Research On Segmentation Method Of Breast Tumor CT Image

Breast cancer is the most common type of cancers among women worldwide, which hasbecomes a serious threat to the health of women in multiple diseases. How to effectivelyimprove the diagnoses and treatments of breast tumor, and receive a reduction infringement topatients, has becomes an urgent problem. For this reason, whether the segmentation of breasttumor is accuracy or not can provide an important basis on the selection of clinical diagnosisand radiation therapy, which is also directly related to the patients’ treatment effectiveness.Therefore, how to improve the accuracy of breast tumor segmentation has been the focus ofprofessionals’ research.The study on this thesis is to segment breast tumor CT images. In view of breas t tumorexisting different features, such as inhomogeneous gray scale and rough surrounding, astrategy is proposed. Firstly, the suspected tumor lesion is extracted according to the grayscale and shape characteristics; Secondly, the pre-processing is employed to enhance thetumor lesion; Then two different methods of segmentation of breast tumor are utilized, whichis based on the condition of whether the boundaries of tumor and its surrounding tissues be ingclose or not.This thesis mainly focuses on the following points:(1) The pre-processing is necessary before segmenting breast tumor. It mainlyincorporates of extraction and enhancement of suspected tumor lesion. The former step isbased on the tumor’s internal features, suchas gray scale, shape and statistical characteristics;The latter step is performed with morphology filters to stretch contrast, which can remove thenoises and details and highlight the gray levels of breast tumor area and non-tumor areas.(2) When breast tumor having sharp edges and obvious gray levels with the surroundingtissues, a coarse-to-fine strategy is proposed for breast tumor segmentation. Firstly,morphological reconstruction filters is applied to remove the small burrs and irregular edgesin order to avert the new boundaries and boundary migration. Secondly, the local maximumarea is marked and the minima imposition is employed to modify the gradient image, thenstandard watershed transform is used to implement the coarse segmentation based on themodified gradient image. Finally, for the purpose of achieving fine portion, the curve formedby selected seed points on the watershed line was chosen as the initial contour of snake modelto dynamically approach the target’s real edges.(3) When breast tumor existing adhesive or relatively close edges with its surroundingtissues, the above method can produce over-segmentation, that is, the segmentation resultincluding tumor area and non-tumor areas, which directly segmenting these images. In orderto avoid over-segmentation, point distribution model (PDM) is utilized to train the sample images and extract the average shape of breast tumor, then superimpose and deform theaverage shape so as to attach the real tumor contour in a new image. Ultimately, thesegmentation results are received with snake approximating and tracking targets boundaries.To demonstrate the effectiveness of segmentation results, some clinical CT images arechoosed to perform tests on MATLAB2012b platform. Experiments show that differentmethods segmenting different breast tumor CT images can receive a higher accuracy.