Geometric Model Reconstruction - Computational Hemodynamics: Theory, Modelling and Applications - page 62

Vingmed, Horten, Norway). The volume was acquired during apnea over 4 heart
cycles, from the apical window, in harmonic mode, one QRS triggered sub-volume
acquired per heart cycle. The frame rate was 27 per cycle.
The endocardial border was generated using the AutoLVQ tool (Hansegard et al.
2009), EchoPAC workstation (version BT 11), GE Vingmed Ultrasound, Hortem,
Norway. AutoLVQ represents the left ventricle boundary as a deformable model and
relies on 3D energy minimization for evolving it. A combination of internal, external
and temporal forces ensures shape continuity, while adapting the model to a particu-
lar 3D echo recording. The endocardial contour process was initialized by manual
positioning of the apex and the mitral valve attachment points in a long-axis view
(e.g. four chamber), both at end-diastole and end-systole. After manual selection,
the endocardial border is automatically generated throughout the cardiac cycle. The
proposed contour was then evaluated in both short and long-axis cut-planes of the 3D
volume. If deemed necessary the border can be further refined by adding additional
attractor points that pull the model towards the endocardium. In this case, the border
was adjusted by placing a limited number of attractors. The papillary muscles and
major trabeculae were included in the left ventricle cavity both in diastole and systole.
The segmented endocardial left ventricle wall derived from RT3DE consisted of
closed three dimensional surface meshes for 27 different time steps. Together they de-
scribe the inner ventricular wall movement. The algorithms were constructed in such
a way that user intervention was necessary only in the first frame, i.e. start of systole.
To create our computational transient 3D model we have to include a model of the
mitral valve and the left ventricular outflow tract into the segmented left ventricular
surface mesh for all time frames. The mitral valve and left ventricle outflow tract
model was reconstructed from the same RT3DE data as the ventricle. Figure 3.17
shows the geometry of the complete 3D model at start- and end-systole.
To illustrate the correlation between the model and the echocardiographic record-
ings, the model was realigned with the original RT3DE data as shown in Fig. 3.18 .
The left ventricle was placed back at the same position as when it was segmented in
AutoLVQ software (Fig. 3.19 ).

3.5 Summary

Reconstruction of the cardiovascular anatomy begins with obtaining the relevant
data from medical imaging. While a number of imaging modalities exist this chap-
ter presented CT and MRI based modes. Regardless technique, the scanned images
typically form 2D contiguous slices that are separated by a known interval distance.
The reconstruction of this data from 2D to 3D requires extraction of the region of
interest in what is known as the segmentation process. A number of algorithms
have been developed by researchers and biomedical scientists which can be ac-
cessed through either open-source/free software or commercial software. Access