Seed selection criteria for breast lesion segmentation in Ultra-Sound images
Joan Massich, Fabrice Mériaudeau, P Elsa, Robert J. Mart, J. A. Mart, J. Trueta · 2011
Abstract. Purpose: Segmentation plays a central role in medical imaging, though is not a trivial task to perform in some screening modal-ities such as Ultra-Sound images. This paper addresses the role of auto-matic seed placement when segmenting breast lesions in B-mode Ultra-Sound images, and proposes a new algorithm to automatically locate seed regions for further region growing expansion. Methods: In this work some state-of-the-art methodologies for seed placement are reviewed and a new method basing its region selection on assigning a probability of belonging to a lesion for every pixel depending on intensity, texture and geometrical constraints of the pixel is proposed. Results: The proposed algorithm has been evaluated using a set of sonographic breast images with accompanying expert-provided ground truth, and successfully compared to other existing algorithms. Conclusions: The experimental results show the performance and ro-bustness of the method when placing seed regions in noisy environments.