In this paper, we focus on the analysis of image textures modeled by anisotropic fractional Brownian fields. We propose some anisotropy indices intended to characterize the anisotropy of these textures. The construction of these indices relies upon the use of multi-oriented quadratic variations. It is based on directional quantities appearing in their asymptotic expectation. We show that anisotropy indices are invariant to some image transforms, and estimable from observed data. Eventually, we use anisotropy indices in combination with a measure of texture roughness to detect lesions in mammograms.