MIT’s Computer Science and Artificial Intelligence Lab has developed a new tool that uses artificial intelligence (AI) and deep learning to predict the development of breast cancer up to five years earlier than current detection techniques.
Unlike other AI-based tools that use data mostly from acquired from white patients and may therefore be algorithmically biased, MIT’s model works equally well for both white and black patient populations. This is particularly important as black women are 42% more likely to die from breast cancer than white women – a possible contributing factor could be that current detection techniques don’t work as well in black women.
MIT’s AI tool was shown to be able to identify a woman at high risk of breast cancer four years (left) before it developed (right).
To create the tool, data was obtained from over 90,000 mammograms and outcomes from over 60,000 patients treated at the Massachusetts General Hospital. A form of machine learning known as “deep learning” was then used to identify patterns in the images and data that are too subtle to be routinely recognized by human physicians. The results have so far have been more accurate than current diagnostic approaches, presumably because the model is not based on existing knowledge and assumptions about patient risk factors.
MIT’s project is intended to help doctors compile the best personalized screening program for their patients and hopefully eliminate the heartbreaking outcome of a late breast cancer diagnosis. We eagerly await to hear if and when this could become widely available.