2 min Analytics

MIT’s AI tool can recognise breast cancer five years before development

MIT’s AI tool can recognise breast cancer five years before development

MIT’s Computer Science and Artificial Intelligence Lab has developed a predictive artificial intelligence (AI) model, which can predict the development of breast cancer five years before that time. The model is also designed to be as accurate for white women as it is for dark women.

The researchers working on the project knew that similar projects often have prejudices in the AI, because they were mainly trained on data from white patients. That’s why they designed their own model in such a way that it was informed on the basis of “fairer” data, says TechCrunch.

According to the researchers, this is also necessary, because black women are 42% more likely to die from breast cancer than white women. One reason for this may be that this group of women is not so well helped by the current techniques for early detection of the disease. The aim of the development of the technique currently used by the researchers was to improve the accuracy of the assessment of this type of health risk among minorities, who are often less well represented in the development of deep learning models.

Training tool

The MIT tool has been trained for mammograms and patient results – with the eventual development of cancer being an important result – of nearly 60,000 patients. A total of 90,000 mammograms were used, all from patients at Massachusetts General Hospital. The tool starts with the data and uses deep learning to identify patterns that human doctors cannot see.

Because the tool is not based on existing assumptions or received knowledge about risk factors – which is a suggestive framework – the results appear to be much more accurate for the time being. This applies in particular to predictive discoveries, even before a diagnosis has been made.

The project aims to help health professionals build a good screening programme and prevent the consequences of late diagnosis. MIT further hopes that the technique can be used to improve the detection of other diseases that also have problems with existing risk models with low accuracy.

This news article was automatically translated from Dutch to give Techzine.eu a head start. All news articles after September 1, 2019 are written in native English and NOT translated. All our background stories are written in native English as well. For more information read our launch article.