Study On Breast Mass Detection And Classification Method

Breast cancer incidence of a disease of whole body as a malignant tumor of7-10%of the risk of disease is increasingly attracted people’s attention, the best way of Prevention and treatment of the disease is to develop breast X radiography the census. Because our country population and medical personnel imbalance, make computer aided testing (CAD) has become an important research direction. Through computer aided physician X radiography diagnosis. CAD detection method for there are many shortcomings, such as miss rate is high, the feature extraction redundancy, computational complexity,and fail to invest in clinical use.This paper in order to improve the detection performance of the CAD based on single mass image, completed the following three aspects:(1)This paper puts forward three the discriminant conditions based on doctors diagnosis method used to remove the obvious false positive area. After extracting the suspect area, this paper uses the mass of round degree, mostly independent existence, mass and mammary edge’s distance is large these three based on prior experience discrimination standards. Through the screening test area can remove a lot of false positive regional. Test results show that the application of this assessment criterion, can be more than40%of the false positive regional removed. In reduce the computation time of also sharply reduce the area on subsequent classification influence.(2) Put forward a image segmentation method based on fuzzy sets. Previous regional segmentation algorithm can only get absolutely boundary, can’t get around mass divergent information. This paper introduced the concept of fuzzy sets into regional growth segmentation algorithm. Through the definition of membership functions giving each pixel a membership parameters to show the degree that point belongs to seeds point. Effective won the mass edge’s detailed information and the parameters shows the degree the mass extensive the area around. These parameters are benign and malignant mass in the judgment of the important basis.(3) Ensuring low in miss rate, reduced of the number of characteristic used for classification vector. After get suspicious area, this paper choose to use machine learning classification, machine learning method we choose the support vector machine (SVM) which have very good performance in limited samples. In this paper the choice of characteristic vector, mainly through three aspects of parameters to support the classification results, including mass brightness mean grey value and the surrounding area background region average gray which reflect mass was significantly higher than the surrounding tissue; mass core region circular degree and edge regions round degree which reflect the contour features; degree of membership distribution which reflect the surrounding area mass invasion. Ensuring low in miss rate, reduced the number of characteristic used for classification, to some extent, solve the problem of the parameter selection redundancy. Test results show that the classification achieved good results. The characteristic parameters this paper chooses can be very good become the basis of classification. The selection of characteristic vector thought also provides direction for later research. After the follow-up improvement, detection method is put forward in this paper will has the value of application in clinical. After test, each image has only1.1false positive point on an average, for good and malign bump recognition rate is above80%.This paper chooses the characteristic parameters can well become the basis of classification, to a certain extent, solved the parameter selection redundancy problem. In simplified calculation at the same time, this paper won the results which can provide to doctors as a reference.