This problem is related to the path extraction tool developed specifically for virtual endoscopy, Deschamps et al. Colorectal cancer represents the third most frequently diagnosed cancer worldwide. If we consider malignant tumor, the yearly incidence of colorectal cancer probably approaches 160,000 cases. This disease begins in the cells that line the colon, as polyps.
A polyp is a growth that occurs in the colon and other organs. These growths, or fleshy tumors, are shaped like a mushroom or a dome-like button, and occur on the inside lining of the colon.
Colon polyps start out as benign tumors but in time may become malignant. The larger the polyp, the more likely it is to contain cancer cells.
A great deal is known about why and how polyps form. There now is strong medical evidence that there are abnormal genes for colon polyps and cancer that can be passed from parent to child. Diet and foods may also be very important.
Importantly, colon cancer is one of the most curable forms of cancer. When detected early, more than 90 percent of patients can be cured. Early detection of colon polyps and cancer is performed usually with
We propose in the following to adapt the method developed in the thesis and already applied to cerebral aneurysm segmentation, to develop a initial framework of semi-automatic polyp extraction. We further explore possibilities of detection with visualization techniques, using the curvature information of the object surface.
Classical CT scanner are generally very large. Instead of treating entire images, we used small volumes of interest which were selected by specialized physicians because of the presence of a particular pathology.
Before acquisition, the patient goes through a particular preparation during which the colon is emptied as much as possible. During the scan it is distended by inflating room air. The resulting image intensity in the colon lumen is rather uniform and lower than in the rest of the image, with a relatively good contrast. Therefore, the critical step of the segmentation process is the variability of the topology and geometry of the pathological structures.
Since the contrast is really important, as shown in figure 6.9-(a), it is a very easy task to set a seed point inside the colon, and another outside.
the four datasets used for segmentation | ||||
Results of the Fast-Marching competition algorithm | ||||
Results of the Level-Sets algorithm |
Colon polyps appear as convex regions in the lumen surface, in intraluminal 3D views. We tried to enhanced these suspect regions using a color information on the surface.
The specific shape of the colon polyps settle the use of the curvature information, mapped on the surface of the object, using an adequate color-map to highlight the cups. This technique has been already used in the surface of a segmented cortex, by Zengh et al, using a measure defined originally by Koenderink et al.
Segmentation step was achieved using the same parameters for each datasets, and the curvature mapping is done with the same color-map. On several datasets, this mapping highlights other non-pathological regions: folds can be highlighted because of the sign of their principle curvatures. However our approach might be consider as a valuable start for the automatic detection of polyps, and currently viewed as an assisting tool for their visualization. The polyps are emphasized, and discriminated from the whole surface. Therefore, segmentation and visualization is achieved with a simple and fast process, leading to a pre-detection of the polyps which can already be used by any clinician.
In conclusion, the use of this kind of curvature filter outputs information relative to small and spherical polyps. Those polyps can grow and develop malignant tumor with non-smooth shapes where the curvature information is not suitable. Our tool finds its application in the early detection of the small polyps. The high precision of the implicit level-set representation of the surface obtained through the segmentation process, enables to map on the surface informations for small objects like polyps. Our curvature measure is dedicated to this visualization.
Having in mind the settle of a non-supervised method of polyps detection, next step is recognition: Other non-pathological objects that are pre-detected can be discriminated with classification of the shapes.