Registre d'imatges mèdicas d'ultrasò
García Corisco, Elisa · UPCommons institutional repository (Universitat Politècnica de Catalunya) · 2014
[ANGLÈS] This thesis propose a processing algorithm to elastically register two ultrasound prostate images (before and after deformation) consists of three steps: (a) preprocessing, (b) global rigid registration and (c) elasticity modeling by registering image sub-patches. Preprocessing involves image filtering, down-sampling and feature extraction, here a Fisher-Tippet filter. Areas with high corner density, obtained with SUSAN corner detector, are regarded to contain significant information. Global registration is utilizing normalized mutual information (NMI). For the following elastic part, the globally registered images are sub-divided into dynamically sized patches; those showing a significant corner density are selected and matched between pre- and post-deformation images by normalized cross correlation (NCC). Elasticity is modeled as deformation field by smoothly interpolating between the corresponding partial patch registration outcomes.