Brain Mr Image Segmentation Techniques

Medicine plays an important role in human history. With the development of meidical imaging, like CT, X-ray, MRI, it turn out to be impossible that people analyze all the growing medical images. So it call for computer to take the task.The aim of this paper is to segment brain MRI, and break it into gray matter, white matter, Cerebrospinal Fluid(CSF), and background. MRI is always corrupted by intensity inhomogeneity or bias field, which is caused by equipment limitations and patient-induced electrodynamic interactions and is inherent to MRI. Bias field vary smoothly and spatially. So it become impossible for intensity-based methods to segment MRI correctly. The work in this paper is as follows:1. Research on methods to remove bias field. Chapter three focuses on polynomial fitting method to remove bias field, and explaining Expectation Maximization(EM) algorithm. Leemput use digital brain atlas for initialization, which is miscellaneous, including affine transformation. This paper use another way that using kmeans to get initial prior estimate of the classes probabilities. However, polynomial fitting deals with a single pixel, resulting in non-smooth segmented boundary results. This paper will fix this problem in the next chapter.2. Research on methods to segment brain tissue. Firstly, chapter four introducs Snake model, analyzing its energy function, forces, and its computation, and then introduces Chan-Vese (CV) model, explains its relationship with Mumford Shah(MS) model, as well as it's relationship with level set and analyses it's computation through level set. Finally, the rest introduces LSR (Level Set Regularization (LSR) term) proposed by Chunming Li to improve segmentation results, which makes segmentation smoothly. The segmented boundary by CV keeps smooth, so it fix non-smooth problem in previous chapter.3. Based on above two methods, proposing a framework including the above two methods to segment brain MRI with bias field. Firstly, the method removes the bias field through polynomial fitting methods, then uses 4 phases CV model to segment the restored image.In future, we plan to do the bias removal procedure and segmentation procedure simultaneously.