Factors such as the shape of bacteria, their refractive indexes - or how much they bend light - the types of substances secreted by a particular bacterium and the distance between individual bacteria in a colony, all contribute to how a colony distorts light. The procedure identifies a bacterial colony by comparing an image of its scatter pattern against a template that contains 120 features described by Zernike polynomials.

"A good analogy is the method used by law enforcement to identify a person's face using specialized recognition software," Rajwa said. "You could describe the face as being made up of a combination of geometric shapes, like ovals, squares and triangles, but each face has a unique blend of these shapes. We did something similar. We reduced complicated scatter patterns to 120 numbers based on Zernike polynomials."

This reduced collection of numbers describes how well the colony fits the template, and then pattern recognition software is used to classify the bacteria.

"One of the most important developments is being able to convert images to numbers, which makes it possible to classify the patterns," Rajwa said. "We are able to take images and convert them to numbers that uniquely describe every picture."

The researchers used the new system to classify six species of listeria, only one of which is a dangerous food-borne pathogen for humans.

"If you have a mixture of different listeria, you would like to know which is the one that can kill you," Rajwa said. "We took pictures of the scatter patterns from different listeria, and we were able to classify all of them accurately."

The system also was able to accurately identify other types of bacterial colonies, including salmonella, vibrio, E. coli and bacillus.

"We were able to classify bacterial colonies with greater than a 90 percent probabi
'"/>

Contact: Emil Venere
venere@purdue.edu
765-494-4709
Purdue University
27-Jul-2006