Improving U-Net based segmentation in cardiac parametrical T1 mapping by incorporating bounding box information
Darian Viezzer1,2, Thomas Hadler1,2, Edyta Blaszczyk1,2, Maximilian Fenski1,3, Jan Gröschel1,2, Steffen Lange4, and Jeanette Schulz-Menger1,2,3
1Charité Universitätsmedizin Berlin, Working Group on Cardiovascular Magnetic Resonance, Experimental and Clinical Research Center, a joint cooperation between the Charité Universitätsmedizin Berlin and the Max-Delbrück-Center for Molecular Medicine, Berlin, Germany, 2DZHK (German Centre for Cardiovascular Research), partner site Berlin, Berlin, Germany, 3Department of Cardiology and Nephrology, Helios Hospital Berlin-Buch, Berlin, Germany, 4Faculty for Computer Sciences, Hochschule Darmstadt (University of Applied Sciences), Darmstadt, Germany
Parametric mapping images contain to a large extent irrelevant background information. In order to hide some of it, we incorporated bounding box information into a U-net based segmentation network. Our dataset consisted of 845 training, 102 validation and 146 test T1 maps of native and post-contrast myocardium from different clinical studies, including healthy volunteers and patients with inflammatory heart disease, muscular dystrophies or chronic myocardial infarction. While cropping the image input improved the segmentation itself, a second input of the bounding box mask reduced the mean absolute and mean squared T1 deviation, which is clinically preferred.
This abstract and the presentation materials are available to members only;
a login is required.