Mathematics could be used to diagnose breast cancer

Two Flinders University lecturers have developed a computer algorithm that can be used to detect an invasive form of breast cancer that is often undetectable on screening mammograms.

For the past year and a half, Dr Murk Bottema and Dr John Slavotinek have worked together to develop a computer program to identify invasive lobular carcinoma, a moderately rare form of breast cancer that represents 8% of all cases and is extremely difficult to detect.

"Radiologists can detect most breast cancers by visual inspection of screening mammograms," Dr Bottema said. "However, invasive lobular carcinoma is extremely difficult to identify using this method which is why over half of the cases of this type of cancer are undetectable. It is for this reason we believe it is imperative to find a method of using computers to help make the diagnosis more accurate."

Dr Bottema and Dr Slavotinek's process involves scanning mammogram images through a digitiser, a device which translates pictures into an array of numbers based on how light or dark certain sections of the image are.

The numbers are then fed into the algorithm, which is designed to look at small, low texture features of the breast tissue, rather than the large, obvious or high contrast regions that are examined to detect other forms of cancer.

According to Dr Bottema, the changes in texture, or the presence of cancer, are represented by different shades of grey in the image and the algorithm is able to identify any anomalies or patterns in the numbers that correlate to these areas. The numbers are then compared with those belonging to a 'normal' or non-cancerous mammogram image to record the differences between the two.

"We have already carried out a preliminary study of the program using 48 mammogram images - 24 'normal' ones and 24 with cancer," Dr Bottema said. "Of the 24 with cancer we were able to detect abnormal areas in half of them, but when those same images were handed to a radiologist, they were unable to tell the difference between them and 'normal' images."

Item provided courtesy of Flinders University