Connie Lehman, MD, Ph.D., Diagnostic Radiologist, Chief, Breast Imaging Massachusetts General Hospital speaks about These Doctors Are Using AI to Screen for Breast Cancer.
Link to Article:
CONSTANCE LEHMAN WAS Persuaded TO Adjust THE WAY MASSACHUSETTS GENERAL HOSPITAL SCREENED WOMEN FOR BREAST CANCER After COVID ARRIVED IN MASSACHUSETTS. Owing to concerns about the infection, several people skipped routine checkups and scans. As a result, the Lehman-coordinated center started using an artificial intelligence algorithm to determine who is most likely to contract cancer.
According to Lehman, the AI technique has assisted in identifying a number of women who, after being advised to come in for annual screening, have early symptoms of cancer. The algorithm-identified women were three times more likely to contract cancer than women who were not identified by the algorithm; previous predictive approaches were little better than chance.
The algorithm looks at previous mammograms and seems to operate even though doctors didn't find any warning signs in those tests.
Researchers have long lauded AI's promise in medical imaging, and several instruments have even made their way into clinical practice. For several years, Lehman has collaborated with MIT researchers on how to adapt AI to cancer screening.
However, AI has the ability to be much more useful in terms of predicting danger with greater accuracy. Breast cancer screening may entail more than just looking for cancer precursors on a mammogram; it may also entail gathering patient data and feeding both into a mathematical model to assess the need for follow-up screening.
Before Covid, Adam Yala, an MIT Ph.D. student, started designing the algorithm Lehman is using, named Mirai. According to him, the aim of using AI is to increase early detection and reduce the stress and costs associated with false positives.
Yala had to tackle obstacles that had stymied previous attempts to use AI in radiology to establish Mirai. To allow for disparities among radiology devices, he used an adversarial machine learning technique, in which one algorithm attempts to fool another. This may suggest that patients with the same risk of breast cancer get different ratings. The model was also built to combine data from several years, making it more reliable than prior attempts that used fewer data.
The MIT algorithm examines the usual four views of a mammogram, inferring detail about a patient that is sometimes overlooked, such as surgical history or hormone causes like menopause. If the information hasn't already been gathered by a doctor, this may be useful. The thesis is described in detail in a paper published today in Science Translational Medicine.