Optimization of CAD System Using Adaptive Simulated Annealing for Digital Mammography

摘要

Mammography is the most effective method to date and is becoming a high volume X-ray procedure for screening and diagnosing breast cancer. The performance of computer-aided detection and diagnosis (CAD) scheme determines its clinical effectiveness as an objective 'second reader' in aiding radiologists mammogram interpretation. Following research work of initial grant year, the major research works in the second grant year are: (1) to construct CAD system robust to FFDM and SFM, (2) to fully optimize the CAD system for its overall performance improvements in both sensitivity and specificity. The major accomplishments in the second grant year are as follows: (1) New modules have been developed, including preprocessing for normalization of mammographic images from PFDM and SFM, adaptive Fuzzy-C means algorithm for segmentation, support vector machine (SVM) technique for classification. Adaptive modules have been modified based on existing modules. (2) Adaptive CAD system has been constructed using developed and modified modules. (3) Fully optimization of CAD system by simulated annealing (SA) algorithm has been developed and performed. Key parameters affecting performance of CAD system have been selected as optimization variables. Modular and full system optimizations have been performed, respectively, on CAD system.