The atrial-model-generation tool is an almost automatic workflow for the generation of patient-specific volumetric atrial models. More...

Detailed Description

The atrial-model-generation tool is an almost automatic workflow for the generation of patient-specific volumetric atrial models.

It can be used to define single atrial chambers or bi-atrial geometries. It comprises six major sequential processing stages. First, starting from a tomographic scan (either CT or MRI), a convolutional neural network is employed to automatically segment the clinical image and label key anatomical structures, including heart blood pools and the left ventricular myocardium. The workflow then automatically augments the atrial blood pool with specific landmarks to identify the appendages, the ostia of the veins, and part of the vein tissue that will be removed in the next step to ensure the veins are open.

Based on these landmarks, an extrusion process is carried out to create the endocardial and epicardial layers, prescribing different wall thicknesses for the body of the atria and the veins. In this phase, the veins and valves are opened by avoiding extrusion at landmarks indicating removable tissue and at the interface between the atria and the ventricles. Next, a clustering algorithm is used to more precisely identify the orifices based on their position and distance from the appendages. Landmarks for the LSPV, LIPV, RSPV, RIPV, MV, IVC, SVC, CS, and TV are thus finally assigned to the corresponding openings.

Based on this information, a rule-based automatic labeling process is employed to define the major anatomical structures of the atria, including the CT, PM, ring of the FO, FO itself, and the lateral and (optionally) the anterior bands of the BB. These landmarks are finally used to define rule-based fiber bundles. As a last step, the workflow allows for the computation of the universal atrial coordinate system.