Study On Lesion Detection Based On Mammograms

In the modern society, incidence of breast cancer is increasing yearly because of many factors such as circumstance, pressure of survival, structure of diet and so on, which heavily threats women's health. It is the key to decrease the rate of death caused by breast cancer whether it could be early found and cured. The molybdenum target X-Ray imaging is the primary method to diagnose breast cancer. However, inferior image quality, benign representation of malignant pathological changes and vision fatigue or negligence of observers can easily cause false judgment and leakage judgment. Therefore, many organizations and institutions have been studying and developing CAD (Computer Aided Diagnosis) systems in order to help doctors to diagnose.Using an actual project as background, the author introduces mammoCAD, deeply studies detection algorithms of mass and calcification which are the key problems of mammoCAD. In terms of mass detection, the paper detects mass by having the image multithreshold delaminating. This method can detect the masses with low intensities, which enhances the sensitivity of the original mass detection algorithm. In terms of calcification detection, the paper puts forward two algorithms for comparison. The first one introduces singular point detection theory in binary wavelet into calcification detection to get the exact margins of calcification points, and then sets thresholds for the margin image to detect calcification. The second one applies the theory of Hessian matrix and wavelet onto calcification detection, firstly uses Hessian matrix onto the image for multiscale enhancement, and then calculates the local variation of the enhanced image to detect calcification. The latter one combines local high-intensity information of calcification points with their shape characteristic as rotundity, simultaneously considers the scope of calcification points'size, so that gets a better detection result. Compared with the prior method, it is more practical.The algorithms in this paper have been adopted in the actual project, and work well.