home 2002-04-18

Lasse: a labeled subdivision surface editor

Lasse is a tool for editing and interactive fitting of labeled subdivision surfaces, implemented using Open Inventor. It was developed to support research in echocardiography, that is, developing 3d models of heart anatomy based on ultrasound images. The screen shots below show an experimental model of the left ventricle of the human heart, fit to image data from a single patient. Click on the images to see them fullsize, then use your browser's back command to return to this page.

Dragging a control vertex Lasse allows interactive changes to geometry, by dragging any collection of control mesh vertices, edges, and faces. The image at right shows a labeled piecewise smooth subdivision surface model of the left ventricle at end systole. Distinguishable anatomic regions are marked in color. For example, the general epicardial surface is brown, the epicardial apex is orange, and the endocardium is green.

The data to which the surface was fit is visible as colored points --- dark brown for epicardial data, green for endocardial data, light blue for data from the aortic annulus, dark reddish brown for mitral annulus data.

The control mesh for the subdivision surface is shown as a wireframe with small cubes as handles for the control vertices. One of the control vertices, in the region between the aortic and mitral annuli, has been selected for dragging, using a standard Inventor 3d dragger.

After an edge split Before an edge split Lasse allows interactive changes to topology, through a set of operations that allow editing the structure of the control mesh. The first image shows a selected control edge highlighted bright green, together with a menu of mesh editing operations. The second image shows the result of splitting the selected edge into 2 new edges, adding a new control vertex and splitting the 2 adjacent control faces into 4 new faces.

End systole model with end diastole data.

Lasse allows interactive fitting of the surface geometry to 3d point data. This image shows 3d point data from the end diastole phase of the heartbeat together with a left ventricle model originally fit to end systole data. The data points are colored dots, the color reflecting anatomic region, with lines connecting them to their projections on corresponding region of the surface model. For example, the light blue aortic annulus data points project on the surface edges comprising the aortic annulus surface feature.
After geometric fitting. After rigid fitting.

The first image at right shows the result of rigid 'Procrustes' fitting, that is, translating, rotating and scaling the end systole model to best fit the end diastole data.

The second image shows the result of general geometric fitting, that is, moving the control vertices independently to optimize a loss function that combines goodness of fit and surface smoothness.

A VRML 2.0 model of a human left ventricle If your browser supports VRML 2.0, you can view the 3d model that resulted from fitting to the end diastole data. (If you have an older VRML plugin that has trouble with VRML 2.0, try this VRML 1.0 model.)

Related Web Sites

home