Mammogram Segmentation Based On Improved Level Set Model

Breast cancer is one of the most common malignant tumors. And mass is an important symptom of breast cancer. Mammography is the most efficient and common tool for early diagnosis of breast cancer. However, due to the masses in mammograms have blur edge, irregular shapes as well as factors such as low contrast, which made it difficult to diagnose. The goal of our paper is to use our method to segment mass and assist radiologists for diagnosis. Our method employ spiking cortical model and improved CV model to segment mass in mammograms. First of all, selecting the largest connected region, seeded region growing, and nonlinear un-sharp masking scheme(NLUM) are used to remove the label and enhance the mammogram. Secondly we apply the Spiking Cortical Model(SCM) on the pre-processed image to locate the lesion as an approximate parametric circle which would be used as the initial contour of CV model. Finally, the mass boundary is accurately segmented by the improved CV model using Local Region-Scalable Force. The validity of the proposed method is evaluated by two well-known digitized datasets(digital database for screening mammography and mammography image analysis society database), and the detection rates of the proposed method are 90.63% and 93.75% respectively. The main contributions of this dissertation can be summarized as follows:1. We give a detailed description of the level set method, as well as the advantages and disadvantages of the level set algorithm’s practical application.2. We introduce improved level set method and apply this model to mammography. The simulation results obtained by comparing our method with the other six comparative methods, which prove that our model has a better effect than the others for mass segmentation.In order to get more accurate results, we applying the methods of preprocessing, locating the outline of masses to mammograms. The methods are as follows:Firstly, to processing the mammograms, such as removing the backgrounds and label, image enhancement. After that, applying the Spiking Cortical Model(SCM) to locate the mass as the level set is sensitive to the initial counter. In order to make the contours converge to the masses with low contrast and fuzzy boundary, we proposed an improved level set method based on local gray-level statistics characteristic. The results of our method is better than the other six compared algorithms, giving a precise contour lines and preserving the detail information of the contours,which has a significant meaning for practical application and assisting detection.