Unsupervised Renal Cortex Image Segmentation Utilizing Cellular Automata And Markov Random Field Model

Kidney, as one of the most important metabolic organs in human body, plays animportant role to maintain metabolic balance. In clinical, the doctors can determine kidneyhealth according to some kidney indicators, such as the thickness of renal cortex, theoverall pattern and so on. For making the kidney medical images more suitable for doctorsto observe and the subsequent analysis and operation, we need to segment these images andextract the target regions from the complex background around. However, most of theresearches aim at segmenting the whole kidney. The research focusing on the segmentationof some kidney internal structure is rare. Because the kidney internal structures arerelatively complex and the abundant background information is very easy to cause somemis-segmentation.Generally we can classify the image segmentation algorithms into two different types.The first one is called classic segmentation algorithm. Another one can be called newsegmentation algorithm combined with the specific tools and new theoretical method. Thispaper will introduces them concisely, including their basic definition and characteristics.We will also discuss their deficiency and points needed to be improved. What’s more, wewill focus on the image segmentation methods based on CA and MRF models.In order to segment kidney automatically, based on the markov random field (MRF)model and the GrowCut algorithm which utilizes Cellular Automata theory, a newunsupervised cortex segmentation method is proposed. To most of the traditional methods,if we want to achieve the goal of segmenting an image sequences, we should mark theseeds for the foreground and background respectively in every slice of the sequences. Butthe new methods can get it automatically. The selected slice is marked by the proposed seedpoint marker to get the initial seed points; According to these initial points, the seeds forevery slice can obtained automatically by the seed point dynamic evolution mechanism;What’s more, a global CA model and local MRF models are established based on the seedsin every slice; Then the CA state transformation rules, determined according to the energyfunction minimization principle, guide the update of cells. Finally the image will besegmented. The experimental results prove the effectiveness of the new method proposed inthis paper, a new algorithm, achieve no under human intervention to increase and the increase of the accuracy of the segmentation of CT images3d kidney. It realized the acutesegmentations of enhance and non-enhance3D kidney images without any manualintervention.