The Research and Implementation of Lgorithms on Microcalcification Detection
|Keywords||Computer Aided Diagnosis calcification detection Bayes discriminant function Mammogram registration|
As a powerful instrument, more and more doctors use the mammo X-Ray images to diagnose breast cancer, and Mammo Computer Aided Diagnosis can contribute to increase the sensitivity and specificity for diagnosis. So Mammo Computer Aided Diagnosis has been the hot topic in the recent research of early diagnosis of breast cancer.Calcification detection is the important part of MammoCAD, we present two detection methods by combining the shape and size of calcification .One is using wavelet transform and Hessian matrix; first, the mammography images were decomposed at scales from 1 to 3, and then the result images for the nodular components were enhanced by Hessian matrix; then we get the threshold value by using the histogram of local variance and mark the pixel whose value is higher than the threshold as calcification. The other method is involving the Gaussian function and Hessian matrix; first we calculate the second difference of Gaussian function and correlate the calculated value with original images, and then the correlation results were enhanced by Hessian matrix. In order to reduce the texture false position of calcification, on the basis of wavelet transform we added the linear components of Hessian enhancement, and employed the Bayes discriminant function to distinguish the texture and the calcification so as to eliminate the texture. Combining the Bayes classifier with two kinds fo detection methods, and then according to the evaluation gained from many hospitals,compared with the old method, both of the two methods accept higher sensitivity, The detection ratio of first method is increased by 12.3%, false position is decreased by11.5%, and the detection ratio of second method is increased by 14.9%, false position is decreased by 23.1%.By combining the second method with the Bayes classifier, the detection ratio is increased and the false position is decreased. According to the effective demand of the item, we add the method, involving the Gaussian function and Hessian matrix, and the Bayes classifier to the mammoCAD platform.