A statistical model of texture for medical image synthesis and analysis
Christopher James Rose, Chris Taylor · 2003
Abstract. We address the problem of building generative statistical models of the appearance of highly variable medical images, in particular mammograms. We treat appearance as a texture that can vary over the image plane. We present a model motivated by one of the most successful algorithms in the texture synthesis literature. Our approach has significant advantages over existing methods: it can learn from very large data sets, does not need to assume spatial ergodicity and can be used for synthesis and analysis. We present early results in the form of synthetic images. 1