Magnetic resonance imaging for gastric motility and function

Gastric morphology and volume measurement: Image acquisition and analysis

The stomach has a complex three-dimensional (3D) anatomy that must be reconstructed from two-dimensional (2D) cross-sectional MR images acquired in standard anatomical planes (i.e. sagittal, coronal, transverse) . The first step in achieving this goal is to filter the images digitally to achieve better contrast for the region of interest (ROI). To facilitate this step, paramagnetic contrast agent (e.g. Gadolinium DOTA) can be added to test meals. Then the ROI is segregated from the other adjacent anatomical structures captured in the images, a step known as image segmentation. The conventional way of doing this is to manually identify sufficient number of points on the edge of the ROI and join them with straight lines to generate the contours of the ROIs, i.e., segmentation of the images. Once all the contours belonging to the ROI in all the 2D image slices are identified, then these can be covered with a surface using 3D reconstruction computer algorithms . Manual analysis of this data is tedious and time consuming, as forty or more image slices are required to capture the ROI in its entirety at each time point during gastric filling and emptying (typically every 5–15 minutes during a 2 hour gastric emptying study). Moreover, this method of image analysis requires expert personnel for accurate image segmentation. These limitations have restricted the adoption of this technique to a few specialist centers.

The simplest method used to extract objective measurements of 3D volume from MR images is based on thresholding, i.e., segregation of regions within images depending on whether they fall below a particular image intensity threshold and are adjacent to neighboring pixels of similar intensity, to form contiguous areas . In the analysis of MR images from the stomach, the search domain is not restricted to neighboring pixels in one particular image slice through the stomach. Rather it can be expanded to include neighboring pixels in adjacent slices to generate the volume of “objects” within a ROI that share the same image intensity. This approach is useful for (relatively) rapid estimate of gastric content volume with gastric air and gastric meal volumes assessed separately . Image quality and the contrast play a crucial role in controlling the efficiency of this algorithm because the semi-automatic measurements are confounded by the presence of adjacent objects with similar image intensity. This can be an issue with the stomach, especially if a normal, mixed solid and liquid test meal with heterogeneous content is ingested. Further the resolution of the 3D images acquired by this technique is not high. Nevertheless, thresholding is very useful for measurement of complex structures with relatively homogeneous content like the small intestine .

Recently, a semi-automatic method has been developed, to generate high-resolution 3D reconstructed geometry in a rapid and efficient manner . The user generates virtual MR images (usually 3–4) along arbitrary planes by manually selecting a few points on the ROI (e.g. stomach wall) in a number of 2D images within the stack such that the planes pass through the stomach in multiple images ( Fig. 14.1A ). The complexity of the ROI determines the number of virtual slices to be segmented. The segmentation of the virtual images generates contours, which intersect the original images and these “seed points” at the edge of the ROI in the original image give cues for the computer to further segment the original images ( Fig. 14.1B ). The order of joining the points is determined using an algorithm, referred to as “magnetic linking” based on the local image intensity gradient in the image and joined automatically using edge detection algorithms (e.g. livewire) to generate high-resolution 3D images the original 2D MR slices . Further this technique can be adapted to assess changes to 3D geometry over time by using a representative image from within a series of images as reference to process adjacent image stacks automatically. This generates four-dimensional (4D) reconstructions that document dynamic gastric function in terms of volume change and alterations in gastric morphology during gastric filling and emptying . The advantage of this method over the manual approach is that it substantially reduces user input and the time required for image processing. The number of slices to be segmented is reduced drastically (approximately 90% less) and requires an estimated 100 mouse clicks, as opposed to that 10,000 mouse clicks in the conventional procedure . The 3D reconstruction of gastric morphology is expedited using this semi-automated method; however, this method is dependent on access to high-end computing technology and still requires more time to perform than the above mentioned thresholding technique .

Insight into the functional anatomy of the stomach can be obtained by dividing the 3D geometry of the organ in the computer environment into different segments and finding the variation in the volume of these segments over time. Usually the stomach is divided into proximal and distal region, however, more division may provide more information about how the stomach accommodates food and delivers nutrients to the small bowel ( Fig. 14.3 ) . Detailed results from non-invasive imaging confirm the functional division of the stomach. After initial filling of the distal and mid stomach (antrum and corpus), the meal is accommodated in the proximal stomach (fundus). Similarly, once the meal is completed, volume in the distal and mid-stomach remains relatively stable until the proximal stomach is almost empty .