CAD Techniques for Detection of Polyps

To date, most of the CAD schemes developed in academia and in industry comprise of the following four fundamental steps: (1) extraction of the colonic wall from the CTC images, (2) detection of polyp candidates in the extracted colon, (3) characterization of false positives, and (4) discrimination between false positives and polyps. A brief descrip tion of each of these steps is provided below. More technical details on the fundamental CAD scheme can be found in recent review articles (Yoshida and Dachman 2004, 2005).

In the first step of the extraction of the colonic wall, either fully automated (MasutaNi et al. 2001; NAppi et al. 2002a, 2004b; Wyatt et al. 2000) or semi-automated (CheN et al. 2000; IordaNescu et al. 2005; Summers et al. 2000) methods are used. Most of these methods use the thresholding of the CTC data based on the CT values characteristic of the colonic wall and the contrast between the colonic wall and the air in the colonic lumen as a means of extracting the colon.

In the second step, polyp candidates are detected by use of morphologic features that characterize the shape differences among polyps, folds, and the colonic wall. Figure 11.2a shows an example colonoscopy image of a 6-mm polyp in the sig-moid colon. As schematically shown in the middle column, polyps tend to appear as bulbous, caplike structures adhering to the extracted colonic wall, whereas folds appear as elongated, ridge-like structures, and the colonic wall itself appears as a large, nearly flat, cup-like structure. To characterize these morphologic differences, various methods have been developed, including use of a volumetric shape index and curvedness (Yoshida et al. 2002a; Yoshida and NAppi 2001), surface curvature with a rule-based filter (Summers et al. 2000), sphere fitting (Kiss et al. 2002), and overlapping surface normal method (PaiK et al. 2004). Figure 11.2b shows pseudo-coloring of the colonic lumen that visualizes the result of the shape analysis based on the volumetric shape index. The shape index determines to which of the following five topologic classes a voxel belongs: cup, rut, saddle, ridge, or cap. Color coding of the anatomic structures in the colonic lumen based on these classes can thus differentiate among polyps (green), folds (pink), and colonic walls (brown) effectively (NAppi et al. 2005b).

Typically, the polyp candidates thus detected include a large number of false positives, many of which are caused by prominent folds and by feces (Yoshida et al. 2002a, 2002b). Various methods characterizing false positives based on geometric and texture features have been developed for reduction of their number, include volumetric texture analysis (NAppi and Yoshida 2002), CT attenuation (Summers et al. 2001), random orthogonal shape section (GoKturK et al. 2001), and optical flow (Acar et al. 2002).

Fig. 11.2a,b. Schematic illustration of the geometric modeling of the structures in the colonic lumen. (Reprint, with permission, from Yoshida and Dachman 2004)

Fig. 11.2a,b. Schematic illustration of the geometric modeling of the structures in the colonic lumen. (Reprint, with permission, from Yoshida and Dachman 2004)

The final detected polyps are obtained by application of a statistical classifier based on the image features to the differentiation of polyps from false positives. Investigators use parametric classifiers such as quadratic discriminant analysis (Yoshida and Nappi 2001), non-parametric classifiers such as artificial neural networks (Jerebko et al. 2003b; Kiss et al. 2002; Nappi et al. 2004b), a committee of neural networks (Jerebko et al. 2003a), and a support vector machine (Gokturk et al. 2001). In principle, any combination of features and a classifier that provides a high classification performance should be sufficient for the differentiation task.

The CAD output is displayed, in a 3D workstation, as a list of detected polyps (Yoshida et al. 2004b) (Fig. 11.1) or integrated in 2D MPR and 3D endolumi-nal views of the colon by use of, for example, the coloring scheme that delineates the detected polyps and the normal structures in the colonic lumen (Nappi et al. 2005b) as shown in Fig. 11.3. For each pair in this figure, the left image shows an axial CT image containing a polyp (arrow), and the right image shows its 3D endoscopic view by perspective volume rendering. The color coding is based on the above shape index analysis (Fig. 11.2). Figure 11.3a shows a 6-mm sessile polyp in cecum and Fig. 11.3b shows a 5.3-mm polyp in cecum, both of which were missed by a radiologist at first reading. Figure 11.3c shows a 7-mm polyp in the transverse colon, and Fig. 11.3d shows an 11-mm sessile polyp in the hepatic flexure. All of these polyps are clearly segmented from folds and the colonic wall by use of CAD.

Managing Diverticular Disease

Managing Diverticular Disease

Stop The Pain. Manage Your Diverticular Disease And Live A Pain Free Life. No Pain, No Fear, Full Control Normal Life Again. Diverticular Disease can stop you from doing all the things you love. Seeing friends, playing with the kids... even trying to watch your favorite television shows.

Get My Free Ebook

Post a comment