Losel P., Heuveline V.
in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 10129 LNCS (2017) 121-128. DOI:10.1007/978-3-319-52280-7_12
© Springer International Publishing AG 2017. Segmenting the blood pool and myocardium from a 3D cardiovascular magnetic resonance (CMR) image allows to create a patient-specific heart model for surgical planning in children with complex congenital heart disease (CHD). Implementation of semi-automatic or automatic segmentation algorithms is challenging because of a high anatomical variability of the heart defects, low contrast, and intensity variations in the images. Therefore, manual segmentation is the gold standard but it is labor-intensive. In this paper we report the set-up and results of a highly scalable semi-automatic diffusion algorithm for image segmentation. The method extrapolates the information from a small number of expert manually labeled reference slices to the remaining volume. While results of most semi-automatic algorithms strongly depend on well-chosen but usually unknown parameters this approach is parameter-free. Validation is performed on twenty 3D CMR images.