An IBM algorithm combining machine and deep learning to analyze health records and mammograms was able to predict the development of breast cancer up to 12 months before its onset with nearly 90 percent accuracy, according to a study published June 18 in Radiology.
Scientists from the health informatics department of IBM Research’s laboratory in Haifa, Israel, developed the artificial intelligence model.
It was trained on the mammograms and health records of almost 10,000 women to predict biopsy malignancy and assess whether screening exams were normal or abnormal.
When applied to images collected from more than 13,000 women who had health records dating back to at least one year before undergoing a mammogram, the algorithm predicted biopsy malignancy with 87 percent accuracy.
The AI also identified breast cancer in 34 of 71 women who had initially been determined to be cancer-free by a radiologist but had then developed cancer within one year.
According to the study’s authors, the algorithm achieves breast cancer detection results better than those of the popularly used Breast Cancer Risk Assessment Tool, or Gail model, and equal to those of trained radiologists.
It also identified new factors linked to breast cancer but not currently included in risk assessment scores, including white blood cell profiles and thyroid function tests.
News of IBM’s algorithm comes just weeks after researchers from the Massachusetts Institute of Technology and Massachusetts General Hospital published a study detailing a deep learning model of their own that predicts the development of breast cancer up to five years in advance, based only on mammography.
