Study On Computer-aided Detection Method Of Masses Based On Multiple Mammograms

ABSTRACT:Breast cancer is one of the most common malignant tumors, which impacts physical and mental health of contemporary women. Now it has become the leading cancer killer of 35-60 urban women. Early detection, early diagnosis and early therapy play a decisive role in reducing the mortailty rate of breast cancer. For that the diagnostician usually affected by experience, continuous working hours and other subjective factors, breast X-ray diagnosis accuracy has been difficult to be guaranteed until the emergence of computer-aided detection technology.However, the computer-aided detection methods widely used currently all have high false positive rate and low correct identification rate of mass detection, whose fundamental reason is the effectiveness and finiteness of information in single mammogram. Computer-aided detection method based on information fusion of multiple mammograms is a very good solution to this problem and becoming an international study focus of computer-aided detection of breast cancer gradually. In order to improve the level of breast mass detection, information fusion of multiple mammograms method is used in the paper to achieve computer-aided detection based on cranio-caudal (CC) view mammogram and ipsilateral medio-lateral oblique (MLO) view mammogram.In this paper, we mainly studied on breast mass segmentation and recognition two aspects and achieved the following results:1. "Segmentation of Breast Masses Based on Improved Stratified Detection" algorithm is proposed in this paper. According to layer-by-layer detection method proposed by Wangying which demanding high image clarity and obtaining many false positive regions, improvement plan and more accurate layering model are presented. The experimental results shows that through image preprocessing removing high gray interference regions, elimating false positive regions of segmentation result in each layer and using the improved hierarchical model, the masses obtained by this method have low false-positive rate whose margins marked accurately and the detection rate of effective masses reached 98.1%.2. A mass recognition method based on multiple mammograms is implemented. Firstly, according to the proposed nipple positioning method based on breast outline and pectoral muscle, selecting nipple proximity position and pectoral muscle as reference positions, matching cranio-caudal view and ipsilateral medio-lateral oblique view and segmented regions in two views to achieve the information fusion. Secondly, according to the characteristics of breast masses, extract the information of matched segmented regions. Finally, use BP neural network to realize the classification of masses and normal tissue. Practice shows that using same design method of BP neural network and image library, the correct rate of mass recognition based on single mammogram is 84.2% and our detection method based on multiple mammograms is 88.7%.Comparing with the computer-aided detection method based on signle mammogram, the known information of method based on information fusion of multiple mammograms increased, the reliability and validity of information improved the correct rate of mass segmentation and recognition also be improved. The research result of this paper gives a good solution to the insufficient information problem in computer-aided detection systems widely used currently based on single mammogram and provides a new solution to the improvement of breast cancer computer-aided detection system.