A novel method for measuring bowel motility and velocity with dynamic magnetic resonance imaging in two and three dimensions

Increasingly, dynamic magnetic resonance imaging (MRI) has potential as a noninvasive and accessible tool for diagnosing and monitoring gastrointestinal motility in healthy and diseased bowel. However, current MRI methods of measuring bowel motility have limitations: requiring bowel preparation or long acquisition times; providing mainly surrogate measures of motion; and estimating bowel‐wall movement in just two dimensions. In this proof‐of‐concept study we apply a method that provides a quantitative measure of motion within the bowel, in both two and three dimensions, using existing, vendor‐implemented MRI pulse sequences with minimal bowel preparation. This method uses a minimised cost function to fit linear vectors in the spatial and temporal domains. It is sensitised to the spatial scale of the bowel and aims to address issues relating to the low signal‐to‐noise in high‐temporal resolution dynamic MRI scans, previously compensated for by performing thick‐slice (10‐mm) two‐dimensional (2D) coronal scans. We applied both 2D and three‐dimensional (3D) scanning protocols in two healthy volunteers. For 2D scanning, analysis yielded bi‐modal velocity peaks, with a mean antegrade motion of 5.5 mm/s and an additional peak at ~9 mm/s corresponding to longitudinal peristalsis, as supported by intraoperative data from the literature. Furthermore, 3D scans indicated a mean forward motion of 4.7 mm/s, and degrees of antegrade and retrograde motion were also established. These measures show promise for the noninvasive assessment of bowel motility, and have the potential to be tuned to particular regions of interest and behaviours within the bowel.


| INTRODUCTION
Bowel motility disorders can arise from several conditions and affect patients to varying degrees across a wide demographic. Inflammatory bowel conditions, such as Crohn's disease 1 and ulcerative colitis, 2 are often associated with bowel motility disorders. 3 Crohn's disease affects growth in children, 4 and has serious long-term implications, while ulcerative colitis is associated with a high risk of colon cancer. 5 Globally these two conditions affect more than 11 million people, 6 with the causes being poorly understood. Irritable bowel syndrome (IBS) is another type of functional gastrointestinal disease that affects 10%-15% of people in the developed world. 7 It is associated with bowel dysmotility but, crucially, it demonstrates no structural changes detectable by current routine diagnostic tools. 8 Causes of IBS range from psychological stress 9 to bacterial aetiology 10 and vitamin deficiency. 11 Treatment options for the aforementioned disorders differ per condition and depend on the degree of symptoms. Examples include lifestyle changes, medication, and surgery. 12 However, as the incidence of conditions such as Crohn's disease increases, elective surgery is becoming commonplace to allay long-term effects. 13 To improve treatment of these illnesses and to develop our understanding of underlying conditions (as well as areas like drug delivery), it is vital that we develop quantitative means of discriminating pathological and healthy bowel segments.
Magnetic resonance imaging (MRI) is an ideal noninvasive tool for imaging the gastrointestinal tract due to its exceptional soft-tissue contrast.
In studying both peristalsis and laminar flow in the bowel the noninvasive nature of MRI offers a safer alternative to barium-enemas, 14,15 x-ray investigations, and intraoperative assessment 16 ; and, crucially, the current reference standard for bowel motility assessment, antroduodenal manometry, cannot measure flow. A number of MRI approaches exist, as reviewed by de Jonge et al., 17 including tagging 18,19 and displacement mapping 20 ; however, scan times can be long, with coarse temporal resolution, and some methods provide only surrogate measures. Some manual 21 and semiautomated 22 MRI methods give meaningful bowel motility metrics, but these can be labour-intensive and can require bowel preparation. Several recent reviews 17,23,24 have made a case for the development of quantitative and noninvasive measures of bowel function.
However, MRI measures of bowel motility and transit are challenging, being subject to many of the same limitations as cardiac MRI methods, 25 and there are obstacles to the widespread adoption of these techniques. First, most approaches to date have been applied to two-dimensional (2D) data, leading to erroneous measurements as the bowel moves in and out of the 2D slice; this is a reason to move to three-dimensional (3D) imaging. Further, bowel motion occurs over a range of different spatial and temporal scales, and imaging protocols need to be optimised accordingly. 26 Indeed, protocols must be adapted to focus on specific bowel regions and time scales to obtain clinically useful measurements.
Another consideration is that bowel preparation using oral contrast agents, while useful for distending the bowel and inducing wall motion that can be studied by MRI, may lead to nonphysiological motility patterns, and thus there is demand for methods that can be used both with and without such preparations.
In this proof-of-concept study we introduce a novel technique that is sensitive to flow within the bowel and is complementary to existing MRI protocols. It consists of a dynamic MRI acquisition that requires minimal bowel preparation (a 6-h fast) and can be applied using off-the-shelf 2D or 3D cine imaging pulse sequences. The subsequent image analysis supplies useful metrics for quantitatively assessing flow in bowel loops, and gives measures of laminar flow and peristaltic motion in standard units (mm/s). Here, we demonstrate this method in healthy volunteers and verify the results against invasive intraoperative measures from the literature.

| THEORY
Motility measures are concerned with capturing peristaltic motion in the bowel, where peristalsis is the contraction and relaxation of smooth muscle in the bowel wall, propagating in a wave in an antegrade direction. Such waves propel material, which flows in a laminar fashion due to its high viscosity. Measures of this laminar flow can be used to assess the bulk motion within the bowel. The function of the bowel can be characterised by the amplitude and propagation of peristaltic contractions, and how these translate to antegrade and retrograde flow. Previous work has looked at local measures of signal intensity variation, 27 or area change, within the small bowel, but not at the quantitative tracking of flow within the lumen, 28 with the exception of Hoad et al., who studied flow in the colon. 29 Therefore, new approaches that measure the degree of motion with respect to the vector described by the bowel are of great interest.
As shown in Figure 1, motion can be cylindrically integrated over φ and measured in terms of α, allowing detection of contraction, translocation, and peristalsis in the bowel. However, any rotational information will be lost. Defining motion with respect to the long axis, b, allows complex motion to be interpreted more easily. Furthermore, any signal of motion is summed over φ, increasing its significance in this inherently noisy technique.
The motion assessment method introduced here uses cost-function minimisation, the value of which determines how well motion has been detected, and the parameters used in the function determine the velocity. Spatially adjacent voxels in the image are sequentially fitted in the indirect time domain or 'motion domain'. All possible vectors are fitted, and previous minimisation results in surrounding voxels are overwritten if a better result is obtained. Consider a as a list of points or voxels, p, corresponding to the vector a in Figure 1: where the number of voxels, n, determines the 'lengthscale' of the fit. Each voxel, p, has a spatial position and an associated time series: where ΔS t is the relative signal at a given time point, t, and T is the total acquisition time. A time shift, Δt, is nominally applied to the first adjacent voxel at the next time point, as material is assumed to have moved to this voxel. This Δt is then scaled to subsequent adjacent voxels (Δt 0 ) to give the signal, ΔS tþΔt 0 , as the same material is assumed to pass through each point. The mean signal can then be defined as: The mean-squared difference at each time point, relative to the mean signal, is given as: The cost function, A, is then defined as: The minimised Δt is then used to calculate the scalar component for velocity, v: where the vector component is defined by the vector a.  3 | MATERIALS AND METHODS

| Simulations
To demonstrate our approach and the contributions of noise and the relative signal of the motion to our results, a set of 2D synthetic time series data was generated. The simulation was structured as a three-level intensity image of bright circles within an envelope, where each timeframe advanced the structure by two pixels.
A matrix of simulated images, with varying contrast and noise, was fed through the pipeline described above. Contrast levels between background and simulated bowel ranged from 0.2 to 0.9, and relative noise was introduced with standard deviations ranging from 1% to 50%.

| MRI
Two healthy male participants (aged 31 and 37 years) were recruited. Each participant received a comprehensive description of the study, including possible risks, and gave informed consent according to the local ethics guidelines. Both participants were asked to fast for 6 h prior to the scan, and no oral contrast was used in order to maintain normal bowel reflexes. Scans were performed with participants in head-first supine orientation.

| Bowel imaging in two dimensions
In 2011, two-dimensional bowel scans were acquired in the 31-year-old participant as part of work carried out by Farghal et al. 27 The scans were performed on a 1.5-T Avanto MRI scanner (Siemens Healthcare, Erlangen, Germany) using two flexible six-element anterior body matrix coils for signal reception. The whole abdomen was covered using 10 single-slice scans, acquired using a dynamic 'True FISP' balanced steady-state free

| Image analysis
Bespoke code was written in Python (version 3.5. To highlight motion on the spatial scales of the bowel, each slice or section of the acquired imaging data was Fourier-filtered in the spatial domain for each time point using a dual Gaussian filter with σ inner = 13 and σ outer = 50, masking structures both larger and smaller than the bowel. The filter was broad enough to preserve bowel signal for the small and large intestines and for size changes relating to motility. Also, structured noise or off-resonance artefacts such as Moiré fringes were minimised using this filter. A relative difference image was then calculated for each time point. Figure 2 shows an example of the preprocessing steps. Note that, in this case, the first time point of the 2D bSSFP acquisition had not yet reached equilibrium, thus this time point was dropped and subsequent slices were normalised to the median signal intensity of each image before preprocessing. It is worth noting that off-resonance banding artefacts in bSSFP can act to highlight motion when difference maps are used, as we do here. Variability maps were also generated using the value of the minimised cost function, A(Equation 5), and the RMS of the sequence in time.

| Region-of-interest and vector delineation
These are comparable with other methods currently used to assess bowel motility.    Figure 4A varies for different lengthscale profiles. The signal is persistent for the larger lengthscales but disappears when the lengthscale is reduced too far. Further, it can be seen that changing the lengthscale also alters the observed mean velocity, as the method is sensitised to different sources of motion.

| Reproducibility of 2D scans
When considering the reproducibility of the measure we must first consider how variable the motion may be in a given region of bowel. To look at variability over the 10 repeat scans, the loop of small bowel shown in Figure 4C was used, but with both regions defined by two values for b at  data. Also shown are data from the liver, which is assumed to be stationary throughout imaging. The 2D histogram shows similar proportions of vectors across all angles, which represent the influence of noise. Further, most vectors are found in the range of velocities to which the analysis was sensitised, which was $ 4-6 mm/s. In the pylorus and the bowel loops shown, these noisy vectors fade into the background due to relative scaling of the histograms, highlighting discrete signals of motion. Figure 9 shows the averaged values of the minimised cost function and the RMS of the sequence in time.

| Variability maps
F I G U R E 6 A series of 2D histograms generated for 10 sequential scans, and analysed for the loop of bowel shown in Figure 4C. in a manner that will enhance the field. Further, it is one of only a handful of approaches that have been applied with minimal bowel preparation, with others having been reported by de Jonge et al. and Khalaf et al. 31,32 The quantitative measure of velocity we obtain, along with the degree of antegrade or retrograde motion along the long axis of the looping bowel, permits the definition of expected laminar flow, provided there is sufficient image contrast between materials in the lumen. The information on retrograde motion we show here also has value: for example, when the ileocaecal valve is incontinent, as a result of surgery or an inflammatory condition such as Crohn's disease, retrograde motion can lead to large bowel bacteria colonising the small bowel. With our method we aim to bridge the diagnostic gap between the bowel's capacity to perform peristalsis and the impact this has on material moving through the digestive tract, helping to connect how motion then relates to absorption and related symptoms.
In addition to peristalsis and laminar flow, our method may also be capable of detecting segmentation contractions, a common type of mixing motility that is particularly apparent in the small bowel. The antegrade and retrograde signals of motion that we observe are indirect measures of bowel segmentation contractions and peristalsis, together. In theory, these segmentation contractions could be directly detected by our method, provided they are of a size and duration to which our scan is attuned. A specialised 'faster' sequence would detect smaller, faster segmentation contractions, for example. However, segmentation contractions can occur over both long and short time scales, both in close proximity to one another and far apart. Thus, the 3D acquisition we present here, with properly tuned time scales and lengthscales, would be required to adequately explore this phenomenon.
One area where our method has particular potential is for studying paediatric motility disorders. 20 The scan volumes for 3D acquisitions in children will likely be much smaller, permitting increased signal-to-noise, improved temporal resolution, and sensitivity to a wider range of velocities. However, limited breath-holding ability in younger patients might pose challenges for the detection of slower and longer time-scale periodic motions, and free-breathing acquisitions may be required to capture these. The current best available solution for bowel motility measurement is a displacement mapping approach, which is offered as a commercially available tool (Motilent, Ford, UK; http://www.motilent.co.uk/); however, this method is restricted to a 2D coronal slice. The main output measure is the degree of variation in the displacement, which reflects the F I G U R E 8 Extraction of motion vectors from a 3D scan. Two slices are shown on the left with the vectors of motion inset, and approximate positions are shown with arrows for guidance. The corresponding 2D histograms for each region are shown on the right. Blue: peristalsis of the pylorus; orange: large bowel down to the sigmoid; green/red/purple: loops of small bowel with degrees of content from high to low. Also shown is a 2D histogram obtained from the liver (grey box), which can be used to illustrate the effect of noise, as the liver is assumed to be stationary. Note how the histogram appears homogenously dark, across all angles, for the range of velocities to which this analysis was sensitised ($ 4-6 mm/s). For motile structures such as the pylorus, and other bowel loops shown here, this noise fades to highlight clear and discrete signals of motion contraction and peristalsis of the bowel; however, the flow within the bowel-anterograde or retrograde-is not considered in this approach.
While this displacement mapping method provides measures for stratifying conditions like Crohn's disease, it has not been extended to 3D scanning in the 9 years since initial publication. Pure measures of variability show levels of activity, 27 but much is hidden in the third dimension.
Although 3D scanning has been performed on the bowel 33 to achieve diagnostic quality images, the method introduced here does not require high-quality images to characterise motion. Moving to three dimensions reduces the signal, which manifests both as lower contrast-to-noise for the fitting (addressed to some extent by the method described here), and difficulty in delineating loops of bowel for applying a vector of measurement (i.e. the scans' utility as a visual tool). The advantage conversely is an order of magnitude increase in the number of voxels available when considering any vector of motion. The primary challenge, therefore, is in delineating the bowel region in which the motion is to be detected. In this study, only the 3D scans and their corresponding cost-function maps were used to determine bowel-containing motion ( Figure 9A); however, in future studies a static high-contrast scan should be included to assist localisation.
In terms of published data, there are relatively few quantitative measures of peristaltic or laminar flow in the literature. A recent invasive study, 16 involving the intraoperative wrapping of the proximal jejunum with flexible printed circuit boards following a laparotomy procedure, measured wave-propagation patterns and a mean (SD) longitudinal velocity of 9.0 (0.7) mm/s. Circumferential velocity of the wave was seen to be faster at 10.8 (0.8) mm/s. For comparison, a histogram of forward motion (α < 20 ) for the region in Figure 4A can be seen in Figure 10, with the intraoperative result included. Our 3D scan data showed similar velocities (mean = 4.7 mm/s) in the small bowel, but no strong signal above 6 mm/s, suggesting laminar flow. However, the loop of bowel examined in Figure 4 is split bi-modally between a signal at 9 mm/s and a spread of velocities at around 5.5 mm/s. This suggests that both laminar flow and peristaltic flow are captured in our data. Previous MRI tagging work by Pritchard et al. obtained semiquantitative velocity data in the ascending colon, 34 showing simultaneous antegrade and retrograde flow, as well as complex flow in the hepatic flexure, and evidence of retrograde central 'jets' travelling in excess of 48 mm/s.
Regarding the varying spatial and temporal scales of bowel motion, the maximum detectable velocity is determined by the smallest value for Δt, so faster movement relies on the reliability of the fit when Δt is less than one time point, therefore encouraging finer timing resolution. It also follows that the minimum detectable velocity is governed by, variously: the number of voxels, the inverse of the number of time points, and the degree of overlap in time between the first and last voxels. The temporal scale (i.e. rate of change of motion) therefore imposes a trade-off with respect to the detectable velocities and the spatial scale that can be used. Scans should be maximised for signal-to-noise, but not at the expense of these considerations, to avoid the loss of important motion signals. In this study, we examined a relatively short period of bowel motion, due to breath-hold constraints of the healthy volunteers ($20 s), and our analysis largely focused on laminar flow. Clinical applications for this method are many-including chronic intestinal pseudo-obstruction, 35 Parkinson's disease, 36 and IBS 37 -and an extended free-breathing implementation of our approach could both confirm the initial measurements and provide information on longer time scales. 38,39 Figures 6 and 7 show how this motion evolves over repeated scans. However, due to the elastic nature of how peristalsis affects the bowel content, retrograde motion is also expected at a higher velocity, 40 and this was not seen. To capture these higher velocities, including the aforementioned retrograde central jets, requires a higher temporal resolution than was used in our study. Similarly, observing the generally slower moving large bowel requires either extending the lengthscale and the number of time points in the fit to the full extent of the 20-s breath-hold, or including sequentially acquired scans or free-breathing data in the analysis. We observed a small amount of motion in the large bowel; however, given the small signal in the bowel content in this case, any estimate of slow laminar flow is probably unreliable. Nevertheless, how this technique is applied is dictated by the area of clinical applicability, and thus we targeted, and successfully characterised, a broad middle ground of velocity profiles.
In this study we used product MRI pulse sequences made available by vendors. Given that our processing pipeline involved Fourier-filtering, a bespoke 3D sequence would be beneficial, as view-sharing in the TRICKS sequence is optimised for vascular scale detail in the centre of k-space that is ultimately filtered out in our method. The central region of the radially acquired k-space is repeatedly sampled (as per the specified tradeoff between spatial and temporal resolution) alternately with annuli of k-space surrounding the centre. TRICKS divides k-space into four concentric regions (A-D, inner to outer) and samples in the order ABACAD, with the frame rate defined by the time between visits to the central region, which is filtered out in our approach. Altering the sampling scheme to regions BABC would avoid unnecessary detail loss and improve temporal resolution, which could be traded for greater signal-to-noise if desired.
In addition to developing a more specialised imaging sequence for our technique to optimise k-space filling, future work could focus on automated region of interest determination. Parallel computing could be used to fit every possible vector in the bowel, in all directions, and those that represent 'true' motion could be isolated using thresholding. From an applications perspective, the repeatability of our method should also be assessed in healthy volunteers and its results validated against the approach used by Motilent, which is routinely used in the clinical setting.

| CONCLUSION
We have shown in this work that it is possible to obtain linear measures of motion within the bowel in both two and three dimensions, and this can be achieved using product scanning sequences and minimal bowel preparation. This method provides a promising complementary tool for the noninvasive assessment of bowel motility, and has the potential to be tuned to particular bowel regions-such as the small bowel, large bowel, and ileocaecal sphincter-as well as to specific applications, including paediatric cases.