In cardiac imaging, a high-resolution geometric representation of the heart is desired for accurate assessment of its anatomical structure and function. This is not easily available due to the limit of acquisition duration and respiratory/cardiac motion in clinical practice. Stacks of multi-slice 2D images are usually acquired in clinical routine and segmentation of these images provides a low-resolution representation of cardiac anatomy, which may contain artefacts caused by motion. Here we propose a novel latent optimisation framework that jointly performs motion correction and super resolution for cardiac image segmentations based on generative learning. Moreover, quality control of the automatic segmentation results is realised via the proposed framework.
Meeting ID: 925 7540 3744