Automatic Lung Tumor Segmentation with Leaks Removal in Follow-up CT Studies

Abstract

We present a novel automatic algorithm for lung tumors segmentation in follow-up CT studies. The inputs are a baseline CT scan and a delineation of the tumors in it; the output is the tumor delineations in the follow-up scan. The algorithm consists of four steps: (1) deformable registration of the baseline and follow-up scans; (2) segmentation of the tumors in the follow-up scan; (3) geometry-based segmentation leaks correction; and (4) tumor boundary regularization. The key advantage of our method is that it automatically builds a patient-specific prior that increases segmentation accuracy and robustness and reduces observer variability. Our experimental results on 80 pairs of CT scans from 40 patients with ground-truth segmentations by a radiologist yield an average overlap error of 14.5 % (std = 5.6), a significant improvement from the 30 % (std = 13.3) result of stand-alone fast marching segmentation.