Research On Feature Extraction Of Breast Tumor Ultrasound Images

Breast cancer is one of the most commonly diagnosed cancer types among women in the world.Early detection,early diagnosis,early treatment are the principle of prevention and treatment of breast cancer.Breast ultrasound(BUS)imaging has gradually become an important tool for early screening of breast cancer due to the advantages of wide adaptation,low cost,no radiation,improved sensitivity and non-invasive.In this paper,an accurate and fast automatic segmentation algorithm for breast ultrasound images is presented.The shape features and texture features are extracted from the segmented images of the suspicious lesions.The features are the foundation for the identification of malignant and benign breast tumors.It provides a reference for the clinical diagnosis of breast cancer,improves the diagnostic efficiency,and decreases the rate of misdiagnosis and missed diagnosis.The contents of this paper mainly consist of the following three parts: pretreatment,image segmentation and feature extraction.In the part of pretreatment,the median filter,mean filter,Gaussian filtering and anisotropic diffusion filtering model are analyzed of the effect of noise reduction in the ultrasound images.The anisotropic diffusion model is chosen to improve the quality of the image while decreasing speckle noise.In the part of segmentation,the advantages and the disadvantages of graph cut theory and DRLSE model are introduced,and a semi-automatic segmentation algorithm of graph cut and level set are proposed to improve the speed,accuracy and objectivity of the segmentation.Then,an automatic segmentation algorithm is proposed based on Otsuthreshold segmentation and an improved CV model for breast ultrasound tumor.The initial region which is used as the initial contour is obtained using the Otsu threshold segmentation algorithm and morphological filtering.The final segmentation result is gained by using modified CV model.In the part of feature extraction,eleven texture features of breast ultrasound image and seven morphological features are extracted combined with the clinical doctor for diagnosis of breast tumor ultrasound images commonly used parameters and breast imaging reporting and data system(BI-RADS)describe 7 aspects of mass.In this paper,a series of parameters extraction based on gray level co-occurrence matrix and some parameters based on Tamura texture feature are extracted.Morphological characteristics reflect the difference of benign and malignant tumor onshape.In this paper,the extractedmorphological characteristicsmainly include circularity,aspect ratio,compactness,and so on.The class distance is used to better identify the characteristics of the strong ability to distinguishthe benign and malignant tumors.