? A novel full optimization methodology is originally proposed and the further development and optimization of a new class of computer-aided diagnosis (CAD) methods is designed for mass detection in digital mammography in this Phase R21 proposal. The final optimized CAD system will be evaluated by computed FROC analysis as a retrospective case study, using a separate image data base including 1200 four view cases containing all mass types. This Phase R33 research project is to modify the optimized single-view CAD algorithms from Phase R21 for constructing an adaptive ipsilateral multi-view concurrent CAD system for improving the early detection of breast cancer by focusing on the computerized detection of tiny mass in digital mammography. The feasibility study of the full optimization new technology is achieved through the Phase R21 research, which explores all the potentials for the single-view full modular computer-aided diagnosis (CAD) algorithms for the automatic detection and diagnosis of masses from digitized screen/film mammography (SFM) and full field digital mammography (FFDM). As worldwide reported in recent literature, the CAD misses early stage breast cancer and results in a relatively large false-positive (FP) detection rate in order to achieve a high sensitivity rate. This project is inspired by the interpretation procedure from mammographers. We found that the abnormal diagnosis can be derived from multiple views but not available through single-view image analysis. To explore this important information will result in significant improvements on CAD performance. In consequence, this proposal aims at modification and optimization of the single-view CAD methodologies optimized in Phase R21 project for ipsilateral multi-view digital mammograms, which creates an entire new multi-view CAD scheme. The overall design procedure of proposed CAD system is as follows: each view image of ipsilateral breast including mediolateral oblique (MLO) view and craniocaudal (CC) view will be processed using advanced preprocessing and segmentation methods, concurrent analysis method will be employed in multi-view images to extract features from segmented suspicious regions and to analyze the feature matching between different views. The analysis result will be fed back to single-view image processing for their further analysis. Such iterative processing and analysis will be conducted between single-view and multi-view images to differentiate tiny suspicious regions for early stage breast cancer detection and reducing false-positive (FP) detection rate in order to achieve a high sensitivity rate. ? ?
