3D Right Ventricle Reconstruction from 2D U-Net Segmentation of Sparse Short-Axis and 4-Chamber Cardiac Cine MRI Views

Tautz, LennartLennartTautzWalczak, LarsLarsWalczakManini, ChiaraChiaraManiniHennemuth, AnjaAnjaHennemuthHüllebrand, MarkusMarkusHüllebrand2022-09-152022-09-152022https://publica.fraunhofer.de/handle/publica/42551810.1007/978-3-030-93722-5_382-s2.0-85124019561We reconstruct a 3D model of the right ventricle from short- and long-axis image data and evaluate the benefits compared to quantification based on the 2D image stack. Deep learning is used to extract short-axis contours. An initial surface representation based on the contours is refined using long-axis images. Using a deformable model, the surface around the basal plane is adapted to image data. The resulting models capture the shape of the right ventricle better than segmentation from short-axis images alone and allow for a more precise volumetry.enDeep learningDeformable modelright ventriclesegmentation3D Right Ventricle Reconstruction from 2D U-Net Segmentation of Sparse Short-Axis and 4-Chamber Cardiac Cine MRI Viewsconference paper