| Magnetic Resonance Imaging (MRI) plays a more and more important role in medical image area for its advantages such as no radiation, multiple imaging, high-resolution in space.Now the scope of MRI application has been extended to all parts of the body and widely used in lesion tissue detection and minimally invasive surgery. The application of image-guided surgical navigation system based on MRI has also been a good development.As a follow-up operation and providing good condition for image-guided surgical navigation system, complete and accurate segmentation of the liver or other organs of the abdomen is very important and necessary.However, the segmentation of abdominal organs is facing many problems:1) Soft tissue is the main part of abdominal tissue, and there is little difference between their intensity. The adjacent organs are closed to each other, even overlapping, thus lead to obvious boundary leakage.2) The shape of each individual organ changes complex, they are varies from age, gender, weight and so on.3) Due to the inherent factors of MRI apparatus, there is large image noise. There is a certain partial volume effects and intensity inhomogeneity. These factors mentioned above make the segmentation of MR images becoming difficult.In order to overcome these problems, this article makes a deep research of segmentation algorithms, the main works of this article as follows:1) Making a detailed description of physical principles of MRI and a summary of intensity inhomogeneity, taking some experiments of two kinds of representative filed-correction algorithms.2) Making a detailed literature review of all current common segmentation algorithms, analyzing the advantages and disadvantages of various algorithms and application scenarios, and giving the research directions in the future.3) Proposing a novel method for abdominal organs’segmentation of MR images which incorporates the kernel Graph cuts (KGC) with shape prior information. Kernel Principal Component Analysis (KPCA) was used to train the shape template set after registration procedures with initial contour. The shape prior information is integrated into the KGC energy function to form a new model. The experimental results show that we can get a satisfied result without boundary leakage and error segmentation for similar tissues. |
