• Courses
• in4086p
• Exercises
• Volume

IN4086P Volume Visualisation

Welcome to the Wonderful World of Wolume Wisualisation! In this exercise, you will learn how to apply the three basic volume visualisation techniques (MPR, surface extraction and DVR) to a medical dataset. All three of these techniques originated from visualization research and are now a valued tool in clinical practice.

IMPORTANT NOTE: For the shading part of this exercise, you need a specially modified ParaView. The Linux installation at the Drebbelweg has already been modified. If you want to make use of your own build, please use the patch appropriate for your version: ParaView 3.2 / ParaView 3.4 This patch is only relevant if you want to compile your own ParaView and you want to activate shading during direct volume rendering. In this case you'll also want to activate Python support in the CMake configuration.

Loading the data

Especially for you, we have prepared a DICOM dataset containing an abdominal angiogram acquired on a 16 detector CT scanner of a patient with an abdominal aortic aneurysm (AAA). DICOM, or Digital Imaging and Communications in Medicine, is a specification for file formats and communication protocols that are used by most medical imaging equipment. Because ParaView can't read DICOM directly, we've pre-converted the data to the VTI format. Whilst we were at it, we image processed away certain external structures (for example the table that the patient was lying on) and downsampled the data with a cubic interpolation filter in order to facilitate this exercise. The dataset filename is panoramix_m_voi_0.8.vti. Download and unpack the zip archive containing this file from the in4086p dataset page.

MPR visualisation

MPR, or multi-planar reconstruction / reformatting, is the most basic and probably most popular volume visualisation technique, and refers to the interpolation of an arbitrary slice of data from a volume. The scanner usually has a primary slicing direction, often axial. With MPR, coronal or saggital slices, or even oblique slices, can also be extracted from the data.

Extracting surfaces

The well-known Marching Cubes algorithm can be used to extract triangulated iso-surfaces directly from volume data.

To be able to extract meaningful iso-surfaces from medical data, one needs to derive suitable iso-values first. There are several ways to do this. One could make use of a histogram to get an idea of the distribution of scalar values in the data.

One can also probe (sample) the data more directly. In ParaView, this is done by making use of the selection interface. Activate View | Selection Inspector, and in the window that appears make sure that your slice is the Current Object, that Selection Type is set to IDs and Field Type to POINT. Now click on the Create Selection button. On the Point Label tab at the bottom, set Visible to on and Label Mode to Scalars_. Now deselect everything in the Pipeline Browser.

Now you can select the Point Selection tool (icon in the bar at the top with a triangle with 3 red corners) and then click on the slice to see the intensity at that position.

Continue using the point selection tool, and then clicking somewhere on one of your slices. It should show the scalar value at that point.

Based on the histogram and direct probing, one can experiment with thresholding to determine which threshold values reveal interesting structures in the data. ParaView Linked Views, explained in the first question along with histograms, are useful when doing this type of exploration. We will now see how.

You should now have a good idea of the scalar makeup of the CT dataset that you are working with.

Direct volume rendering

Now the fun really starts! Direct Volume Rendering (DVR) is a technique for the direct visualisation of volume data. Please study the linked wikipedia article. What's not explicitly mentioned in the article, is that there are two main types of DVR, image-order, for example raycasting, and object-order, for example splatting.

ParaView supports the raycasting type of DVR. Recall that in raycasting, there is a strong relation between the size of the rendered image and the speed at which it can be rendered. Practically, this means that if you zoom out far enough, you will get high interactivity in return. This is useful when you are tuning raycasting parameters for example. Zoom out, tune, zoom in for a final high quality view.

An important element of DVR is the transfer function. This function maps data values to optical characteristics, such as colour and opacity. In raycasting, it is applied at every step of the integration, usually after the sampled data value has been interpolated. This was also discussed in the wikipedia page on Volume Rendering, which you studied a few paragraphs ago.

Up to now, you have been using a straight-forward raycasting implementation without shading. Each interpolated volume sample is transformed to a colour and opacity via the transfer function and then accumulated with other values along its ray using the volume rendering integral. However, in most modern volume rendering implementations, lighting (shading) is also integrated. This entails that the effect of light (under the Phong model ambient, diffusive and specular light) is simulated, based on the image gradient and the looked-up colour and opacity.

You are now going to activate shading in ParaView! The ParaView GUI does not yet support this, so we have hacked the source to make it possible via the Python scripting interface. Activate this interface by selecting "Tools | Python Shell" from the main menu. Cut and paste the following script into the scripting interface, line by line using Ctrl-C Ctrl-V (middle button paste will NOT work, due to Qt). If you have multiple views, the index to GetRenderView should correspond to the view containing the volume rendering.

   1 from paraview import servermanager
   2 view = servermanager.GetRenderView(0)
   3 names = [i.__class__.__name__ for i in view.Representations]
   4 r_idx = names.index('UniformGridRepresentation')
   5 r = view.Representations[r_idx]
   6 r.Shade = 1
   7 

There are some beautiful examples of direct volume rendering on the web. See for example the Voreen image gallery, or do a Google Image Search for "direct volume rendering". One of the most important factors in creating good volume renderings is a suitable transfer function. In ParaView, you can zoom out to get faster updates as you finetune the transfer function (as explained above). You could also consider using the "Extract Subset" filter to extract an interesting sub-volume of the dataset and focus on that.