Fully automatic colon segmentation in computed tomography colonography

Abstract

Colon cancer is the second leading cause of cancer-related death in the United States, and can be prevented by the removal of precancerous colon polyps. For colon diagnosis, computed tomography colonography (CTC) has been proposed as a minimally invasive technique, and computer aided diagnosis (CAD) systems using CTC data are a rapidly evolving tool to localize, detect, and identify colon polyps. Colon segmentation is an essential and challenging step in the development of CAD systems. To accurately segment the whole colon using CTC data, we propose a fully automatic method. In this work, the whole body region excluding the lungs is first localized to narrow the search region and lower computation burden. Inside the body of the test case, a pre-trained colon atlas probability map is fitted using anatomy constraints to localize parts of the colon as seeded regions. Then, region growing is applied to generate an initial 3D segmentation. Below colon air, discriminative classifiers are used to classify regions into colon-tagged materials or non-colon regions, and a fuzzy connectedness segmentation method is applied. Combining colon air and tagged residuals, the whole colon is extracted from CTC data. Experiments were conducted on publicly available CTC database which results in better accuracy and error rate compared with other methods.