Abstract
In recent years, computer-aided diagnosis (CAD) has been expecting to improve the diagnostic performance and to reduce the oversight of a lesion. CAD is used for breast X-ray images. However, there is a challenge about the poor detection accuracy of architectural distortion that is the most missing findings. In this study, we propose the hybrid detection method that integrates two detection methods of massive patterns and architectural distortion to improve the lesion detection sensitivity. We applied to the adaptive Gabor filter for the detection of the architectural distortion and to the iris filter for the detection of the mass. Candidate regions are obtained by calculation of the concentration index using filtered images. In the experiments, we compared the results of proposed method and physician interpretation report using 200 images (63 architectural distortions) of digital database of screening mammography. As a result, the conventional method had true positive rate (TPR) of 73.0% and the number of false positives (FPs) of 0.65 per image. Whereas TPR of proposed method was 79.4% and FPs were 1.88 per image. These results indicate that our method may be effective to improve the performance of computer aided detection in breast X-ray images.
