Resumen de Modelo estadístico de textura para detección de masas en mamografía

Nicolás Gallego Ortiz, David Fernández Mc Cann

• In this article we demonstrate the use of principal components analysis to visualizea high dimensional texture representation and understand its geometry. Our targetapplication is mass detection as texture abnormalities. We analyse a texture represen-tation based on the gray levels of a local neighborhood, we estimate a gaussian mixturemodel by means of the EM algorithm and graph the data along with the probabilisticmodel. Our results explain nature of the feature space and at the same time keepsaround 98% of the total variation.Kewords: Biomedical Engineering, Radiodiagnostic, Breast-Cancer, Statistical Meth-ods, Mathematical Models.