Three-dimensional Image Analysis
Measurement and analysis add another dimensionDean Sequera
It wasn't that long ago that three-dimensional (3-D) rendering of microscopic objects was an elusive dream of microscopists. The only available technology was expensive, difficult to configure and hard to use. The computers required to run these systems were highly specialized, with unique architectures, and required a high degree of sophistication to operate. But times change, and affordable computing has suddenly brought 3-D imaging to mainstream laboratories everywhere.
It shouldn't be surprising, but the technology to render 3-D image volumes is a result of the PC computer gaming industry and its constant drive to create ultra-detailed, highly realistic virtual environments. Now, the very same technology allows accurate modeling of the 3-D workspace that has been sought by the research community. Many off-the-shelf PCs incorporate the necessary hardware to render, rotate, section and quantify image volumes. So, tracking down the right hardware is not hard or cost-prohibitive.
In parallel developments, the microscopy world has made its own technology advancements. Confocal microscopes are capable of producing impressive and highly resolved image volumes. Not to be outdone, microscopists with widefield epifluorescence microscopes coupled with deconvolution software are also creating highly resolved volumes. The resulting image volumes are revealing information that has never before been exposed, much less comprehended. More importantly, these finely resolved 3-D volumes reveal details that lead to broader understanding of the spatial relationships of the structures under investigation. When quantitative 3-D analysis tools are added to remove subjectivity, a new dimension in scientific inquiry is born.
A tool that has long been used to examine relationships in image volumes is a concept commonly known as extended depth of focus (EDF). This method takes a Z-stack of image planes and compresses it into a single composite image. Based on such characteristics as local contrast, pixels are se
Figure 1: A 3-D picture representing a kidney glomerulus surrounded by tubules. Fluorescent probes are used to localize and identify the structures' molecular components. By using probes such as these, details can be revealed that would otherwise go unnoticed.
Robyn Rufner, Director of the Center for Microscopy and Image Analysis (CMIA) at George Washington University Medical Center, is a big supporter of the technology. "3-D imaging has enabled us to image cellular relationships that we were not able to observe before. In the past, we could not completely visualize the morphological architecture of samples. We just didn't have the means to do it," said Rufner.
Since the CMIA is a multidisciplinary core facility, a wide variety of applications and imaging problems are seen through the oculars of the microscopes located there. For example, one team of researchers is trying to characterize corneal stem cells of mouse eyes by evaluating the expression profiles of specific proteins (integrins). Using a combination of laser confocal microscopy and 3-D imaging, they are able to visualize the different layers that comprise the surface of the cornea. The implications of this research include potential stem cell therapies associated with corneal repair and regeneration.
Figure 2: Image-Pro Plus with 3-D Constructor Plug-in Module, built using the underlying Open Inventor libraries, is used to visualize a three-color micrograph of a kidney glomerulus taken with an Olympus Fluoview confocal microscope.
At the University of California Medical School in San Diego, Satish Eraly makes his living as a nephrology researcher. His own investigation into the function of kidneys provides insight into the importance of studying the morphology of the glomerulus.
"Glomerular diseases, including those caused by diabetes, are one of the most common causes of kidney failure," Eraly said. "So, the ability to image the glomerulus in different states is central to the understanding of these diseases."
To study these tiny pumps and to show the intricate forms they create, fluorescent molecules are directed toward receptors on the surface of the glomeruli. When fluorescent light excites the molecules, a particular wavelength of light is emitted, revealing details in an otherwise unobservable environment. To show the relation of glomeruli to the surrounding tubules that pick up the waste and re-absorb vital salts and proteins, fluorescent molecules emitting a different wavelength are attached to those tubules.
Confocal and widefield epifluorescent microscopes create image stacks by focusing at subsequently deeper optical sections in a sample. And microscope objectives are capable of resolving fine details in both spatial and axial dimensions as they focus through a specimen. The image stacks that are created contain enough information to be built into impressive volumes. Flat detail becomes brilliantly animated. Features relate to those around them. Spatial arrangements become much more understandable.
In Santa Barbara, at the University of California, Brian Matsumoto runs the Integrated Microscopy Facility for the Neuroscience Research Institute and the Department of Molecular, Cellular, and Developmental Biology. Part of his work involves high-performance imaging as a means to evaluate emerging technologies. So, a highly detailed specimen like a glomerulus makes for an ideal test subject.
"The complex tuft of vessels provides an ideal specimen to determine fine 3-D architecture," he says.
Figure 3: This pseudocolored image displays the relative location of two probes: wheat germ agglutinin (green) and phalloidin (red). They show, respectively, the position of sialic acid and N-acetylglucosaminyl residues, and f-actin. Blue Nuclei- DNA.
Here is how it all comes together:
1) Image volumes are imported into the software, usually in the form of a TIFF image stack.
2) The software then renders the image stack in 3-D, taking into account spacing between image planes to create a lifelike and accurate model of the 2-D planes.
3) Voxels, the 3-D world's equivalent of the pixel, are built up like the ultimate adult version of building blocks, each one containing spatial and intensity information. The voxels are created by first determining the calibrated size of the pixel and then adding the distance between image planes as the third dimension.
4) Calibrated standards are applied and a measurable volume is created.
The advantage of 3-D visualization is the ability to rotate a stack through space, observing relationships through a number of perspectives. Frames collected at regular intervals can be assembled into movies that add visual appeal and the sense of being there with the objects.
Although qualitative rendering is insightful, the key to scientific advancement is measurement. Quantitative analysis removes subjectivity and offers the ability to repeat experiments and to know that the same results are generated. A vast array of measurement possibilities unfold in 3-D - from surface area of a single object to angles and distances between objects.
However, a potential problem exists with all this data. How in the world to model it all? After all, an individual voxel is comprised of six faces and eight corners. Millions of voxels can be represented in a single volume! Trying to keep track of every face and vertex of every voxel would wear out the standard-issue desktop. Processing times also would be affected, resulting in slow rotation speeds and unsatisfying performance.
Most video cards used for visualization of volumes cannot display voxels as semi-transparent 3-D blocks -- this is supported by advance graphic processors. To overcome this limitation, three sets of 2-D semi-transparent slices perpendicular to each axis are created. Only one set, which faces the observer, is shown. The rendering task is done by the processor of the video-card, which uses short, OpenGL commands sent by the computer, so that the main processor can be used more effectively. Some volumes may not be loaded with full resolution due to memory limitations. In the example above, the software uses sub-sampling which, when applied, retains the original resolution of the volume.
In today's world, the microscopist's dream of affordably rendering 3-D stacks has not only been achieved, it has been surpassed. Measurement and analysis add another dimension to this third dimension.
Dean Sequera is Vice President, Marketing & Product Development at Media Cybernetics. He may be contacted at firstname.lastname@example.org.