Synthesising 3D Cardiac CINE-MR Images and Corresponding Segmentation Masks using a Latent Diffusion Model

Nina Cheng, Zhengji Liu, Yash Deo, Haoran Dou, Ning Bi, Kun Wu, Fengming Lin, Zeike A. Taylor, Nishant Ravikumar, Alejandro F. Frangi

Abstract: We propose a novel pipeline for the generation of synthetic full spatial cine cardiac magnetic resonance (CMR) images via a latent Denoising Diffusion Implicit Models (DDIMs). These synthetic images can be used as viable alternatives to real data in deep learning model training for downstream cardiac image analysis tasks such as cardiac segmentation. To demonstrate the effectiveness of this approach, we generated synthetic CMR images along with their corresponding segmentation masks. We evaluated model performance using a variety of methods, including generated image fidelity, diversity and calculated the volumes of the generated segmentation masks and compare it with the real segmentation masks. The proposed pipeline has the potential to be widely applied to other tasks in various medical imaging modalities. Effective and efficient generation of 3D cine cardiac images with corresponding segmentation masks can supplement real patient datasets and help reduce the burden of manually annotating images.

External IDs:dblp:conf/isbi/ChengLDDBWLTRF24