Neural networks diagnose cancer

May 31, 2001 | Source: Wired News

Artificial neural networks have succeeded in diagnosing cancers based on gene expression signatures for the first time, according to a National Institutes of Health study published in the June issue of Nature Medicine.

Using data from gene chips (wafers filled with DNA that are analyzed to identify which genes are expressed, or turned on), the researchers fed the neural network software 6,000 genes containing 88 types of cancer. They taught the neural system to recognize 65 of them.

The researchers then presented the computer with 25 additional cancer types that the computer hadn’t seen before to see if it could categorize them accurately. It did.

A tool based on this research could be available in less than three years, the researchers said.