High performance lung nodule detection schemes in CT using local and global information

Abstract

Purpose: A key issue in current computer-aided diagnostic (CAD) schemes for nodule detection in CT is the large number of false positives, because current schemes use only global three-dimensional (3D) information to detect nodules and discard useful local two-dimensional (2D) information. Thus, the authors integrated local and global information to markedly improve the performance levels of CAD schemes.

Methods: Our database was obtained from the standard CT lung nodule database created by the Lung Image Database Consortium (LIDC). It consisted of 85 CT scans with 111 nodules of 3 mm or larger in diameter. The 111 nodules were confirmed by at least two of the four radiologists participated in the LIDC. Twenty-six nodules were missed by two of the four radiologists and were thus very difficult to detect. The authors developed five CAD schemes for nodule detection in CT using global 3D information (3D scheme), local 2D information (2D scheme), and both local and global information (2D + 3D scheme, 2D - 3D scheme, and 3D - 2D scheme). The 3D scheme, which was developed previously, used only global 3D information and discarded local 2D information, as other CAD schemes did. The 2D scheme used a uniform viewpoint reformation technique to decompose a 3D nodule candidate into a set of 2D reformatted images generated from representative viewpoints, and selected and used "effective" 2D reformatted images to remove false positives. The 2D + 3D scheme, 2D - 3D scheme, and 3D - 2D scheme used complementary local and global information in different ways to further improve the performance of lung nodule detection. The authors employed a leave-one-scan-out testing method for evaluation of the performance levels of the five CAD schemes.

Results: At the sensitivities of 85%, 80%, and 75%, the existing 3D scheme reported 17.3, 7.4, and 2.8 false positives per scan, respectively; the 2D scheme improved the detection performance and reduced the numbers of false positives to 7.6, 2.5, and 1.3 per scan; the 2D + 3D scheme further reduced those to 2.7, 1.9, and 0.6 per scan; the 2D - 3D scheme reduced those to 7.6, 2.1, and 0.8 per scan; and the 3D - 2D scheme reduced those to 17.3, 1.6, and 1.0 per scan.

Conclusions: The local 2D information appears to be more useful than the global 3D information for nodule detection, particularly, when it is integrated with 3D information.

FIG. 1.

Histogram of the nodule size.

FIG. 2.

Overall scheme of the computerized detection technique using 2D and 3D information.

FIG. 3.

Uniformly distributed viewpoints on a sphere generated by spiral-scanning technique.

FIG. 4.

Illustration of the coordinate system showing a nodule O and a viewpoint P. The reformatted 2D image generated from the viewpoint passes the center of the nodule and is perpendicular to the line connecting the center of the nodule and the viewpoint.

FIG. 5.

A nodule (left column) and a blood vessel (right column) in the consecutive CT slices [(a) and (b)], the 2D reformatted images [(c) and (d)], and the segmented images [(e) and (f)]. As expected, the nodule appears circular in all the consecutive CT slices and 2D reformatted images. Although the blood vessel appears as nodule-like circular objects in all the consecutive CT slices, it appears clearly as noncircular linear structures in some “effective” 2D reformatted images. These “effective” 2D reformatted images enable us to well distinguish nodules from blood vessels.

FIG. 6.

Relationship between two 2D features for nodules (circle) and false positives (dot).

FIG. 7.

Two rules for removing many false positives. The lines, circles, and dots indicate the rules, nodules, and false positives, respectively.

FIG. 8.

FROC curves of the 2D + 3D scheme obtained with different numbers of viewpoints. The performance levels using 24 and 42 viewpoints are considerably higher than that using 11 viewpoints.

FIG. 9.

FROC curves of the 2D + 3D scheme obtained with different percentage threshold (T). The performance levels using different percentage thresholds are close to one another.

FIG. 10.

FROC curves of 2D + 3D scheme obtained with no rule, one rule, and two rules. The performance levels using no rule, one rule, and two rules are close to one another.

FIG. 11.

Time required to classify all 2D nodule candidates at different levels of sensitivity for the 2D + 3D scheme with no rule, one rule, and two rules (initial sensitivity = 91%). When the sensitivity is set at 80%, the needed time with two rules was about a third of that with no rule.

FIG. 12.

FROC curves of the 2D + 3D scheme obtained with nodules confirmed by two, three, or four radiologists. If the nodules were confirmed by more radiologists, the performance of our CAD schemes for these nodules were higher.

FIG. 13.

FROC curves for the 2D scheme, 3D scheme, 2D + 3D scheme, 2D − 3D scheme, and 3D − 2D scheme. The performance levels of nodule detection using local 2D information and global 3D information (2D + 3D scheme, 2D − 3D scheme, and 3D − 2D scheme) are higher than that using 2D information alone (2D scheme). The performance of nodule detection using 3D information alone (3D scheme) is the lowest.