Multi-view-based Computer-aided Detection Of Mammographic Masses

Early detection and diagnosis of breast cancer is specially a pivotal step of mammary-therapy. Computer-aided detection and diagnosis techniques could offer a cost-effective alternative to double reading as a means of reducing errors. Most of the development of CAD systems has been based on the analysis of single views. The false-positive detection rate of single-view-based CAD is somewhat high. On the basis of single-view-based CAD results, research multi-view-based CAD and find the same mass area from two ipso-lateral view to deduce false-positive detection rate and buildup the practicability of multi-view-based CAD system.The performance of multi-view-based CAD depends on single-view-based CAD, and mass segmentation is an important factor. Maximum entropy threshold based method automatically produces initial contour, then adopt GVF snake segmentation method. The results yield a high nicety of mass segmentation.Implement single-view-based CAD to obtain suspicious mass area for every image. For every suspicious mass area, define a search area (match strip). Make links between the mass area and the areas in the match strip and extract multi-view features for every link. On the basis of the identifiers of doctors establish links and bring classifier. In the next step, every multi-view feature vector are inputted a classifier and decide the region pair as a TP-TP pair or a not TP-TP pair. The final step is to regroup suspicious region pairs for deciding positive mass region.The dataset comes from Digital Database for Screening Mammography of South Florida University. We choose 210 cases, one breast for every case, including 420 images, each image contains a positive mass. When the multi-view based CAD yield a sensitivity of 92%, its false positive reduces from 2.83 per image to 1.51 per image .