Level Set-based Vessel Segmentation Of Medical Images

Vessel segmentation is essential for many clinical applications, especially the image-guided radiation therapy (IGRT) by magnetic resonance imaging(MRI). In practice, the implementation of surgical resection for hepatocellular carcinoma (HCC) is limited due to poor tumor location. Therefore, IGRT become a crucial treatment for HCC, which requires accurate analysis on the structures of liver vessels (such as portal vein, hepatic vein, hepatic artery, and bile duct) and the locations of tumors related to these structures. Users can manually segment vessels well through visual judgment, but it is almost unpractical that every slice is segmented one by one for hundreds slice of a dicom dataset. Therefore, automatic segmentation algorithm is required.This paper presents an improved variational level set method, which uses non-local robust statistics to suppress the influence of noise in MR images.The non-local robust statistics, which represent vascular features, are learned adaptively from seeds provided by users. K-means clustering in neighborhoods of seeds is utilized to exclude inappropriate seeds, which are obviously corrupted by noise. The neighborhoods of appropriate seeds are placed in an array to calculate the non-local robust statistics, and the variational level set formulation can be constructed. Bias correction is utilized in the level set formulation to reduce the influence of intensity inhomogeneity of MRI. Experiments were conducted over real MR images, and showed that the proposed method performed better on small hepatic vessel segmentation compared with other segmentation methods.Besides, we present a two-stage hybrid method to distinguish normal brain tissue from lesion regions based on the T1-weighted and T2-weighted MR images of the same anatomic structure. In the first stage, normal brain tissue was segmented using an iterative Otsu’s thresholding method, which is extracted from T1-weighted MR images, and lesion regions were extracted from T2-weighted MR images. In the second stage, an MRF-MAP approach was used to classify areas dually identified as normal brain tissue in T1-weighted and lesion regions in T2-weighted images. The approach was validated using synthetic and real MR images to demonstrate its ability to clearly distinguish normal brain tissue from lesion regions.