Ai Detects Breast Cancer Before Doctors!
Imagine this: an algorithm reviews recent scans of your body in an effort to identify cancer. If there is cancer, another algorithm examines your genetic makeup and generates a customized treatment plan. Simultaneously, a third algorithm reviews your health history and determines your prognosis. This computational trio of doctors doesn’t get tired, isn’t particularly concerned with pay, and is far more accurate than even the best human physicians.
This is the future of medicine, and it’s being made possible by artificial intelligence and machine learning. AI will impact nearly every facet of medicine and has the potential to dramatically increase our quality of care and disease survivability.
One such example is a recent breast cancer diagnosis tool developed by researchers at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL). This tool, which uses deep learning, was trained on over 90,000 mammograms and can recognize disease patterns that are imperceptible to human doctors. Amazingly, it can predict the development of breast cancer up to five years in advance from just a single mammogram.
Deep Learning can detect cancer that is impossible for a human eye to observe!
One of the major focuses of this project was the inclusion of “equitable” data. An issue increasingly in the spotlight, oftentimes AI models are trained on datasets that are skewed towards a specific demographic. For example, Amazon has received flack for inadvertently creating a hiring AI that discriminates against women, simply due to the fact that it mostly received men’s resume data for training. Similar challenges often pop up with facial recognition software, which tends to perform much more poorly on minorities. In this instance, developers paid careful attention to their dataset, including more minorities leading to equal performance for both black and white women. This is particularly important, since black women are currently 42% more likely to die from breast cancer.
Though this model has yet to be launched in a true healthcare setting, its performance is noteworthy and certainly indicative of good things to come. It’s important not to let important developments like this stall in the research stage for years; these models must be deployed and put into practice quickly so that their impact can be realized as soon as possible.
Written by Daniel DiPietro, Edited by Matthew Durborow & Alexander Fleiss