A New Adaptive Method for 3 D Liver Segmentation with Active Contours

Jialin Peng,Dexing Kong,Fangfang Dong

2011-01-01

Abstract:Liver segmentation is not only the key process for volume computation but also fundamental for further processing to get more anatomy information for individual patient. Due to the low contrast, blurred edges, large variability in shape and complex context with clutter features surrounding the liver that characterize the liver CTA images, it is a convoluted problem and still a challenge task to robustly and accurately segment the liver. In this paper, we overcome these difficulties with a novel variational model based on the idea of intensity probability distribution propagation and region appearance propagation with which we can focus on the target liver regardless of how complex the uninterested background is. Our model consists of an edge based term and two region based term which integrate both region intensity information and region appearance. The model we proposed is robust to model parameters, image noise and greatly alleviate the strict requirement of the scanning protocol and data quality limitation. Moreover, with a spatially varying weight to balance between the region term and edge term, our model can properly switch between the region item force and the edge term force, which makeS our model much more robust and perform well for the difficult liver segmentation problem. While segmentation in a slice by slice style often neglects the consecutiveness between slices, we segment the liver from the 3D volume data directly. Experimental results show that our model segments the liver region with accuracy: the liver can be effectively segmented distinguished from the complex background and the vessels are also simultaneously isolated from the liver with accuracy. Our system is promising for stable practical use. Last but not least, Our model is a nearly automatic one which needs only an initial contour/surface inside the liver and is also can be used to segment other abdominal organs.