Detectability of patches in fractal textures for assessing Hölder exponent-based breast cancer risk evaluation

Early detection of breast cancer is key to patient’s survival. Recent works showed that the distribution of local Hölder exponents in a mammogram can quantify breast tissue disruption, and hence assess breast cancer risk. This work proposes a systematic study of the detectability of disrupted tissues embedded inside either fatty or dense tissues leveraging simulated piecewise homogeneous fractal textures modeling the breast tissues. A novel filtered fractional Brownian field model for stationary isotropic fractal textures is proposed, based on a genuinely designed isotropic filtering. Intensive simulations on synthetic textures generated either from the previously introduced fractional Gaussian field or from the novel filtered fractional Brownian field show that a state-of-the-art local Hölder exponent-based segmentation algorithm is capable of detecting large patches of disrupted tissues in fatty environments, but that segmentation accuracy drops down for small patches, while for dense environments performance are good and decrease slowly with the patch size.