Towards Patient-Specific Heart Models for Decision Support System

2. Perspective: From 1-D clinical parameters to 3-D anatomical model of the heart

2.1 Introduction

The second approach that can be used to build a patient-specific heart model is to use 1-D measurements acquired during clinical evaluation instead of 3-D images. This idea comes from the observation that advanced imaging hardware is not systematically available to the clinicians and diagnostic is mostly based on 2-D ultrasonography data. In this way, we propose to integrate into the disease-based heart model 1-D clinical parameters related to the patient's physiology. Simulations would be personalised and hidden parameters that apply to the patient might be extracted without the need of 3-D imaging. Moreover, the generated 3-D patient-specific heart models could be used to improve the accuracy of some measurements, like the volume of the cavities for instance.

As a first step, we rely on geometrical parameters only to define patient-specific heart models. Though mono-dimensional and sparse, these measurements are of great importance to the clinicians since they allow the assessment of the heart anatomy, the efficiency of the myocardium, the cardiac function and, above all, the diagnostic of cardiac pathologies. Moreover, their acquisition is systematic and performed through standardised methods.

Next figure illustrates various measurements that are acquired during the diagnostic process. Size of the ventricles, internal lengths and diameters of the cavity, diameter of the valves and the thickness of the myocardium are the parameters that we plan to use to build the patient-specific heart model.

Clinical measurements used to describe the right ventricle (image from Gaslini).
The white arrows are diameters measured from 2-D or 3-D images.

2.2 Method

Overall considerations

The overall algorithm used to generate the patient-specific anatomical model is made up of three main steps:

acquisition and input of geometrical parameters,
building of the skeleton of the synthetic heart according to these measurements,
fitting of deformable surfaces to the skeleton to create the synthetic organ.

The result can then be converted into a mesh and imported, in a near future, into our cardiac simulator.

To create the 3-D representation of the synthetic organ we make use of the NURBS framework. NURBS, for Non-Uniform Rational B-Spline, is a mathematical model that allows the intuitive generation of sophisticated geometries. Derived from the Bezier's methodology and the B-spline theory, NURBS represents any curves (or surfaces) by:

a set of control points,
the related "knots" that specify the way the control points modify the curves (or the surface),
one (for the curves) or two (for the surfaces) integers that define the order of the NURBS.

However, in most cases, only the control points and the order of the NURBS are available to the user. The procedure to create the desired shape consists then in modifying the number and the position of the control points of the NURBS (see next figure).

Another interesting property of NURBS is the possibility to assign to each control point a weight, which offers more freedom during the generation process.

NURBS surface created using Blender. Left panel: wireframe visualisation of the NURBS surface (in white). The yellow points are the control points with which the surface is modelled. Right panel: solid visualisation of the surface.

Owing to their powerful features, NURBS-based tools can be found in any 3-D modelling software. They are widely used in computer graphics to design sophisticated shapes and are implemented in the most common 3-D modelling suites (3ds Max, Blender...). Advanced techniques based on NURBS have been developed to assist the user. The one which most interests us is the "skinning technique". It consists in modelling the desired shape by defining its skeleton, the software fits automatically NURBS surfaces to this structure to get the final object.

NURBS-based tools are thus suitable for our aims. Blender, a free but powerful 3-D modelling software, have been chosen to generate the patient-specific heart anatomy. Supported by an important community of developers, it implements the essential 3-D modelling tools and offers a Python API for the development of advanced scripts.

A script with a graphical user interface (GUI) has been developed to allow the clinicians to enter the measured parameters and to create automatically the synthetic anatomical heart. In the following the three main steps of the algorithm are detailed.

Step 1- Input of the clinical data

The first stage of the algorithm consists in providing the clinical measurements. No assumption about the acquisition method has been done and therefore any imaging technique can be utilised, from 2-D ultrasonography to cardiac MRI. The parameters used to generate the synthetic heart have been selected by consensus between the clinicians and the engineers of the Health-e-Child project. The way they are acquired is presented in the diagnostic protocol of the project (Health-e-Child D9.1, 2006) and in the ASE committee recommendations (Lang et al., 2005). Finally, their default values have been set according to the normal values available in the literature (Lang et al., 2005; Ho and Nihoyannopoulos, 2006). The following table lists the parameters required to create the patient-specific heart model along with their default values.

Left ventricle		Right ventricle
Parameters	Default value	Parameters	Default value
Internal length	9.0 cm	Internal length	8.0 cm
Internal diameter	5.0 cm	Internal diameter, major axis	5.0 cm
		Internal diameter, minor axis	3.0 cm
Diameter of the mitral valve	2.8 cm	Diameter of the tricuspid valve, major axis	3.5 cm
		Diameter of the tricuspid valve, minor axis	2.5 cm
Diameter of the aorta valve	1.3 cm	Diameter of the pulmonary valve	1.3 cm
		Diameter of the outflow tract	2.0 cm
		Length of the outflow tract	2.0 cm

Interface used to create patient-specific heart models.

Step 2- Generation of the skeleton

Once the script is provided with the clinical parameters, it creates the skeleton of each ventricle. Because the ventricular geometry cannot be modelled by a single NURBS surface, each ventricle has been divided into two different parts. The first one consists in the "inflow" section of the ventricle, from the atrio-ventricular valve to the apex. The second part corresponds to the "outflow" section, from the arterial valve to the apex. However, this sub-division is transparent to the user, he only has to click on one single button to create the skeleton of a ventricle, Draw LV for the left ventricle and Draw RV for the right ventricle.

Skeleton of the right ventricle generated automatically from clinical parameters

Step 3- NURBS fitting

Next, two NURBS surfaces are fitted to the two parts of the skeleton by means of an automated tool that comes up with Blender. Various parameters can be modified to adjust the fitting process but in general their default values offer satisfying results.

NURBS surfaces fitted to the skeleton of the right ventricle

Step 4- Mesh creation

Finally, the two NURBS surfaces are converted into meshes and merged by using an automated Boolean union. Only the external envelop is kept and all the "internal" and artificial walls are removed to get an empty cavity. At last, the resulting mesh can be smoothed to improve the result.

INTERACTIVE 3D MESH

Right ventricle automatically created from clinical parameters.
Left panel: 2-D view of the 3-D mesh.
Right panel: interactive 3-D mesh (use the mouse to interact with the 3-D model)

2.3. Conclusion

Relying on the observation that 3-D imaging is not systematically available, a preliminary algorithm that automatically creates 3-D heart geometries from well-known and standardised 1-D clinical parameters has been presented. The underlying principle are promising and clinicians involved in the Health-e-Child project have shown a great interest in the method.

The exploratory algorithm developed so far relies on geometrical parameters only and on the NURBS framework to model the cavities. Blender software has been used for the implementation and a graphical user interface has been developed.

Though NURBS-based tools for the modelling of the heart anatomy are available in the literature, the overall majority, to our knowledge, depends on the manual definition of landmarks in the image to build the NURBS surfaces and generate the synthetic heart model (Segars et al., 1999). Our approach is thus distinct from these methods since no landmarks are used and the shape is controlled by clinical and meaningful measurements.

2.4. References

Health-e-Child Integrated Project, 2006. Report on diagnostic coding system and data collection protocols.
Ho, S. Y., Nihoyannopoulos, P., 2006. Anatomy, echocardiography, and normal right ventricular dimensions. Heart, Vol. 92, pp. 2-13.
Lang, R. M., Bierig, M., Devereux, R. B., Flachskampf, F., A., Foster, E., Pellikka, P., A., Picard, M., H., Roman, M., J., Seward, J., Shanewise, J., S., Solomon, S., D., Spencer, K., T., Sutton, M., St J., Stewart, W., J., 2005, Recommendations for Chamber Quantification: A Report fromt the American Society of Echovardiography's Guidelings end Standards Committee and the Chamber Quantification Writing Group, Developed in Conjunction with the European Association of Echocardiography, a Brandch of the European Society of Cardiology. J. Am. Soc. Echocardiogr, no 18, pp. 1440-1463
Segars, W. P., Lalush, D. S., Tsui, B. M. W, 1999. A realistic spline-based dynamic heart phantom. Nuclear Science, IEEE Transactions on, Vol. 46, No. 3., pp. 503-506.