Bone segmentation

1. A method of automatically identifying bone voxels in a medical image data set of voxels that includes at least a portion of the large bones of the legs and at least a portion of the bones outside the legs, the method comprising:

a) identifying at least a portion of the voxels belonging to large bones of the legs by applying a first procedure to the voxels of the data set, comprising finding first connected sets of volume greater than a minimum volume, of voxels having density greater than a first minimum density;

b) identifying at least a portion of the voxels belonging to bones of the body outside the legs, by applying a second procedure to the voxels of the data set, comprising finding second connected sets of voxels having density greater than a second minimum density; and

c) excluding those voxels from the second connected sets of voxels that satisfy a criterion involving one or more of average density, uniformity of density, and thickness, thereby excluding voxels belonging to the aorta exceeding a threshold level of contrast agent;

wherein the second minimum density and the criterion are chosen so that voxels belonging to the large leg bones would also be excluded if the second procedure and the criterion were applied to them.

2. The method according to claim 1 , wherein the data set is defined with a z-coordinate where the z axis is a longitudinal axis of the body, the method also including identifying at least a portion of the voxels belonging to the knee bones by:

a) distinguishing femur voxels from lower leg bone voxels by their z-coordinates; and

b) identifying as belonging to the knee bones, voxels exceeding a threshold image density, whose z-coordinates are intermediate between the range of z-coordinates of the femur voxels and the range of z-coordinates of the lower leg bone voxels.

3. The method according to claim 1 , wherein the image data set is defined in an x-y-z Cartesian coordinate system having an x, y, and z axis where the z axis is a longitudinal axis of the body, and the x axis is a left-right axis of the body, and wherein the criterion comprises:

a) finding connected regions of voxels which have an image density exceeding an image density threshold value;

b) for each connected region, finding a range W in x-coordinate, a range H in y-coordinate, and a range D in z-coordinate, of the centers of the voxels in the region; and

c) identifying as belonging to the large leg bones a set of the connected regions for which the range W is similar in to the range H, and the range D is larger than for the other connected regions.

4. The method according to claim 3 , wherein finding connected regions of voxels comprises excluding connected regions which are less than a minimum volume, and the minimum volume and the image density threshold value of the image density are chosen so that any connected regions comprising voxels of blood vessels in the legs are excluded, but at least one connected region comprising voxels of large leg bones is not excluded.

5. The method according to claim 1 , wherein the data set is defined with a z-coordinate where the z axis is a longitudinal axis of the body, and wherein the first procedure comprises selecting voxels having image density greater than a first threshold value, such that only voxels in a central portion of each of the large bones of the legs are identified, the method also comprising identifying voxels belonging to end portions of the large bones of the legs, by selecting voxels which have image density greater than a second threshold value lower than the first threshold value, and which have their z-coordinate within a range determined by the locations of the voxels in the central portions of the large bones of the legs.

6. The method according to claim 5 , wherein the range excludes a central portion of the femur where blood vessels are close to the femur.