Universal Liver Extraction Algorithm: An Improved Chan–Vese Model

Abstract

Liver segmentation is important to speed up liver disease diagnosis. It is also useful for detection, recognition, and measurement of objects in liver images. Sufficient work has been carried out until now, but common methodology for segmenting liver image from CT scan, MRI scan, PET scan, etc., is not available. The proposed methodology is an effort toward developing a general algorithm to segment liver image from abdominal computerized tomography (CT) scan and magnetic resonance imaging (MRI) scan images. In the proposed algorithm, pixel intensity range of the liver portion is obtained by cropping a random section of the liver. Using its histogram, threshold values are calculated. Further, threshold-based segmentation is performed, which separates liver from abdominal CT scan image/abdominal MRI scan image. Noise in the liver image is reduced using median filter, and the quality of the image is improved by sigmoidal function. The image is then converted into binary image. The Chan–Vese (C–V) model demands an initial contour, which evolves outward. A novel algorithm is proposed to identify the initial contour inside the liver without user intervention. This initial contour propagates outward and continues until the boundary of the liver is identified accurately. This process terminates by itself when the entire boundary of the liver is detected. The method has been validated on CT images and MRI images. Results on the variety of images are compared with existing algorithms, which reveal its robustness, effectiveness, and efficiency.

Keywords

• liver segmentation
• chan–vese model
• histogram-based threshold segmentation
• computerized tomography
• magnetic resonance image