A comparison of 3D methods for volume quantitation in SPECT

Three methods for detecting the edges of objects for segmentation of SPECT (single photon emission computed tomography) imagery are compared. The methods investigated are 3-D edge detection with nonmaximum suppression, 3-D Lacroix edge linking, and 3-D zero-crossings of the second directional derivative. Studies with simulated SPECT images show that the 3-D second directional derivative operator produces significantly more accurate volume quantitation than the other methods, especially when image contrast is low.<
>