Neural nets help diagnose cancer

Using computer-based neural networks, researchers at the University of Maryland Greenebaum Cancer Center have devised a new method to differentiate and diagnose several types of colon tumours.

The method, which uses ‘artificial neural networks,’ or ANNs, to analyse thousands of genes at one time, could ultimately help doctors to identify the cancers earlier and spare some patients from unnecessary, debilitating surgery, says Stephen J. Meltzer, MD, professor of medicine at the University of Maryland School of Medicine. Dr. Meltzer is the senior author of a study to be featured on the cover of the March issue of Gastroenterology, the journal of the American Gastroenterological Association.

Patients with Crohn’s disease and ulcerative colitis, the two forms of inflammatory bowel disease (IBD), have an increased risk of developing cancer, but the cancer can be one of two forms. ‘Sporadic,’ or common, colon cancers can often be removed without radical surgery, while IBD-related growths and cancers are much more aggressive and are generally treated by taking out the entire colon.

‘Until now, we had no reliable way to discriminate between these two types of lesions, especially in their early stages,’ says Dr. Meltzer, who is also associate director for core sciences at the University of Maryland Greenebaum Cancer Center and director of the cancer centre’s Genomics Core Facility.

‘This study helps to establish a new method, called artificial neural networks (ANNs), that can be used in a wide variety of disease settings, not just in cancer,’ he says. ‘These networks can be trained to recognise specific disease lesions or subtle differences within disease categories,’ he says.

Dr. Meltzer says the study is one of the first to use this type of ‘artificial intelligence’ to diagnose disease and is believed to be the first involving colon cancer or other gastrointestinal diseases. Last year, researchers used a similar method to differentiate four types of tumours in children.

In the Greenebaum Cancer Center study, researchers took 39 tissue samples from patients at the Baltimore VA Medical Center and Mount Sinai Hospital in New York who had well-documented cases of ‘sporadic’ cancers or cancers related to IBD.

The researchers extracted the DNA from the samples and then used gene microarray equipment to analyse 8,064 genes to determine the level at which they were present in each colon sample, according to Florin M. Selaru, MD, research associate in the Department of Medicine at the University of Maryland School of Medicine, director of bioinformatics and data analysis at the Greenebaum Cancer Center, and the lead author of the study.

These ‘gene expression’ levels were translated into numbers, which were processed by ‘artificial neural networks,’ multi-layer mathematical programs that are capable of recognising complex patterns in large amounts of data.

Using gene information from 27 of the 39 samples, researchers ‘trained’ the neural network to recognise the two types of colon cancer, and then gave it information from 12 samples it had never seen. It made the correct diagnosis in all 12 cases.

‘We now have a tool that is extremely precise, which may prevent misdiagnoses and unnecessary surgeries and help us treat patients most effectively,’ says Dr. Selaru, who developed the computer algorithm used in the study.

The researchers were also able to reduce the number of genes necessary to make the correct diagnosis from 8,064 to 97, which would make the method easier and less expensive if this technology became more widely available.

The University of Maryland researchers believe the use of ANNs may eventually have a significant impact on the early detection and treatment of colon cancer in people with IBD.

This study was conducted using technology that was funded, in part, by the Maryland Cigarette Restitution Fund Program.