algorithm and clinical validation study

We present a new algorithm for a nearly automatic liver segmentation including liver, tumors and vessels segmentation and volume estimation in abdominal CT datasets and its clinical evaluation and validation. Our algorithm uses a multi-resolution iterative scheme which repeatedly applies smoothed Bayesian classification followed by set of morphological operators and active contours refinement. Its advantages are that it requires only one seed point, in case there is no tumor in the CT, or two seed points - in case there is a tumor in the CT. It is fast, accurate, and robust. In a retrospective study on 26 CT datasets, we compared our method to a ground truth manual segmentation performed by a radiologist, and to semi-automatic and automatic commercial methods. For each method, we computed the liver volume from the segmentation and recorded the time required to produce it. Our liver segmentation method is accurate, robust, efficient, and easy to use. It is an effective tool for hepatic volume estimation in a clinical setup.