Segmentation of Liver in Low-contrast Images Using K-Means Clustering and A Priori Knowledge

摘要

In this paper, we address the problem of liver segmentation from low-contrast CT-scan datasets. We consider a 'Gaussian Mixture' model for intensity distribution of liver and non-liver tissues and use a priori knowledge to find statistical parameters of liver. Then, we apply thresholding in a narrow range round the mean of each component of liver's mixture model to find liver candidate pixels. K-means clustering is used to discriminate between liver and non-liver index pixels. Then, we establish a liver probability map by assigning a probability number to each pixel of the original image. We can find initial boundary for liver by thresholding the image map and use it as the input to a 'Geodesic Active Contour' algorithm to find final liver boundary. We tested the proposed algorithm on non-contrast liver datasets. Assessment of the results proves that the proposed method is both robust, resistant to leakage, requires minimum level of interaction.