Cine MRI assessment of motility in the unprepared small bowel in the fasting and fed state: Beyond the breath‐hold

The symptoms of functional bowel disorders are common in postprandial but investigations are generally undertaken in the fasted state using invasive procedures. MRI provides a noninvasive tool to study the gastrointestinal tract in an unperturbed, fed state. The aim of this study was to develop a technique to assess small bowel motility from cine MRI data in the unprepared bowel in fasting and fed states.


| INTRODUC TI ON
Conventional manometry of small bowel motility has provided valuable insights into motor function pathophysiology of the gastrointestinal tract. 1,2 The technique has limitations, with naso-duodenal or oro-duodenal intubation being a difficult, uncomfortable, and invasive procedure for patients. Moreover, manometry techniques are not generally used in the lower sections of the small bowel due to difficulties with access and the invasiveness of the technique. 3,4 Furthermore, misinterpretation of the manometry recordings can occur if nonocclusive contractions occur and large spacing between ports mean that motor patterns can be mis-defined. 5 The tube may also interfere with normal feeding making it particularly difficult to study physiological changes from the fasting state and the effect of nutrient intake.
Over the last 10 years, MRI has proven to be a useful tool to probe the unprepared physiology of the gastrointestinal tract. [6][7][8][9] It is particularly suitable for longitudinal or repeated studies, and its versatility allows for multiple physiological parameters to be monitored in a single scanning session. Magnetic resonance enterography (MRE) is used to evaluate the small bowel after the ingestion of an oral contrast agent. It involves distending the bowel artificially to produce detailed images of the bowel wall 10

and induces
bowel wall motion to move the large amount of oral contrast agent through the GI tract, which can then be studied using cine MRI.
Motility measurements following oral contrast preparation using a cine MRI acquisition have made significant advances in recent years, 8,[11][12][13][14][15][16][17] but to date quantification of wall motion has either involved looking for contractions across the lumen 13,14,16 or using registration methods, 15,17 which work well in the deliberately distended bowel where the walls are clearly visualized. However, bowel distension with a hypo-osmotic solution may not be truly physiological and so cannot study true fasting motility patterns and may not represent the full range of motility patterns in the postprandial state. The ability to study motility in the postprandial state, or the transition between the 2 states, has many potential advantages in furthering our understanding of physiology and the origin of symptoms which many patients experience after feeding. Moreover, it is particularly important for the pharmaceutical sciences because the rate and extent of drug dissolution and absorption from solid oral dosage forms is highly dependent upon gastrointestinal motility. 18,19 Furthermore, the use of bowel distension limits the use of MRE in pediatric and elderly populations.
The unprepared small bowel can be imaged using the same high spatial and temporal resolution cine acquisitions as for the prepared bowel. 11,12,14 However, the postprocessing techniques used to parameterize the motility may need to be refined for images, which do not delineate the small bowel wall clearly and show different patterns of motility.
The aim of this study was to develop an analysis technique to assess motility from cine MRI data acquired in the fasted and fed, unprepared small bowel. Inter-and intra-observer variability, and the sensitivity to changes in motility caused by feeding were investigated.

| MATERIAL S AND ME THODS
This study was approved by the local Ethics Committee of the University of Nottingham (H19062014). This study is registered on www.clinicaltrials.gov with identifier NCT02717117. All subjects gave informed written consent. The study design, subjects, and data sets used have been reported previously. 20

| Unprepared bowel data acquisition
Fifteen healthy volunteers (age 29 ± 10 years, BMI 24 ± 5 kg/m 2 ) were recruited from the local campus population. Subjects with any disease or taking medication (eg loperamide, codeine, metoclopramide, hyoscine butylbromide, mebeverine, ondansetron) that affects gastric emptying or small bowel transit were excluded. Standard MRI exclusion criteria were applied.
This study was open label. Subjects were scanned using a were scanned using a range of sequences. At each timepoint, scans were acquired to assess small bowel motility 8 using a single slice bTFE cine MRI acquisition (with reconstructed in-plane resolution 1.49 × 1.7 mm 2 , slice thickness 10 mm, echo time (TE) = 1.52 ms, repetition time (TR) = 3.0 ms, flip angle 80º, SENSE 2.0), of 1 minute free-breathing, temporal resolution of 1 s, this was repeated at 6 contiguous parallel coronal planes through the small bowel as previously described. 20 The total scan time for motility was 6 minutes.
The subjects were instructed to take shallow gentle breaths for the duration of the motility acquisition.

| Motility assessment
Free breathing data were processed using GIQuant TM (Motilent, Ford, UK). The algorithm corrects respiratory motion 22 before applying the nonlinear optic flow registration as described previously 15 to correct local deformation caused by bowel wall motion and model intensity changes caused by luminal flow. The data output from the image registration were further analyzed using a customized graphical user interface written in MATLAB ® (MathWorks, Natick, MA).
On MRI, the unprepared small bowel has a very different appearance to the prepared bowel required for MRE ( Figure 1). In the prepared bowel, there is clear definition of the bowel walls and obvious peristaltic motion is visible through the time series across most of the small bowel. In the unprepared bowel, the bowel wall is not always visible and bolus movement of the chyme between segments is common postprandially. Therefore, a different approach for quantifying the motility of the unprepared small bowel was developed, based on the registration parameter C 15 which represents the change in signal intensity between timepoints, within a defined region of interest (ROI) placed over the small bowel loops. This parameter is modeled simultaneously with the deformation during the registration process and is intended to capture any signal intensity changes not occurring from in-plane motion (ie through plane motion and flow). 15 To allow sensitivity to both oscillatory events such as mixing of contents during peristalsis and forward propulsion of boluses of chyme, the power spectrum analysis of the image registration parameter C was developed similar to that proposed by van der Paardt et al 23 and Sprengers et al. 24 Initially, to remove zero frequency data, the mean of C through time for each pixel was calculated and subtracted from each pixel value ( cally seen postprandially. 25 When regions of interest were defined, average data for the ROI was calculated from the pixel by pixel measurements within the ROI of the AUC power spectrum maps.

| 2.2.2.Observer variability
To determine the variability in the results due to observer definition of the region of the small bowel loops, the following analyses were carried out.

| Changing motility in response to feeding using 2 motility metrics
We examined the strength of correlation between 2 motility analysis techniques using the total power (AUC power spectrum ) and the standard deviation of the Jacobian (SD JAC , a previous published metric for

| Statistical analyses
All statistical analysis was carried out using Graph Pad Prism 7.0 (La Jolla, CA). All data were tested for normality using the D'Agostino and Pearson's normality test. Interobserver variability was investigated using a Bland-Altman plot to determine the 95% confidence limits of agreement. Correlation between observers was measured using the Intra-class correlation coefficient (ICC) using a 2-way random effects model, with a single rater and absolute agreement. 26 Intra-observer variability was also investigated with Intra-class correlation coefficients using a 2-way mixed effects model and single rater and absolute agreement. The 95% confidence limits of agreement were also calculated. Pearson's correlation coefficient was used to measure the strength of correlation between the measurement of motility using AUC power spectrum and SD JAC for the fasting and immediately postprandial data sets.

| Interobserver variability of small bowel motility
The variation in both AUC power spectrum and SD JAC across different timepoints, averaged over all healthy volunteers at each timepoint, and covering all regions of small bowel from the 6 slices, measured by each observer is shown in Figure 4A (error bars shown are SEM).
This graph shows low measured motility at baseline in the fasting state followed by a significant increase postprandially which then persists for the majority of the imaging period. The degree of correlation between 2 observers was assessed using the ICC to be 0.979 and P < 0.0001, n = 195 ( Figure 4B). Interobserver variability was assessed using the Bland Altman plot 27 ( Figure 4C) which showed a mean difference of 8.5 au between small bowel motility measurements, with a 95% confidence interval of −28.9 to 45.9 au as indicated by the upper and lower dotted lines ( Figure 4C).

| Intra-observer variability of small bowel motility
The correlation between the 2 analyses for AUC power spectrum performed by each observer was also assessed using the ICC and Bland-Altman limits of agreement (Table 1), showing good F I G U R E 3 Example of motility maps generated by the software for a single volunteer across the 6 slices acquired, visualizing the areas of high motility. A and B illustrate the fasting and the fed state motility maps. C represents the different motility maps generated by the area under the power spectrum (AUC power spectrum ) and SD JAC motility parameters. S: slice number. Regions of small bowel have been highlighted on the images agreement between the analyses (ICC > 0.9 for all data). Figure 5 plots AUC power spectrum measurements against time for the 6 subjects with normal and high BMI, showing good agreement between analyses at most timepoints for both body compositions.

| Changing motility in response to feeding
The subjects showed an increase between their fasting and initial postprandial motility measurement for both analysis methods (AUC power spectrum 122.4% ± 98.7%, SD JAC 31.8% ± 20.7%), although there was considerable spread within the data across the subjects ( Figure 6). The correlation between the techniques was significant (r = 0.9, P < 0.0001, n = 30).

| D ISCUSS I ON
This study has described and evaluated an optimized technique for the analysis of gut motility from MRI images, specifically addressing the differences in appearance and motility of the small bowel wall in unprepared bowel MR images, compared to images The data from the normal and high BMI subjects would indicate that there is a potential for over estimating motility in the higher BMI subjects. These subjects all presented with high motility indices postprandially; however, only 1 of the 3 showed high baseline data.
These larger motility indices may have been measured as regions of high signal intensity in the fat as it moves into and out of the imaging plane during respiratory motion. As this may not be fully corrected by the registration algorithm these movements will be interpreted as bowel motility. Further studies of higher BMI subjects is needed to understand the factors contributing to the larger motility metric measured and whether poor registration is a factor.
Other factors, which could also influence the signal intensity, are field inhomogeneities and metallic artifacts. To some extent, overall changes in image intensity across the image due to these factors are removed from the AUC power spectrum analysis using the registration parameter C which models the signal changes and not the absolute values. An empty bowel has a different intensity to a filled bowel; however, movement of the contents either between loops or from 1 section to another show similar changes in intensity levels. Other meal contents should have similar motility patterns to the meal in this study, but would need investigating to determine whether the sensitivity is the same as the soup meal, particularly for a more solid meal, which may have a much lower signal intensity in the small bowel.
There were limitations to our study. Due to the time-consuming nature of drawing all the individual ROIs on the motility data the intra-observer repeated measurements were confined to just 6 subjects, not the full 15 available (used for the interobserver data).
Drawing of ROIs for a single timepoint took around 5-10 minutes depending on the anatomy including the loading of each data set into the software. However, the intra-observer data were chosen from subjects who had very different small bowel anatomical appearances due to their differing BMIs, providing the observers with contrasting data for drawing the regions. Smoothing of the data before calculating the power spectrum reduces the effects of isolated poor mis-registration of the data. However, it will not eliminate the effects completely and these datasets will slightly overestimate the small bowel motility present.
In conclusion, this study describes an optimized analysis tech-

CO N FLI C T O F I NTE R E S T
The remaining authors have no competing interests. AM is the CEO of Motilent Limited, a medical imaging analysis company.