This paper describes a method for automated liver segmentation using two different time-phases of CT images with and without contrast materials acquired from a multislice CT scanner. Because of patient's respiratory motion, images acquired at different phase might be out of alignment. So, we introduce the non-rigid registration, which enable us to get correct alignment of these images. We estimate the intensity distribution of the liver as a gaussian distribution in two dimensional feature space, whose axes correspond to two phases of CT values, and generate the liver region-enhanced image. We can get liver region by thresholding this liver region-enhanced image. We also define the candidate of the liver contour and modify the liver region by fitting this contour. We evaluated this method using four cases of CT data sets, and showed its usefulness.

View full abstract