Besides holding out the prospect of new diagnostic tools and techniques, these cross-campus partnerships provide apt examples of how biologists, engineers and computing scientists are meeting more regularly in the growing field of biological computation. Biocomputing involves using computing methods to make sense of ever-increasing amounts of biological information.

  The first project might eventually yield a computer tool to diagnose the causes of lameness in horses and perhaps in other livestock such as pigs and cattle.

  This collaboration would see information collected about mechanical stresses on horses' hooves fed into a computer-based neural network "trained" to distinguish between normal and lame horses.

  Prof. Jeff Thomason, Biomedical Sciences, envisions a diagnostic device equipped with special software that veterinarians might use to gain more precise information about the nature of a hoof injury.

  He has begun working with Computing and Information Science (CIS) professors Dave Calvert and Deb Stacey, both of whom study neural networks as part of the Guelph Natural Computation Group, along with other faculty in CIS and the School of Engineering.

  Pointing to a schematic on his computer monitor, Thomason explains that gauges hooked to a horse's hoof yield information on the angle and magnitude of strain at different points as well as how those variables change through the horse's stride. (The measurements have been taken from horses at several locations, including Sunrise Equestrian Centre. See the Dec. 8, 1999, issue of @Guelph.)

  Those measurements collected with each stride - along with additional information on whether the horse was being ridden or hand-led, turning or moving straight, shod or not - yield "hundreds of thousands of numbers," says Thomason. He has used statistical analyses to hunt down patterns in the data, but these methods don't lend themselves to making predictions about individual cases. "They may be able to distinguish all lame horses from all healthy ones, but not a single lame horse. This is where neural nets come in."

  By chewing through piles of numbers and grouping and labelling the data, the neural network would quickly "learn" the rules for telling healthy and lame horses apart. "Can you tell all the circumstances from just looking at strain information?" Calvert asks rhetorically. "If yes, then we can start getting records from lame horses and start exploring the use of neural networks as a diagnostic tool."

  Thomason says vets making on-farm visits often have trouble discerning precisely what ails a lame horse beneath its hoof. Armed with a kind of electronic veterinary assistant, "there's no reason someone couldn't go around or be employed by horse owners to test horses on a regular basis," he says.

  Such a system might even be useful to pig or cattle farmers, whose livestock can develop sore feet if they're kept on a concrete floor or a slatted surface designed for manure collection and disposal.

  Calvert says his first task once he starts receiving information from Thomason will be to determine whether the computer can provide reliable information about a horse's condition based on strain data. He plans to hire a student under an undergraduate research assistantship for this project. "We're training neural networks to extract patterns from the data."

  This work builds on the kind of expertise he developed on another project that seems similarly removed from his desktop in the Reynolds Building. Later this year, he will present a paper at an engineering conference about using a neural network, along with Prof. Mary Buhr, Animal and Poultry Science, to pinpoint breeding bulls likely to produce high-quality semen.

  Developing a better way to diagnose the causes of lameness in racehorses is also the purpose of the second OVC/CPES collaboration, this time between Prof. Howard Dobson, Clinical Studies, and Prof. Bob Dony, Engineering. They plan to blend their respective skills in radiology and signal processing to construct a more sophisticated imaging device that might save horses - and their owners - unnecessary grief when the animals are being checked for orthopedic and neurological problems.

  Currently available devices for magnetic resonance imaging (MRI) require vets to anesthetize an animal and lay it down. Because of the risk of complications, racehorse owners - the primary market for this sophisticated imaging technology - shy away from exposing their animals to general anesthetic.

  Using a device envisioned by the Guelph researchers, a veterinarian would need only sedate the horse, which would be imaged while walking between two halves of the device, roughly analogous to an airport electronic detector.

  Besides making the process less stressful, this device would give the vet a clearer picture of the loads on the horse's leg. Dobson explains that the device would typically be used to diagnose the cause of lameness, such as arthritis or infection.

  Dony's work in signal processing has found application before in medical imaging, such as image processing and compression for chest X-rays taken at a Hamilton hospital. "The logical next step was to work with someone at the Ontario Veterinary College," he says.

  Dobson says a new device would either be adapted from available off-the-shelf equipment or built from scratch. Building such a device would cost about $5 million, compared with the $1 million to $2 million for an existing machine. He plans to apply for federal research funding to pay for the equipment.

  Dony envisions working on software design for the MRI. "I would like to be involved with the engineering and development of the physical device and the computer processing that goes into creating the image."

  He would also analyse the images, including trying out new computational techniques for processing images that would give clearer pictures of the health of an animal's hoof and leg. "That follows up from my interest in signal processing," Dony says, adding that the School of Engineering plans to introduce undergraduate and graduate courses in imaging.

  Dobson says neural networks are increasingly being investigated for their use in medical diagnostic imaging. "A neural network will refine my ability to make a diagnosis," he says, noting that such a device would complement existing diagnostic tools and equipment at OVC.