Segmentation Of Intima-Media In Ultrasound Common Carotid Artery Images

Common carotid artery (CCA) intima-media thickness (IMT) is an important indicator to cardiovascular diseases. Ultrasound imaging is widely applied in clinical environment for visualization of carotid artery, in which the traditional measurement of IMT is based on manual tracing. To solve the problems such as laborious work and large intra- inter- observer variability accompanying manual measurement, it is necessary to research an automatic method for IMT measurement via image processing techniques.The key problem in automatic IMT measurement is identification of two nearly parallel boundaries. Based on the characteristics of CCA images and previous research methods, this thesis proposes a distance constraint based active contour model for intima-media segmentation, which comprises an initial contour construction and a curve evolution step. It first uses directional derivative to compute the edge map, which is thresholded and divided into several equal width non-overlapping segments. In each of these segments, according to the characteristics that the local boundaries are nearly parallel and satisfy a certain distance range, two line segments are obtained to approximate the boundaries though Hough transform. The initial contours are constructed by sampling the line segments that are corrected based on the neighboring references. The final step uses a dual snake model to evolve the two initial contours via keeping them smooth, close to boundary and a uniform distance between to obtain the final segmentation results.Fifty images from Zhongnan Hospital of Wuhan University were selected to evaluate the proposed method. The comparisons with respect to the ground truth data among the proposed method, traditional dynamic programming, dual dynamic programming and traditional snake model demonstrated that the new method had the least segmentation error. Moreover, clinical evaluation indicated that the mean and max values of IMT had errors from -0.03mm to 0.08mm, from -0.1mm to 0.09mm respectively within 95% confidential interval when compared to those derived from ground truth data, which satisfies clinical requirements.