Research On Medical Image Segmentation Algorithms

Multiple Sclerosis(MS) is an autoimmune disease of central nervous system. magnetic resonance imaging(MRI) is increasingly used in the diagnosis of Multiple Sclerosis in recent years. Manual delineation of MS lesions in MR images by human expert is time consuming, and would also be subjective and have poor reproducibility. However, the progression of the MS lesions shows considerable variability and MS lesions present temporal changes in shape, location, and area between patients and even for the same patient, which renders the automatic segmentation of MS lesions a challenging problem. Based on traditional medical image processing, we proposed a method for automatic segmentation of MS lesions. For the research of MS lesion segmentation, I did the major work as follows:1. This article summarizes some common methods of medical image segmentation. The application of these methods on MS lesions’ segmentation is not suitable, however, these methods lay a foundation for the new method of MS lesions’ segmentation which is proposed later in this article.2. This article introduces the recent development of MS lesions’ segmentation, and mainly describes two methods of MS lesions’ segmentation, one is based on support vector machine(SVM), the other is based on Gaussian mixture model(GMM). Both methods have some disadvantages but also have some advantages of its own.3. This article proposes a new multi-channel method for MS lesions’ segmentation based on the method for the segmentation of normal brain’s MR images. This method could merge together the complementary information of three modalities’ MR images(T1, T2, FLAIR). The multi-channel segmentation is based on the energy minimization framework for simultaneous estimation of the bias field and segmentation. It’s result will be four different tissues, white matter(WM), gray matter(GM), cerebrospinal fluid(CSF), MS lesion(lesion).4. For the complete segmentation of MS lesions, the MS lesions should be enhanced on the outer CSF before the multi-channel segmentation. This article proposes two methods for enhancing, one is based on the histogram statistics, the other is based on the bias field. From the experiment result, the method based on the histogram statistics is obviously better.5. This article proposes a multi-channel Region-Scalable Fitting(RSF) method for the post processing of MS lesions resulted from multi-channel segmentation. This method makes the multi-channel’s lesions’ result as initial level set, and after the processing the final MS lesions’ result is much better.