Research Of Mammographic Breast Mass Computer Aided Detection System

Breast cancer is a mortal health threat to women. Early detection and treatment is the only effective way to save lives. Mammography is the most widely used breast imaging method currently. However, the high subtlety of breast cancer lesions and the overwhelming amount of image data generated from the routine detection can cause fatigue of radiologists and overlook of lesions. Thanks to the rapid development of artificial intelligence and computer vision, computer aided detection has been gradually introduced into medical image processing, assisting the radiologists to improve detection accuracy and sensitivity.Computer aided detection(CAD) has seen a serious of successes in the past, and several commercial systems have been launched into market. Clinical practice has proven the effectiveness of CAD in improving the detection performance and saving lives. However, there is still space for improvement, especially in breast mass detection. The false positive rate is relatively high for the present breast cancer CAD system. Therefore, the most important technique problem for breast cancer CAD is to lower the FP rate while still maintaining a high sensitivity.This thesis proposed a breast mass CAD system which can achieve a high sensitivity and a low FP rate. We first developed a novel segmentation method by combining a dual morphological enhancing method and superpixel SLIC method to separate the masses from the surrounding tissue. Then a rule-based selection was applied to choose the regions of interest(ROI). After the ROIs were refined by a re-initialization free level set method-DRLSE, an ensemble of under-sampled SVMs(EUS SVMs) scheme was applied to label a ROI as mass or non-mass. Evaluated on the DDSM dataset, our approach achieved a case-based sensitivity higher than 90% with an FP rate lower than 1.1 marks/image.In conclusion, we proposed a novel mammographic breast mass CAD scheme which can achieve a high sensitivity and a relatively low FP rate, providing inspiring ideas for the development of breast cancer CAD technology.