Semi-local active contour for a faint region detection in knee CT images

1. A method of detecting a structure in a three-dimensional (3D) computed tomography (“CT”) knee image in a computer aided diagnosis device, the method comprising:

(a) creating an intensity prior mask based on a two-dimensional (2D) slice of the CT image by the computer aided diagnosis device;

(b) receiving or estimating a seed point in the 2D slice, the seed point being in a Hoffa's pad region by the computer aided diagnosis device;

(c) creating a geometric prior mask based on the seed point by the computer aided diagnosis device;

(d) combining the intensity prior mask and the geometric prior mask to form a region mask by the computer aided diagnosis device; and

(e) segmenting a level-set contour from the 2D slice to detect the structure, wherein the step of segmenting is constrained by the region mask by the computer aided diagnosis device.

2. The method of claim 1 , further comprising: (f) repeating steps (a) to (e) for every slice in the 3D CT knee image.

3. The method of claim 2 , further comprising: outputting the results of step (f) in a 2D stack.

4. The method of claim 2 , further comprising: outputting the results of step (f) in a 3D surface.

5. The method of claim 1 , wherein the step of (a) creating an intensity prior mask comprises:

estimating a probability distribution of an intensity of the 2D slice;

detecting four extrema in the probability distribution; and

labeling the four extrema.

6. The method of claim 5 , wherein the step of estimating a probability distribution comprises: computing a histogram of the intensity of the 2D slice.

7. The method of claim 5 , wherein the step of labeling the four extrema comprises:

labeling a first extrema for a black background;

labeling a second extrema for dark gray matter;

labeling a third extrema for light gray matter; and

labeling a fourth extrema for white matter.

8. The method of claim 1 , wherein the step of (c) creating a geometric prior mask comprises:

raster scanning pixels from the seed point out towards each background boundary;

detecting two of the scanned pixels that hits the background based on the intensity prior mask;

determining a diameter of the geometric prior mask from a width of the two detected scanned pixels; and

creating the geometric prior mask with the diameter.

9. The method of claim 1 , wherein the step of (e) segmenting by a level-set contour from the 2D slice, comprises:

forming an initial 2D contour around the seed point, wherein the contour is represented in terms of a level-set function;

propagating a level-set contour based on the initial 2D contour, the level-set contour being propagated with a region-based speed that assumes that the Hoffa's pad is a piecewise-constant region; and

constraining the propagation of the level-set contour by the region mask;

wherein the step of propagating continues until convergence.

10. The method of claim 1 , wherein the step of (b) receiving or estimating a seed point in the 2D slice, comprises: receiving the seed point from a user interface for selecting the seed point.

11. The method of claim 1 , wherein the step of (b) receiving or estimating a seed point in the 2D slice, comprises: automatically estimating the seed point utilizing the created intensity mask and relative positions of bone regions with respect to fat regions in the anterior region of the CT knee image.

12. A program storage device readable by a machine, tangibly embodying a program of instructions executable by the machine to perform a method of detecting a structure in a three-dimensional (3D) computed tomography (“CT”) knee image, the method comprising:

creating an intensity prior mask based on a two-dimensional (2D) slice of the CT image;

estimating a seed point in the 2D slice, the seed point being in a Hoffa's pad region;

creating a geometric prior mask based on the seed point;

combining the intensity prior mask and the geometric prior mask to form a region mask; and

segmenting a level-set contour from the 2D slice to detect the structure, wherein the step of segmenting is constrained by the region mask.

13. The program storage device of claim 12 , wherein the estimating a seed point comprises:

determining a major axis direction by taking a difference between a position vector of the centroid of a bone region of the intensity mask and a position vector of the centroid of a non-black foreground region of the intensity mask; and

searching along the major axis direction for a seed point in a Hoffa's pad region of the intensity mask.