Bioluminescence Imaging
The objective of this project is to develop new visualisation techniques for bioluminescence imaging.
People: Peter Kok, Charl Botha, Jouke Dijkstra (LKEB - LUMC), Emile Hendriks, Marcel Reinders, Frits Post, Erik Jansen, Boudewijn Lelieveldt (LKEB - LUMC, TU Delft)
Theme: Medical Visualisation
Description
In the past decade, molecular imaging has quickly evolved as an important medical imaging tool, especially for small-animal imaging. Using the different molecular imaging modalities that are available today, such as MRI, various types of optical imaging, PET, SPECT and ultrasound, body processes can be monitored at the molecular level in-vivo. Each of these imaging modalities having their specific properties, combining multiple modalities is becoming very important in increasing the amount of information that can be obtained from them. This is the point where advanced data visualization techniques come into play. The goal of our project is to develop visualization techniques that make full use of the available data to significantly increase the amount of information that can be obtained from it and thus allow for more insight into the data.
The main characteristic of molecular imaging techniques is that they are aimed at imaging certain targets or target processes. Probes, also called biomarkers, can be developed to signal at the presence of a target or at a certain event, such as the activation of a specific gene (gene expression), cell division or cell death. The research area aimed at developing such probes is a subject in itself and we will not focus on this.
Particular attention has been given to the optical imaging techniques called bioluminescence imaging and fluorescence imaging, where optical light that is emitted by certain markers is captured using a sensitive CCD camera. The main difference between the two techniques is that fluorescence imaging uses an external light source to excite the probe/marker, where for bioluminescence imaging the light is naturally emitted by cells that have been genetically altered with DNA of, for example, fireflies. The advantage of fluorescence imaging is that the light source location, wavelength and intensity (on/off) can be controlled, and with this extra information it is easier to determine the depth of the signal. On the other hand, the signal-to-noise-ratio in fluorescence imaging is lower than in bioluminescence imaging, because of autofluorescence.
|
Figure 1 - The IVIS 3D BLI scanner from Xenogen. (http://www.xenogen.com)
For our project, and for clarity, we will focus on a specific case study: a mouse that has been injected with a certain form of cancer cells that emit a bioluminescent signal. This enables us to track the location and size of the tumor, which information is used by biologists to test the effectiveness of a certain treatment. The mouse has been imaged with a 3D BLI scanner as shown in figure 1. In this scanner, the mouse, which has been anesthetized, is put on a tray that rotates around a mirror. In this way a number of images can be taken from different angles, in this case four. Subsequently, the mouse is scanned in a micro-CT scanner. This procedure is repeated after a number of weeks in order to track the tumor growth and metastasis.
|
Figure 2 - A traditional overlay of a BLI signal on a photograph. The data in this image and those below has been acquired by the Department of Endocrinology (Courtesy of Clemens Lowik), Leiden University Medical Center
In the traditional way, using the software that comes with the BLI-scanner, the BLI signal is visualized by applying a color map to the BLI-signal and overlaying this on a normal photograph of the mouse (figure 2). This gives the biologist a global idea of where the tumor is located and of its size. However, the 2D view is not ideal in that it doesn't contain any spatial or structural information, which is the reason that we want to combine the data with a 3D structural modality, in this case a CT-scan.
To integrate the information in all these scans, we have to overcome a number of challenges:
- Create a 3D light source reconstruction from the multi-angle BLI images - Currently, we are using a backprojection algorithm that takes the BLI signal on the multi-angle 2D views and projects them into a 3D space (figure 3). This results in a volume with intensity values, where a high intensity corresponds to a higher probability of the light source being located there. We currently don't take tissue into account, which in reality affects the path the light travels. However, because most light sources are located just below the skin, the reconstructions we create can be assumed to be reasonably accurate.
|
Figure 3 - The BLI signal is back projected into a volume.
- Perform BLI-CT registration - The next step is to align the reconstructed BLI data with the CT-scan. Currently we use a landmark transformation approach, where the user has to interactively indicate a number of corresponding landmarks on the photographs and the CT-data respectively. Ideally, image registration would be done automatically, using the outline of the mouse that can be derived from both the BLI-photographs and the CT-data.
- Create a visualization that intuitively combines the two modalities - We applied a number of visualization techniques that may be useful. First of all there is the carousel visualization as shown in figure 3, where the 2D images are placed in a circle and their relation to the CT-data can be seen. Another visualization is shown in figure 4. It displays an overlay of the BLI data on the CT scan, which is similar the traditional visualization, but now in 3D. The user can interact with the image planes and move them through the volume. Figure 5 shows the same dataset, but here a different visualization has been applied that shows the BLI volume as an isosurface and the CT scan as a volume rendering.
|
Figure 4 - Overlay of BLI data on a CT scan.
|
Figure 5 - BLI data isosurfaces combined with a CT scan volume rendering.
- Map an atlas onto the data - Registering an atlas to the CT-data can be useful for several purposes. When comparing data of a mouse that has been scanned at more than one point in time, this will inevitably be difficult, because the mouse will have changed posture (see figure 6). Even if the mouse was scanned in the exact same posture, there could be physiological changes over time. Another purpose for atlas registration might be to use the atlas as a tissue model to create a more accurate light source reconstruction. The work to register an atlas to the CT-data has been done by Martin Baiker (LUMC).
|
|
|
Figure 6 - Posture and physiology change over time, which makes accurate comparison difficult.
- Develop a visualization approach that can be used to compare datasets of the same mouse, but of different points in time - Currently we are working on methods to visualize multiple time points of the combined BLI-CT data. This data has been mapped onto the atlas in order to be able to visually detect the changes that occurred over time.
The Cyttron Visualization Platform
We are implementing the resulting algorithms of all the challenges listed above into a platform called the Cyttron Visualization Platform (CVP). We are developing this platform as a part of Cyttron (http://www.cyttron.org). The Cyttron consortium is a collaboration between a number of companies and universities, which perform various types of projects. The main aim is to integrate data from resolutions that range from the cellular to the molecular level (this is mainly microscopy data). The idea is to create a tool that can be used to determine the cause of diseases at the molecular level and use this for diagnosis and therapy development.
The fact that the CVP is aimed at integrating multiple modalities makes it perfect for our purpose. Also, the CVP has been designed to be easily extendible, so we have implemented a large part of the work directly into the CVP. Currently, we have a plugin that allows the user to walk through a wizard-like interface, processing the BLI- and CT-data step by step. Because this platform is meant to be used by biologists, they can instantly make use of our algorithms. More information on the CVP can be found on the Cyttron Visualization Platform website.
Media
P. Kok, Multiple Modality Registration, Fusion and Visualization for Bioluminescence Imaging, M.Sc. thesis, 2006. http://visualization.tudelft.nl/publications/kok2006_msc_thesis.pdf