Getting clinical about neural networks

A trained human being is, in many cases, the best diagnostician. This certainly remains true in medicine, but the success of new imaging technologies and clinical applications of molecular biology has meant that doctors are now inundated with diagnostic data. Although rule based systems have had some limited success assisting doctors, much concern remains about the poorly defined factors that physicians must use for diagnoses. Neural networks, which cross the line between artificial intelligence and statistical regression, find relationships that do not clearly spring out from a mass of data. Neural networks attempt, in an abstract mathematical way, to mimic the way the brain processes sensory information and are moving from the laboratory to the medical marketplace in the image processing units used to diagnose cancer or other diseases.