4D Cardiac Shape Reconstruction Using Image-Aided Neural Signed Distance Correction Fields

Motivation: Accurate 4D cardiac shape reconstruction is crucial for myocardial disease diagnosis, yet existing deep learning methods fail to fully utilize cardiac magnetic resonance data, resulting in sub-optimal performance and efficiency. Goal(s): Our goal was to develop a deep learning-based 4D cardiac shape reconstruction method using cardiac magnetic resonance and focusing on the left ventricular myocardium to achieve better shape precision and efficiency. Approach: We leveraged imaging data as guidance to model the 4D cardiac shapes as neural signed distance correction fields. Results: Our method achieved state-of-the-art performance in 4D cardiac shape reconstruction with faster inference on new patients. Impact: The 4D cardiac shape sequence of a new patient can be obtained using our method with less time required and the best shape precision among the deep learning-based methods so far.