Deformable models for segmentation of medical ultrasound images
Vikram Chalana, Yongmin Kim · 1996
Segmentation of anatomical organs is the first step towards obtaining diagnostically important quantitative information from ultrasound images. This research focuses on two clinically important applications to which segmentation techniques have been applied--the detection of left-ventricular boundaries in echocardiograms and the detection of fetal head and abdomen in obstetric ultrasound images. Deformable models provide an effective method for segmenting ultrasound images because they allow the integration of multiple sources of information. A type of deformable model known as the active contour model was used in the ultrasound image segmentation problems. The active contour model has been extended to three dimensions which allows the segmentation of boundaries on image sequences or on spatial 3-D data. We have also developed methodologies to automatically estimate the various algorithm parameters. The boundary detection procedure has been evaluated on clinical data sets by comparing the computer-detected boundaries to boundaries hand-outlined by four experienced observers. Using a boundary distance measure, it was shown that the computer-generated boundaries were as far from an observer's boundaries as the observers' boundaries are from each other. Also, the parameters derived from the computer-generated boundaries such as the areas enclosed by the boundaries or the circumference of the boundaries agreed very well with the parameters derived from the hand-outlined boundaries.