This paper is about an evaluation for a feature selection strategy for mammogram classification. An earlier solution to this problem is revisited, which constructed a supervised classifier for two problems in mammogram classification: tumor nature, and tumor geometric type. The approach works by transforming the data of the images in a wavelet basis and by using a minimum subset of representative features of these textures based in a specific threshold (λT). In this paper different wavelet bases, variation of the selection strategy for the coefficients, and different metrics are all evaluated with known labelled images. This is a suitable solution worth further exploration. For the experiments we have used samples of images labeled by physicians. Results shown are promising, and we describe possible lines for future directions.
展开▼