Four Chamber Whole Heart Segmentation and Reconstruction for evaluating healthy and non-healthy Heart state based on Deep Learning models

Cardiovascular magnetic resonance (CMR) has become a key imaging modality in clinical cardiology practice due to its unique capabilities for non-invasive imaging of the cardiac chambers and great vessels. A wide range of CMR sequences such as Cine CMR, flow CMR, tagged CMR, late gadolinium enhancement (LGE), and perfusion CMR have been developed to assess various aspects of cardiac structure and function, and significant advances have also been made in terms of imaging quality and acquisition times.

The 3D shape of the atria and ventricles is important for studying the mechanisms of disease processes. Left atrial volume is commonly estimated using the bi-plane area-length method from two-chamber (2CH) and four-chamber (4CH) long axes views. However, this can be inaccurate due to a violation of geometric assumptions. We aimed to develop a deep learning neural network to infer 3D left atrial shape, volume, and surface area from 2CH and 4CH views.

In this abstract, we have proposed Attention-guided Generative adversarial and efficient 3D volumetric probabilistic diffusion deep learning models for 4CH whole heart segmentation and reconstruction using private non-annotated clinical MRI and open source annotated MICCAI challenge datasets. The MICCAI dataset has abandoned annotation while our private clinical MRI dataset has no manual annotation. We first generated the target clinical MRI data from unpaired MICCAI challenges datasets using Attention-guided Generative adversarial network and transform the style or miss alignment pixels from unpaired source MICCAI annotated datasets to our private non-annotated dataset and then segment left ventricle (LV), right ventricle (RV), left arterial (RA), and right atrial (RA) using 3D volumetric probabilistic diffusion model. Further A 3D UNet was trained and tested using 2CH and 4CH segmentations generated from 3D coronary computed tomography angiography (CCTA) segmentations. The sparse input label map volume was converted to a dense label map by the label completion network, giving dense volumetric label maps of the LA, LV, left/right pulmonary veins. Our proposed model achieved better performance using the generated clinical MRI without annotated labels. This process aids in analyzing myocardium function and conducting biomechanical analyses from imaging data.