Fast detection of convergence areas in Digital Breast Tomosynthesis

In this paper we propose a fast method to detect spiculated lesions and architectural distortions in digital breast tomosynthesis datasets. This approach relies on an a contrario modeling of the problem. First, an indicator corresponding to the convergence of structures is defined, then the a contrario framework is used to set a threshold on it in order to detect zones where its value is unlikely. We propose, as a main contribution of this paper, a fast algorithm to implement this method, which significantly reduces its computational cost. The method was evaluated on 38 breasts (10 containing a lesion), and a sensitivity of 0.8 at 1.68 false positive per breast was obtained.