Increasing computer-aided detection specificity by projection features for CT colonography

Abstract

Purpose: A large number of false positives (FPs) generated by computer-aided detection (CAD) schemes is likely to distract radiologists' attention and decrease their interpretation efficiency. This study aims to develop projection-based features which characterize true and false positives to increase the specificity while maintaining high sensitivity in detecting colonic polyps.

Methods: In this study, two-dimensional projection images are obtained from each initial polyp candidate or volume of interest, and features are extracted from both the gray and color projection images to differentiate FPs from true positives. These projection features were tested to exclude different types of FPs, such as haustral folds, rectal tubes, and residue stool using a database of 325 patient studies (from two different institutions), which includes 556 scans at supine and/or prone positions with 347 polyps and masses sized from 5 to 60 mm. For comparison, several well-established features were used to generate a baseline reference. The experimental evaluation was conducted for large polyps (> or = 10 mm) and medium-sized polyps (5-9 mm) separately.

Results: For large polyps, the additional usage of the projection features reduces the FP rate from 5.31 to 1.92 per scan at the comparable by-polyp sensitivity level of 93.1%. For medium-sized polyps, the FP rate is reduced from 8.89 to 5.23 at the sensitivity level of 80.6%. The percentages of FP reduction are 63.9% and 41.2% for the large and medium-sized polyps, respectively, without sacrificing detection sensitivity.

Conclusions: The results have demonstrated that the new projection features can effectively reduce the FPs and increase the detection specificity without sacrificing the sensitivity. CAD of colonic polyps is supposed to help radiologists to improve their performance in interpreting computed tomographic colonography images.

Figure 1

The histograms (bottom figure (f)) of CT density distribution of five VOIs, which are indicated by the closed curves in the above five pictures. From (a) to (e), the objects are a polyp, two stools (i.e., the stool_1 and stool_2 in (f)), an ileocecal valve, and a rectal tube.

Figure 2

A 10 mm pedunculated tubular adenoma in the ascending colon of a 62 yr old female, showing a typical translucent color signature. (a) The 3D endoluminal view of the polyp. (b) Translucency display applied to the 3D image in (a). (c) The 3D endoluminal view of the same polyp, but from another direction. (d) Translucency display applied to the 3D image in (c).

Figure 3

Overview of our CADpolyp scheme.

Figure 4

Definition of the LRF for a polyp candidate. (a) The arrow along the protruding direction represents the normal direction of the polyp candidate, starting from a small ball centered at the centroid of the VOI of the polyp candidate. The rectangle through the centroid of the VOI denotes the plane perpendicular to the normal direction. (b) Extraction of the optimized 2D frame (the dotted arrows) of a point set (the points mimicking the results of projecting the VOI to the plane in (a)). The solid arrows represent an arbitrarily selected initial frame. (c) The resulted LRF of the polyp candidate in (a). (d) The LRF of an IPC on a fold. (e) The LRF of another 9 mm tubular adenoma. Unlike the one in (a), this polyp has a large angle from the colon wall.

Figure 5

The sub-volume (the box) of a 10 mm pedunculated tubular adenoma in Fig. 2a.

Figure 6

Illustration of the projection procedure. (a) The projection rays in three directions (arrows) shoot through the 3D polypoid object, and are collected on the three projection planes to form 2D projection images as shown in Fig. 10. (b) Points are evenly sampled on each projection ray.

Figure 7

Plot of the transfer function which was used for the projection images, where the solid, dash dotted, and dotted curves indicate the mapping of the red, green, and blue channels. The white channel is shown with the dashed curve, and the solid curve with diamonds plots the opacity values according to attenuation (HU).

Figure 8

The axial projection images (last two columns) of a 10 mm polyp based on the original (upper row) and cleansed (bottom row) CTC images. The arrows in the left column indicate the IPC findings.

Figure 9

Illustration of the projected gray and color images of several IPCs (including FP and TP findings), where each row represents one IPC. Column 1 shows parts of the original axial, coronal, or sagittal CTC slices with the arrow indicating the IPCs. The sub-volumes and LRFs are shown in column 2. Columns 3 to 5 show the corresponding gray images (the axial, sagittal, and coronal images) of the IPCs. The last column shows the axial color image generated by Eq. 3. True polyps in rows 1 to 3 are 10 (same as in Fig. 2), 9, and 6 mm, respectively. FPs in rows 4 to 6 are three tagged stool, while rows from 7 to 9 are a tube, a thickened fold, and a round ileocecal valve.

Figure 10

Illustration of the projection images of the 3D polypoid object in Fig. 5, where the arrows indicate the characterizing patches. (a) The axial image. (b) The sagittal image. (c) The coronal image.

Figure 11

Extraction of the patches in axial, sagittal, and coronal images of the polyp in row 2 of Fig. 7. Pictures (a), (c), and (e) show the projections of the seed voxel sets Seed(C), i.e., the patches imposed on the axial, sagittal, and coronal images, respectively. Picture (b) shows the highlighted patch (center area) in the axial image. Pictures (d) and (f) show the gray patch (upper and right area) and bright patch (bottom and left area) in the sagittal and coronal images.

Figure 12

A graphical illustration of the disk-likeness of a highlighted patch (the center area), where the circle is centered at the centroid of the patch.

Figure 13

The three color projection images of the two IPCs in rows 1 (TP) and 4 (stool FP) of Fig. 9, where the axial images are repeated here for comparison purpose.

Figure 14

Color projection images (first row) of four IPCs, where each column represents one IPC. The second row shows the IPCs (indicated by the arrows) from zoomed CTC axial∕sagittal slices. Columns 1 and 2 are two stool. One is a tagged adherent stool on a fold, another is partly submerged in the fluid and is less tagged. Column 3 is a thickened fold without any tagged material nearby and column 4 is a tube in another scan.

Figure 15

The size distribution of the 347 lesions in the CTC database.

Figure 16

The fROC curves of the two experiments on two categories of polyps.

Figure 17

Some examples of FN and FP detections. Similar to Fig. 9, one row represents one object. Rows 1 and 2 are two FNs, which are a 30 mm flat mass and a 5 mm polyp on the fold. The last three rows are three FPs, where rows 3 and 4 are two nontagged hard stool balls and the last row is a normal polypoid bump on the fold.