Classifying mammograms using texture information

In an ongoing effort to assist radiologists in detecting breast cancer early, this paper focuses on breast characterisation according to internal tissue characteristics. This is an important feature because it has been demonstrated that women with dense breasts are more likely to suffer breast cancer, and also, the performance of automatic mass detection methods decreases in dense breasts. The strategy of our proposal firstly identifies regions with similar grey-level by using a clustering strategy. Subsequently, texture descriptors are extracted from each cluster by using Local Binary Patterns and Co-occurrence Matrices, and finally used to train a classifier. Results obtained from the complete MIAS database and using a leave-one-out strategy show a correct classification of 78% when compared to expert assessment.