Biventricular statistical shape model of the human heart adapted for computer simulations

This is an adapted version of the biventricular statistical shape model from Bai et al. (2015) that can be used as a basis for computer simulations of cardiac electrophysiology or cardiac mechanics. The original model consists of disconnected surfaces of the left ventricular (LV) myocardium and the right ventricular (RV) blood pool. The surface of the RV blood pool was clipped at the base to add an orifice representing the RV in- and outlets. The resulting RV endocardial surface was shifted along its normals by a fixed wall thickness of 3 mm to obtain an RV epicardial surface. Then all surfaces were merged to form one closed surface of the biventricular myocardium. This surface was remeshed using Instant Meshes (Jakob et al., 2015) and tetrahedralized using Gmsh (Geuzaine. et al., 2009), resulting in the following two meshes of the mean shape:

heart_sur.vtp:
Triangle mesh with 100 k nodes, 200 k elements, mean edge length: 0.83 mm.
This mesh contains a point data array 'class' labeling the following regions: base (1), epicardium (2), LV endocardium (3), RV endocardium (4), apex (5).

heart_vol.vtu:
Tetrahedral mesh with 479 k nodes, 2555 k elements, mean edge length: 0.82 mm.

The principal components and variances of the 100 shape modes were Laplacian interpolated to the nodes of these new meshes and are stored in the following files:

pc_sur.csv.gz or pc_sur.mat: 100 principal components corresponding to the nodes of heart_sur.vtp.

pc_vol.csv.gz or pc_vol.mat: 100 principal components corresponding to the nodes of heart_vol.vtp.

var.csv.gz or var.mat: 100 variances applicable to both meshes.

100 quasi-random instances derived from the model are also available. They were created by drawing the weights of the 100 principal components from a uniform distribution in the range of −3 to +3 standard deviations:

instances_XXX_to_YYY.tar.gz: Archives with 10 out of 100 instances. For each instance, there is a vtp file containing surface classes and a vtu file containing Consistent biventricular coordinates.

instances_weights.csv.gz or instances_weights.mat: Weights in the range [−3, +3] corresponding to the 100 instances. They were created using a scrambled Halton sequence.

Please be aware that the model does not include any variations in RV wall thickness.