Mammograms Recognition Of Tumors Based On Multiple Feature

At present,the prevention and treatment of breast malignant tumors is very important,so the early screening of breast malignant tumors by mammograms has important research significance.Masses and calcification points are important diagnostic criteria for breast tumors.In this paper,we extract the features of tumors and calcification points in mammograms,and recognize the benign and malignant tumors based on the extracted features,which can be used to assist in screening breast malignant tumors.The breast tissues were segmented based on spectral clustering algorithm.The method of maximum grayscale difference constraint between the neighbor classes constraint and the shape constraint were proposed to accurately segment the breast muscles,which avoided the error segmentation caused by the high brightness pectoral muscle whose gray characteristics is similar to that of the mass.In the mammograms which pectoral muscles were segmented,the segmentation of the masses was realized based on the features of high grayscale and small area of the masses.The double windows filter method was used to enhance the micro calcification points,and the contribution matrix method based on the window was used to extract the micro calcification points.The texture features of masses were respectively described based on texture edge direction auto-correlation vectors,gray level co-occurrence matrix and gray level run-length matrix,respectively.The gray features of the masses were described by the gray standard deviation,the average grayscale deviation of local maximum value and the grayscale contrast with adjacent tissues.The morphological features of masses were described by edge complexity,radial length ratio and shape parameter ratio.The number of micro calcification points,the number of micro calcification points in local area and the grayscale contrast with neighbor area were used to describe the features of the micro calcification points.The support vector machine method was used to recognize benign and malignant mammograms,which based on the masses and micro calcification features with the MIAS database images and the mammograms of patients acquired by the hospital.The experimental results showed that the proposed method can segment pectoral muscles and masses accurately.The proposed micro calcification points extraction algorithm could reduce noise points while ensuring the accuracy of micro calcification points extraction.The accuracy rates of micro calcification points and masses recognition were 98.00%and 84.37%respectively,and.The total accuracy rate reached 90%.