Overview Segmentation information from the small intestine and colon can be used for finding suspicious regions on CT or MR images. The regional segmentation process is difficult if the segment in question is not filled with gas or radiopaque contrast (CT) or sorbitol solution or other hyperosmotic contrast (MRI). It is possible to instill contrast into the small intestine via nasojejunal tube (enteroclysis), but the process is difficult and uncomfortable for the patient, so enterography with oral contrast is preferred. It is a rule of thumb in medical imaging that luminal distension is optimized if the lumen is temporarily occluded at both ends; the pylorus and duodenal bulb are not completely competent valves, and the ileocecal valve is often incompetent, making it difficult to totally distend the small bowel. Because it is difficult to expand the entire small intestine with gas or liquid contrast, and peristalsis empties short sections, it is more difficult for radiologists to follow
Fig. 3.83 Example of luminal regions of the stomach. Outside views of luminal regions are rendered
the entire course of the long-small intestine, as opposed to the shorter colon, on a CT scan. Typically, the number of “fragments” (opacified segments separated by peristaltic segments) of the small intestine is 20 and that of the colon is five. The large intestine usually runs almost in a simple shape like a U shape. The small intestines run in an unpredictable shape. Fragment connection is much more important in the small intestine extraction from CT images than that of the colon. Oda et al. (2014) tried to solve this issue by enumerating all possibilities of fragment connection . This method first obtains centerlines of the fragmented luminal regions (Fig. 3.84). The i-th centerline is expressed as Ln .n = 1 ? ?-N/. A terminal point and the other terminal point of the centerline Ln are expressed as v2n and v2n+, respectively (Fig. 3.85). For one ordered sequence of L, we compute
where D is a distance term and R is a direction term. D is described as
Here N is the number of the centerlines of the fragmented luminal regions. Here we introduce Cj.j = 0 ---N — 2/ that expresses the connection of the centerlines. dj is
Fig. 3.84 Example of intestinal luminal regions and their centerlines
Fig. 3.85 Concept of fragmented region connection
the maximum distance in the connections. Direction term R is defined as
where j and j are running directions of the centerlines connected by Cj. Fragments connection can be obtained by finding a set of connection minimizing Eq. 3.22. Figure 3.86 shows an example where fragmented regions are connected correctly by the above process.
Colon lumen segmentation is easier, given the facts that the colon is shorter and can be opacified with contrast or distended with air in a retrograde fashion if necessary, with the ileocecal valve at the beginning of the ascending colon
Fig. 3.86 Example of fragmented region connection
and a rectal balloon at the end of the colon acting as plugs to allow distension. Segmentation data on the colonic lumen can be used for eliminating processing regions in automated polyp detection. They can also be used for computing the centerlines of luminal segments, enabling automated fly-through inside the lumen. Such segmentation includes the lumen and the intestinal wall. Because peristalsis changes the luminal shape, it is difficult to employ statistical methods for segmenting intestinal organs. It is important to connect luminal fragments. A typical method uses the distances between the segments. Two segments are connected if the distance between these segments is lower than a certain threshold value.
Large Intestine Threshold value t for segmenting the luminal regions is usually set as
where fw means a CT value of the large intestine wall and fa shows a CT value of air. Typicallyfw andfa are -90 H.U. and -1000 H.U., respectively.
Figure 3.87 shows an example of luminal region extraction of the colon.
1. Intestinal wall segmentation
Intestinal wall extraction from CT images is a difficult problem when we compare intestinal wall segmentation with intestinal lumen segmentation. This is
Fig. 3.87 Example of luminal regions of the colon. Outside views of luminal regions are rendered
because normal intestinal wall in a distended segment is almost invisible on CT. Masutani et al. tried to segment the colonic wall regions from abdominal CT images. This method extracts the colonic wall regions in combination of thresholding and binary image manipulations .