A Comparative Study On Algorithms In Different Stages Of Breast Mass Segmentation In MRI

Nowadays, breast cancer is one of the most common diseases of modern women. For thegrowing social pressure in recent years, the modern women have many unhealthy habits, such asgiving birth to a child without breastfeeding and late childbirth. Therefore the incidence andmortality of breast cancer are rising rapidly. According to the statistics, the incidence of breastcancer is almost3%~4%every year in China. As to improve the diagnostic accuracy and objectivityof the breast cancer, Breast MRI plays an important role in curing the breast cancer. It is anessential diagnostic tool which can lighten the diagnostic workload of radiologist. In addition, as toautomaticly segment and measure the lesion in medical image, the computer technology isindispensible.On account of the specificity of medical image and the unobvious image contrast of lesions, itis unable to create the stable and universal method to segment a variety of medical images withrandom complex backgrounds. Through the research of several segmentations which were skilled indifferent stages, we could combine them into a new algorithm which is more stable and accurate. Inthe meanwhile, we could improve the algorithm by repeatedly researching mass segmentations inthe field of breast MRI (Magnetic Resonance Imaging). Moreover, this method is more outstandingthan a single segmentation algorithm which only forms the initial positioning to the final division.Through the partitioned and comparative experiment, this report analyzes the inhomogeneousand fuzzy characteristics of the gray value of breast MRI. Furthermore, the report emphasize onusing the features of sFCM(spatial Fuzzy c-means clustering algorithm). It was applied to thesegmentation of the object with fuzzy boundaries, and then the GVF snake model can find thesegmentation of the fuzzy boundary. Additionally, we can obtain a stable Breast MRI segmentationmethod by bringing in the correlation of the adjacent inter-frame: using sFCM to choose severaldifferent clusters, we can obtain the initial contour position; According to this initial contourposition, we can then obtain the fine segmentation results by GVF Snake Model; Owning to theseveral different clusters, there are several different segmentation results. Besides, there are twocharacteristics of Breast MRI: Firstly, the area of the adjacent inter-frame’s tumor is changingslowly. Secondly, the morphological of tumor is similar. Taking advantage of the characteristics,we can screen the area of tumor as a condition to optimize the method.