New mammogram measures to predict breast cancer risk
An international team of researchers, led by The University of Melbourne, has found two new mammogram-based measures for breast cancer risk. Published in the International Journal of Cancer, the new measures could substantially improve the screening process, make it more effective in reducing mortality and less stressful for women.
Since the late 1970s, scientists have known that women with denser breasts, which show up on a mammogram as having more white or bright regions, are more likely to be diagnosed with breast cancer and to have it missed at screening. With this in mind, University of Melbourne researchers collaborated with Cancer Council Victoria and BreastScreen Victoria to study other ways of investigating breast cancer risk using mammograms.
Using computer programs to analyse mammogram images of large numbers of women with and without breast cancer, the researchers found two new measures for extracting risk information: Cirrocumulus, based on the image’s brightest areas, and Cirrus, based on its texture.
First, the team used a semi-automated computer method to measure density at the usual and successively higher levels of brightness to create Cirrocumulus. They then used artificial intelligence (AI) and high-speed computing to learn about new aspects of the texture (not brightness) of a mammogram that predict breast cancer risk and created Cirrus. When their new measures were combined, they substantially improved risk prediction beyond that of all other known risk factors, including breast density and genetics.
Lead researcher Professor John Hopper said that, in terms of understanding how much women differ in their risks of breast cancer, these developments could be the most significant since the breast cancer genes BRCA1 and BRCA2 were discovered 25 years ago.
“These measures could revolutionise mammographic screening at little extra cost, as they simply use computer programs,” Prof Hopper said.
“The new measures could also be combined with other risk factors collected at screening, such as family history and lifestyle factors, to provide an even stronger and holistic picture of a woman’s risk.
“Tailored screening — not ‘one size fits all’ — could then be based on accurately identifying women at high, as well as low, risk so that their screening can be personalised.
“Given mammography is now digital, and our measures are now computerised, women could be assessed for their risk at the time of screening — automatically — and given recommendations for their future screening based on their personal risk, not just their age.”
Adjunct Associate Professor Helen Frazer, Clinical Director of St Vincent’s BreastScreen Melbourne, added that improvements in assessing a woman’s risk of breast cancer would be transformative for screening programs.
“Using AI developments to assess risk and personalise screening could deliver significant gains in the fight against breast cancer,” she said.
Researchers have successfully used machine learning to complete cumbersome materials science...
Using machine learning, researchers have identified distinct patterns of coordinated activity...
Researchers used field data to 'train' a machine-learning model to detect gamba grass...