXION-F and CT images. Figure 6 showed the process of measuring AreaM using both MRI-DXION-F and CT images (Fig. 6a,b). Firstly, the peripouch fat (regions of interest) was circled manually (Fig. 6c,d). Briefly, only the pelvic visceral fat, and the pelvic parietal fat inside the pelvic muscle and the abdominal wall muscle were included. The subcutaneous fat and the fat in the bone marrow of the sacrum, femoral head and lateral pelvic wall, were excluded. And these muscles act as inherent boundary of the peripouch fat area. And then the outside redundant tissue was cleared (Fig. 6e,f). Images were transformed into 8-bit type, and then the thresholds were adjusted. For MRI-DXION-F images, the threshold was automatically set at the midpoint between the two peaks of the identified signal intensities in the histograms (Fig. 6g), and fat area was highlighted automatically in almost all images 21 . For CT images, the thresholds were set at a fixed range (10-200) (Fig. 6h). The red area demonstrated peripouch fat (Fig. 6i,j). And then the red areas were calculated with the "Analyze Particles" tool in the ImageJ software.
Validation of measurements based on MRI images. In order to determine the reliability of the measurement method, intraobserver and interobserver variability was assessed Intraobserver variability was evaluated by comparing repeated measurements made by a single rater. Interobserver variability was evaluated by comparing the measurements made by two independent raters. A formative assessment of intraobserver and interobserver variability was performed by scatter plot with linear regression and Bland-Altman graph.
Statistical analysis. A paired t test or Wilcoxon rank sum test was used for continuous variables as appropriate. Fisher exact or Chi-square test was used for categorical variables as appropriate. Interobserver and intraobserver differences were evaluated by scatter plot with linear regression and Bland-Altman plots. For the Bland-Altman plots, the limits of agreement were calculated from the difference between the raters' measurements for thickness, AreaM and AreaH, respectively. P < 0.05 (two-side) was considered as statistically significant. Bland-Altman graphs were drawn using the MedCalc 9.2.0. All other analyses were carried out using the SPSS 17.0 (Chicago, IL).
Compliance with ethical standards. The study was approved by the Institutional Review Board of Cleveland Clinic. For this type of study formal consent is not required.