Computerized robotics are already easing lab-based wet tasks such as feeding cells and changing media. Several vendors now sell programs that can track cells in flat culture, keeping them in focus and in the field of view.

The latest version of the Nikon BioStation CT can 'memorize' the positions of non-motile cells before a plate is removed for media exchange, and can then continue tracking them when the plate is replaced, avoiding the 'image jiggle' that would disrupt statistical analysis, says Ned Jastromb, a senior application manager at Nikon Instruments in Melville, New York. It also integrates a calendar function with a robotic system that slides culture plates in and out of an imaging area on schedule, allowing one instrument to run several long-term experiments.

But software is poised to solve a wider range of problems. By combining a fast image-acquisition program with a noise-reducing algorithm that compares consecutive images, John Sedat at the University of California, San Francisco, and his colleagues decreased the amount of light needed to image yeast cell division by several orders of magnitude 5. Advances in fully automated cell identification and tracking, and modern continuous cell-imaging techniques can outperform traditional manual methods6. Historically, software advances have spread slowly because programs designed to follow a particular cell type tend not to recognize other types, says Andrew Cohen, a computer engineer at the University of Wisconsin–Milwaukee.

More broadly, Cohen says he may be on the cusp of solving a problem that plagues many live-cell imaging experiments. Many software programs work only when cells are sparse. That limits the technology because some cells can grow only in dense cultures, and some cells divide many times before producing the desired cell types, in which case a single cell produces hundreds of daughters. By the time the most interesting cells appears, it is impossible to tell which cells they came from. Recently, Cohen found that an algorithm he originally wrote to follow hundreds of organelles within a single cell can be applied to trace neural stem-cell fate. “Our ability to track very high-density image sequences is going to improve very rapidly,” he says.

Larger advances, however, may come less from improvement in technology than from biologists' awareness of what software can do, says Cohen. “Sometimes the biologists start out just wanting to characterize data, and they don't think about the big questions they can ask.”

M.B.