Method for identifying colon polyps in a computed tomography colonography dataset

1. A method of formatting data for detecting colon polyps in a Computed Tomography Colonography (CTC) dataset using a computer processing apparatus for providing automated image analysis, said method comprising the steps of:

i) using the computer processing apparatus to receive the CTC dataset and using the computer processing apparatus to extract colon surface data, from the CTC dataset, within a sub-volume of the extracted colon surface data that is centered on a seed point of a candidate polyp;

ii) using the computer processing apparatus to identify individual sets of points corresponding to a body of the candidate polyp and a base of the candidate polyp within the sub-volume;

iii) using the computer processing apparatus to select the points corresponding to the body;

iv) using the computer processing apparatus to re-format the candidate polyp by projecting the selected points corresponding to the body on to a tangent plane, wherein re-formatting of the candidate polyp includes the following steps carried out by the computer processing apparatus:

(a) projecting the set of points corresponding to the body on to the tangent plane via a normal vector, wherein the normal vector is calculated from a best fitting plane to the set of points corresponding to the base,

(b) isolating a new seed point for the candidate polyp from the projected data, wherein isolation of the new seed point comprises the following steps carried out by the computer processing apparatus:

1) labelling the non-flat surface of the body using region-growing,

wherein this process is seeded by the points which are also part of the body,

2) projecting these labelled points onto the tangent plane, via the normal vector of the best fitting plane to the set of points corresponding to the base,

3) applying least-square ellipse fitting to the data,

4) locating the center of a best-fitting ellipse, and

5) projecting this point, via the previously calculated normal vector, onto a plane passing through a centroid of a 3D surface,

wherein this point becomes a new seed for all further processing, and

(c) locating a nearest surface point to the new seed in the extracted colon surface data, and region-growing the surface ab initio from the nearest surface point; and

v) using the computer processing apparatus to generate a series of cutting planes based on the reformatted candidate polyp.

2. A method according to claim 1 wherein the step of using the computer processing apparatus to generate a series of cutting planes based on the reformatted candidate polyp comprises the following steps carried out by the computer processing apparatus:

i) identifying individual sets of points corresponding to the body and the base in the region-grown surface, and determining the normal vector from the best fitting plane to the set of points corresponding to the base;

ii) using the normal vector to slice a structure of the candidate polyp in a number of different orientations; and

iii) generating a series of planes for each orientation, wherein elliptical structures can be identified as representative of flat polyps.

3. The method of claim 1 wherein candidate polyp seed points are located in the CTC dataset utilising a 3D Hough transform.

4. The method of claim 1 wherein region-growing is done on every point whose normal disparity measure is lower than a maximum allowed convexity,

wherein calculating the normal disparity measure for a point comprises calculating a normal vector for the point and determining a standard deviation of normals in the 3×3×3 surface neighbourhood, wherein the standard deviation is used as a measure of local surface curvature.

5. The method of claim 2 wherein the structure of the candidate polyp is sliced in seven different orientations using the normal vector calculated from the best fitting plane to the set of points corresponding to the base.

6. The method of claim 5 wherein the structure of the candidate polyp is sliced in seven different orientations based on orientation of the normal vector and two sets of three orientations at two different angles from the normal vector.

7. The method of claim 2 comprising an edge detection step.

8. The method of claim 7 wherein the edge detection step is performed on each of the planes extracted.

9. The method of claim 8 wherein edges representing the candidate polyp are detected in the edge detection step using a canny filter.

10. The method of claim 1 , wherein the method further comprises:

using the computer processing apparatus to classify candidate polyps using re-formatted candidate polyps.

11. The method of claim 10 wherein the step of classifying candidate polyps is performed by means of a committee of classifiers running on the computer processing apparatus.

12. The method of claim 6 wherein a normal plane and a tangent plane are extracted for each different orientation.

13. The method of claim 11 wherein the committee of classifiers comprises one classifier for each cutting plane.

14. The method of claim 12 wherein the method further comprises:

fitting a tightest bounding box to the colon surface data to extract a sub-volume of the colon surface data, and

extracting a working volume of the colon surface data from the sub-volume of colon surface data.

15. The method of claim 14 wherein the colon surface data of the extracted working volume is padded to get square images.