Volume-based feature analysis of mucosa for automatic initial polyp detection in virtual colonoscopy

Abstract

In this paper, we present a volume-based mucosa-based polyp candidate determination scheme for automatic polyp detection in computed colonography. Different from most of the existing computer-aided detection (CAD) methods where mucosa layer is a one-layer surface, a thick mucosa of 3-5 voxels wide fully reflecting partial volume effect is intentionally extracted, which excludes the direct applications of the traditional geometrical features. In order to address this dilemma, fast marching-based adaptive gradient/curvature and weighted integral curvature along normal directions (WICND) are developed for volume-based mucosa. In doing so, polyp candidates are optimally determined by computing and clustering these fast marching-based adaptive geometrical features. By testing on 52 patients datasets in which 26 patients were found with polyps of size 4-22 mm, both the locations and number of polyp candidates detected by WICND and previously developed linear integral curvature (LIC) were compared. The results were promising that WICND outperformed LIC mainly in two aspects: (1) the number of detected false positives was reduced from 706 to 132 on average, which significantly released our burden of machine learning in the feature space, and (2) both the sensitivity and accuracy of polyp detection have been slightly improved, especially for those polyps smaller than 5mm.

Computer-aided detection of colonic polyps with level set-based adaptive convolution in volumetric mucosa to advance CT colonography toward a screening modality

“…The Gaussian distributed weight represents the ratio of the local smoothness of x j to the global one along the path L, whose details are referred to by Wang and colleagues. 24 However, following the same arguments as we acquired SL above, we believe the middle layers (eg, layers 2 and 3 in Figure 7) of the VM would be more reliable than the other layers (eg, layers 1 and 4 in Figure 7) for evaluating the geometric features of the colon mucosa. Therefore, we introduced the MEI strategy to highlight the middle layers of the VM through the term W j n Taking Figure 7 as a typical example with totally four layers inside VM and starting from layer 1, the breadth-fi rst dilation strategy marches towards layers 2, 3, and 4.…”

Improved computer‐aided detection of small polyps in CT colonography using interpolation for curvature estimationa)

“…Indeed, curvature is the basis of almost all existing CAD schemes for polyp detection. [4][5][6][7][8][9][10][11][12][13] In general, there are two approaches (geometry/surfacebased and image-based) to computing curvature estimates from volumetric images. The first approach (geometry/surfacebased) is to fit locally a surface patch to the data, with a known parameterization in a local coordinate system, and to compute surface curvatures from the parameterized surface.…”

“…When this kernel approach is applied for polyp detection, such problem was pointed out by many researchers. 12,16,19 That is, spurious estimation of curvatures could be observed in two situations: (1) thin flat folds and small polyps. Here, the "thin" and "small" are relative to the window size of kernel used for curvature estimation.…”

“…Similarly, Wang et al 12 directly eliminated the contribution of voxels of different structures during the convolution. Zhu et al applied level-set based adaptive convolution 22 and Knutsson mapping method 19 to improve curvature estimation for colonic polyps.…”

Computer-aided detection of colonic polyps with level set-based adaptive convolution in volumetric mucosa to advance CT colonography toward a screening modality

“…The mathematically sophisticated method used by Wijk and colleagues 22 is time consuming, and the hard segmented air-tissue interface cannot refl ect the partial volume effect (PVE). Similarly, Wang and colleagues 24 directly eliminated the contribution of voxels of different structures in the convolution by Deriche. 17 However, the single-voxel interface layer (referred to as the starting layer (SL) in the following text) used to build up the distance transform was not properly selected either, and due to PVE, their SL would overlook the polypoid structures and incorrectly synthesize nonexisting polypoid structures.…”