Aug 15, 2005 | Confocal microscopes provide a window into the finest details of cellular structure and intercellular dynamics. But most labs using the technology face considerable challenges. Microscopic images require lots of manual intervention to be interpreted, and while there are numerous algorithms for image analysis, many use different data formats. Researchers must summon large amounts of computing power and memory to analyze and visualize anything more than a single confocal scan.
Category: Knowledge Management
Title: Confocal Microscopy Data Analysis
Organization: National Cancer Institute
Partner: Silicon Graphics
To overcome these problems, Jack Collins and colleagues at the National Cancer Institute (NCI) Advanced Biomedical Computing Center and Image Analysis Laboratory partnered with Silicon Graphics (SGI) to develop an automated imaging workflow and analysis process, affording researchers a faster way to study details about cancer cells and tissues.
“Today, we’ve got many more images and lots of data,” says Collins, manager, scientist, at the Advanced Biomedical Computing Center. “[We] needed something to automate tasks like determining the edge of a cell and segmenting the cell.” Software tools can help, but they require some human intervention. Given the size of the data sets being analyzed, the time-consuming calculations make the process protracted and tedious.
For instance, researchers analyzing cellular confocal images typically perform various steps for data analysis — capturing the confocal data, orienting the image (along a cell axis, for example), applying a mask that differentiates various elements within the image, and determining the image intensity. NCI wanted something to accelerate this process. “We wanted the process to be interactive — like recalculating an Excel spreadsheet,” says Collins.
The first priority for NCI and SGI was to examine the existing procedures. “There are many programs to analyze the data and no one workflow for all cases,” says Curtis Lisle, solution architect at SGI. “[People] would often send images off in e-mail, so Outlook was the workflow.”
After examining the processes, a set of requirements was drawn up, with computing power and memory near the top. “Confocal image analysis has typically been done on [a slice] on a PC,” says Daniel Stevens, business manager of SGI Material, Chemical, and Life Sciences. “We want to turn slices into 3-D images for analysis. This requires putting all the data in core memory and having a system with strong visualization capabilities.”
Noting that many of the problems such as loading, viewing, analyzing, annotating, and storing volumetric data sets were common to microscopy scientists and clinical imaging researchers, SGI decided to leverage work that was already done by these communities. To that end, it used 3D Slicer software from an industry-wide open-source project. 3D Slicer allows researchers to interactively examine and process medical imaging data sets. It handles multiple data formats and provides graphical tools for a user to examine features in a data set. In the NCI project, SGI used plug-ins so that the Slicer software could tap hardware-assisted visualization features using SGI’s OpenGL Volumizer API, which pre-loads large amounts of data into memory and speeds rendering of images.
Next, the group looked at software for segmentation that would “ease new algorithm creation,” according to SGI. Again, the team leveraged existing software. It selected the National Library of Medicine Insight Segmentation and Registration Toolkit, an open-source software system originally designed to support the Visible Human project, but with features that could be applied to microscopy-specific analysis.
Pulling these various aspects together, the project gives researchers an interactive tool that allows them to visualize and analyze large confocal microscope data sets. The solution taps hardware-assisted visualization technology and SGI’s shared memory Altix and Prism systems so that huge data sets can be analyzed. “We’re bringing lots of computer power and lots of visualization into large data sets,” says Lisle.
The workflow takes in images, crunches them, and lets scientists interact with them from the desktop. “It allows [scientists] to get to the heart of their research without moving lots of data around.” Researchers now have the luxury of looking not just at a single cell, but at intercellular relationships — a prerequisite if intercellular dynamics are to be better understood.