Image retrieval scheme for mammographic masses by using a local-pattern matching technique

We proposed a concept of content-based image retrieval and demonstrated the potential usefulness in mammography. The approach incorporated a local-pattern matching method based on Nth-order autocorrelation features with KL expansion (principal components analysis) to retrieve similar mass shadows on digitized mammograms. The method can perform image retrieval without carrying out the image segmentation. We confirmed the tendency that similar mass images were retrieved as the initial studies by using the 30 simulated patterns and the 75 images of mammographic masses. The result showed that the image retrieval method might provide a new CAD system on mammograms.