A local Rayleigh model with spatial scale selection for ultrasound image segmentation

Abstract

Ultrasound images are very noisy, with poor contrast and the attenuation of the acoustic wave in the depth of the observed medium leads to strong inhomogeneities in the image. Segmentation methods using global image statistics give unsatisfactory results. The use of local image statistics can solve effectively the problem of attenuation. The contribution of this paper is two folds. First, we propose the study of the adaptation of the global model proposed by Sarti et al. [28]. We kept the variational framework and the Rayleigh model of the observed image statistics. Second, we propose an interesting and generic adaptive scale selection algorithm based on the Intersection of Confidence Interval rule. The latter is also applied to the local Gaussian segmentation model of Brox and Cremers. Results on realistic simulations of ultrasound images show the robustness and the superiority of the local Rayleigh model. The efficiency and the genericity of the proposed scale selection strategy is also demonstrated.