Transforming breast cancer screening with artificial intelligence (AI)
The Breast Cancer AI Project (BRAIx), which was recently awarded a substantial Australian Government Medical Research Future Fund (MRFF) grant aims to transform breast cancer screening in Australia with artificial intelligence (AI). The project involves leading clinicians from St Vincent’s Hospital Melbourne and BreastScreen Victoria, medical researchers from St Vincent’s Institute of Medical Research, University of Melbourne and University of Adelaide, and aims to better use mammography to prevent women dying from breast cancer in a way that improves detection, lowers harms, reduces costs, causes less stress and can be quickly put into practice.
The clinical challenge
Breast cancer is the most common cancer affecting Australian women. Mammographic screening reduces the risk of dying from breast cancer, however, interpretation of mammographic images is challenging, subject to human variability, and has some room for improvement. Despite independent double reading of all mammograms by radiologists (and a third arbitration read if there is disagreement), approximately 33,000 Australian women are recalled annually for assessment and later determined not to have breast cancer (false-positive), whilst approximately 1000 women subsequently discover they have breast cancer after receiving an ‘all clear” result (false-negative). The cost of the public breast screening program, at over $300m annually, is also rising with Australia’s aging population.
The data-driven solution
The BRAIx project has assembled a cross-sectoral and interdisciplinary health research team comprised of epidemiologists, AI computer scientists, data scientists, health statisticians, radiologists and breast surgeon (healthcare practitioners), qualitative researchers, and genomic researchers. The diverse team is investigating utilising novel AI techniques in developing deep learning algorithms to improve analysis and interpretation of mammograms and ultimately transform breast cancer screening.
The BRAIx team has already demonstrated the opportunity to significantly improve screening outcomes, lower harms and reduce costs using AI. The BRAIx team’s current AI algorithm was trained to distinguish between ‘normal’ and ‘cancer’ samples from small patches of mammograms. The team then tested the model using previously unseen images. Their model produced 88 per cent accuracy, which is on par with human performance.
The next stage of research for the BRAIx team is applying their developed deep learning algorithms in a ‘proof of concept’ prospective screening study to measure the model’s real-world performance accuracy at detecting cancers. Concurrently, BRAIx researchers at University of Melbourne are applying models to discover novel aspects of a mammogram that predict a woman’s risk of breast cancer. Underpinning these novel AI research aims, BRAIx researchers are also leading engagement on the ethical, legal and social implications of utilising AI models in healthcare, developing approaches to explain AI prediction and will use a co-creation approach in implementing these models in St Vincent’s BreastScreen services.
Expected project outcomes
The researchers don’t imagine that they will be able to do away with human interpretation altogether: their vision is to replace one of the initial two reads of each mammogram with a read done by artificial intelligence. With more accurate results delivered more quickly, they hope to reduce the burden on the individual as well as on the health-care system, which wastes considerable resources following up innocent abnormalities.
The major and novel outcome of this project is to make it possible, automatically and at the time of a mammographic screen, to identify the presence of breast cancer or a woman's risk of developing breast cancer. This paradigm-shifting discovery will change mammographic screening from the current ‘one-size-fits-all’ protocol. Radiologists will no longer be looking at mammograms blind to key information that can guide them, and screening will be more effective. The door will be opened for more nuanced screening that incorporates a woman’s individualised risk profile based on subtle imaging features the AI-models learn.
To get involved contact email@example.com