This paper presents a new method for classifying tissues in ultrasound imagery at the pixel level. The proposed method is basically composed of 1) image normalization, 2) multiscale description, and 3) classification. The main advantage of the proposed method is its capacity to consider pixel-neighboring patterns, which are not encoded by existing methods based on isolated pixel values.
It was experimentally evaluated in tissue and carotid images. A comparison with the state-of-the-art techniques for pixel-level classification shows the quality of the proposed method and its advantages. A prototype implementing the proposed method is currently in use in the University of Campinas Hospital, Brazil.
D. V. Pazinato, B. V. Stein, W. R. de Almeida, R. O. Werneck, P. R. Mendes Júnior, O. A. B. Penatti, R. S. Torres, F. H. Menezes, A. Rocha. “Pixel-Level Tissue Classification for Ultrasound Images,” in IEEE Journal of Biomedical and Health Informatics, vol. 20, no. 1, pp. 256-267, Jan. 2016. doi: 10.1109/JBHI.2014.2386796