Rectal tumor boundary detection by unifying active contour model

The Project of 3D reconstruction of rectal wall structure aims at developing an analysis system to help surgeons cope with a large quantities of rectal ultrasound images, involving muscular layer detection, rectal tumor detection, and 3D reconstruction, etc. In the procedure of tumor detection, a traditional active contour model suffers some difficulties for finding the boundary of tumor when it deforms from the “seed” in the interior of a tumor. In this paper, we proposed a novel united active contour model with the information of image region feature and image gradient feature for the purpose of tumor detection. Region-based method added in the model, however, introduces a statistical method into the segmentation of the image and hence becomes less sensitive to noise. The originality in this algorithm is that we introduce a Gaussian Mixture Model (GMM) into the statistical model description of “seed” region. This model can perform more accurate and optimal statistical description than a single Gaussian model. A K-means algorithm and an Expectation Maximization (EM) algorithm are used for optimal parameter estimation of GMM. The experimental results show the new model has more optimal performance for image segmentation and boundary finding than classical active contour model.