DeepRecon: Joint 2D Cardiac Segmentation and 3D Volume Reconstruction via a Structure-Specific Generative Method

AbstractJoint 2D cardiac segmentation and 3D volume reconstruction are fundamental in building statistical anatomy models understanding functional mechanisms from motion patterns. However, due to the low through-plane resolution of cine MR high inter-subject variance, accurately segmenting images reconstructing challenging. In this study, we propose an end-to-end latent-space-based framework, DeepRecon, that generates multiple clinically essential outcomes, including accurate image segmentation, synthetic high-resolution image, reconstructed volume. Our method identifies optimal latent representation contains semantic information for structures. particular, our model jointly with structures using representation. We further explore downstream applications shape 4D pattern adaptation by different latent-space manipulation strategies. The simultaneously generated present a interpretable value assess motion. Experimental results demonstrate effectiveness approach on fronts reconstruction, adaption performance.Keywords3D reconstructionCardiac MRIGANLatent space