Ovarian follicle segmentation in ultrasound images

1. A method, comprising:

constructing a variation image from an ovarian follicle B-mode ultrasound image, wherein the variation image is indicative of local variations in the ovarian follicle B-mode ultrasound image, and wherein the variation image includes one of a variance image and an entropy image;

constructing a binary image from the variation image and the ovarian follicle B-mode image;

identifying connected components in the binary image, wherein each connected component corresponds to a different ovarian follicle candidate;

constructing a coarse follicle mask from the binary image where each identified connected component represents a mask for a corresponding different ovarian follicle candidate in the binary image;

optimizing contours of the follicles in the coarse follicle mask; and

segmenting one or more ovarian follicles from the ovarian follicle B-mode ultrasound image with the optimized follicle mask.

2. The method of claim 1 , wherein constructing the binary image includes:

setting a value of a pixel in the binary image to a value of one only if a first corresponding pixel in the variation image is less than a first predetermined threshold and a second corresponding pixel in the ovarian follicle B-mode ultrasound image is less than a second predetermined threshold.

3. The method of claim 1 , further comprising: pre-processing the ovarian follicle B-mode ultrasound image, constructing a pre-processed image, wherein the pre-processing includes at least one of low pass filtering the ovarian follicle B-mode ultrasound image and removing speckle from the ovarian follicle B-mode ultrasound image.

4. The method of claim 3 , wherein constructing the binary image includes:

setting a value of a pixel in the binary image to a value of one only if a first corresponding pixel in the variation image is less than a first predetermined threshold and a second corresponding pixel in the pre-processed image is less than a second predetermined threshold.

5. The method of claim 1 , wherein the binary image includes noise, wherein the noise includes at least one of a pixel with a value of one in a region of pixels with values of zero and a pixel with a value of zero in a region of pixels with values of one, and further comprising: processing the binary image to remove the noise.

6. The method of claim 5 , wherein the binary image is processed with at least one of morphological opening and closing operations, dilation and erosion.

7. The method of claim 1 , further comprising:

extracting only identified connected components having an area that satisfies a predetermined follicle size threshold; and

constructing the coarse follicle mask with the extracted connected components.

8. The method of claim 1 , wherein the binary image includes at least one connected component that has a non-convex hull shape, and further comprising: morphologically filtering the connected component which re-shapes the at least one connected component so that it has convex hull shape.

9. The method of claim 8 , wherein the morphologically filtering employs a disk-shaped kernel with an area on an order of one tenth of the area of the connected component.

10. The method of claim 8 , further comprising:

comparing an area of a morphologically filtered connected component with an area of a corresponding connected component;

dilating the connected component with a first scaling factor in response to a difference in the areas satisfying a predetermined threshold; and

dilating the connected component with a second scaling factor in response to a difference in the areas not satisfying the predetermined threshold.

11. The method of claim 8 , further comprising:

extracting only dilated connected components having an area that satisfies a predetermined follicle size threshold; and

constructing the coarse follicle mask with the extracted connected components.

12. The method of claim 1 , wherein follicle walls are optimized by finding an optimal curve in a polar representation of each follicle such that a predetermined set of follicle characteristics are satisfied.

13. The method of claim 12 , wherein the optimizing includes:

multiplying the ovarian follicle B-mode image by a follicle mask, producing only the follicle in a Cartesian coordinate system;

calculating a center of mass of the follicle mask;

constructing a polar map of the follicle using pre-specified angular and radial resolutions; and

optimizing the contour by minimizing a cost function.

14. An apparatus, comprising:

a follicle identifier that receives an ultrasound B-mode image generated by an ultrasound imaging system, wherein the follicle identifier includes:

a coarse estimator configured to employ morphological filters to identify potential ovarian follicles in the ultrasound B-mode image and determine a coarse outline of their boundaries, wherein the course estimator constructs a variance image or an entropy image from the ultrasound B-mode image, constructs a binary image from the variance or entropy image and the ultrasound B-mode image or a pre-processed ultrasound B-mode image, identifies candidate ovarian follicle in the binary image through connected components, and constructs a course follicle mask which includes the coarse outline of the boundaries, where each identified connected component represents a mask for a corresponding different ovarian follicle candidate in the binary image; and

a fine estimator configured to employ a non-linear optimization to refine the borders.

15. The apparatus of claim 14 , wherein the coarse estimator identifies, automatically or semi-automatically, a dominate follicle in the coarse follicle mask, and further comprises: displaying the coarse follicle mask with indicia highlighting the dominate follicle.

16. The apparatus of claim 14 , wherein the coarse estimator morphologically opens and closes the binary image generating a de-noised binary image, and morphologically filters the candidate ovarian follicles to re-shape candidate ovarian follicles not satisfying a predetermined follicle shape.

17. The apparatus of claim 14 , wherein the fine estimator multiplies the ultrasound B-mode image by a follicle mask producing only the follicle in a Cartesian coordinate system, calculates a center of mass of the follicle mask, constructs a polar map of the follicle using pre-specified angular and radial resolutions, and optimizes the contour through minimizing a cost function based on the polar map.

18. A non-transitory computer readable medium encoded with computer executable instructions, which, when executed by a computer processor, causes the processor to:

construct an entropy image with a B-mode image;

low pass filter the B-mode image;

construct a binary image based on pixels values of both entropy image and the low pass filtered B-mode image;

morphologically filter the binary image;

identify connected components in the morphologically filtered binary image;

create a coarse follicle mask with the connected components having an area that satisfied a predetermined threshold;

refine contours of the follicles in the coarse follicle mask; and

identify ovarian follicle B-mode image based on the follicles with the refined contours.