Methods of obtaining geometry from images
In one aspect, a method of detecting at least on feature associated with a blood vessel in at least one image of at least one blood vessel using a matched filter adapted to respond to the at least one feature is provided. The method comprises applying a scale detection filter to selected voxels in the at least one image to determine a scale for the matched filter at each of the selected voxels, determining an orientation for the matched filter at each of the selected voxels, wherein determining the orientation is assisted by using the scale determined at each of the selected voxels, applying the matched filter at each of the selected voxels at the scale and the orientation determined at each of the selected voxels to obtain a filter response at each of the selected voxels, and analyzing the filter response at each of the selected voxels to determine if the respective voxel corresponds to the at least one feature.
1. A method of extracting vessel geometry from one or more scan images, comprising:
determining a scale for a voxel in the one or more scan images;
determining an orientation of the voxel;
applying a matched filter using the scale and orientation to generate a filter response;
determining, based on the filter response, whether the voxel represents a centerline of a vessel;
repeating the above acts to identify a plurality of centerline voxels; and
linking the plurality of centerline voxels together, wherein the linking comprises selecting a second voxel to link to a first voxel based on a likelihood that the second voxel is adjacent to the first voxel and part of the same vessel, wherein the likelihood is calculated by:
determining a distance between the first voxel and other voxels;
determining a difference between the orientation of the first voxel and other voxels; and
determining a difference between the filter response of the first voxel and other voxels.
2. The method of claim 1 , wherein the scale is determined using a scale detection filter.
3. The method of claim 1 , wherein the orientation is determined using the scale.
4. The method of claim 1 , wherein the vessel geometry comprises a poker chip representation at each of the centerline voxels, the poker chip representation including a location of the centerline voxel, a radius, and a direction.
5. The method of claim 4 , further comprising sorting the centerline voxels into bins based on radius.
6. The method of claim 5 , wherein a binned micro-vessel density (BMVD) analysis is performed using the bins.
7. The method of claim 6 , wherein the BMVD is used to detect and diagnose diseases associated with tubular networks.
8. The method of claim 6 , wherein the BMVD is used to track disease progression.
9. The method of claim 6 , wherein the BMVD is used to monitor a response to therapy.
10. The method of claim 1 , wherein distance between the first voxel and other voxels is weighted more heavily than other criteria.
11. The method of claim 1 , wherein a curve representing the centerline of the vessel is obtained by integrating orientation vectors for all centerline voxels in the vessel.
12. The method of claim 11 , wherein a curvature of the vessel is obtained using derivatives of the orientation vectors.