Novel computer model for breast cancer

A worldly-wise computer model can forecast which breast cancers will spread and so need aggressive treatment. Researchers at University of Hopkins, have developed a neural network system - where a computer is taught to spot patterns - in combination with fuzzy logic, a way of modeling the uncertainty of natural language. The result is a model which can scrutinize images of breast cancer cells, and indicate how likely it is the tumor will spread.

Tested on samples from 90 women with breast cancer, the computer system was 80 per cent accurate in predicting which would spread to the lymph glands and in predicting survival after five years. The system is an extension of an existing analysis technique called image cytometry, where a computer analyses cell properties. It relies on looking at which cells have extra DNA, the patterns of DNA levels, the number of dividing cells, and the shape of the cell nuclei. The fully tested new system, the most advanced of its kind - could help doctors identify those women who need the most aggressive treatment of their breast tumor.