Beth Israel and Harvard Medical Develop AI Method To Detect Breast Cancer

A team of researchers from Beth Israel Deaconess Medical Center in Boston along with researchers from Harvard Medical School recently developed an artificial intelligence (AI) system to help pathologists with making more accurate diagnosis of breast cancer. In a test during a scientific meeting, the system has shown to boost human accuracy to 99.5 percent.

For the past 100 years, pathologists have been largely diagnosing disease by manually reviewing images under a microscope. The new method suggests that computers can help doctors enhance precision and notably alter the way cancer and other diseases are diagnosed.

Pathologist Andrew Beck, MD, PhD, Director of Bioinformatics at Beth Israel and an Associate Professor at Harvard Medical School explains:

"Our AI method is based on deep learning, a machine-learning algorithm used for a range of applications including speech recognition and image recognition This approach teaches machines to interpret the complex patterns and structure observed in real-life data by building multi-layer artificial neural networks.”

Dr. Beck led the team, which included post-doctoral fellows Dayong Wang and Humayun Irshad, and students Rishab Gargya and Aditya Khosla of the MIT Computer Science and Artificial Intelligence Laboratory.

The team recently competed at the 2016 meeting of the International Symposium of Biomedical Imaging, which involved analyzing images of lymph nodes to determine whether or not they contained cancer cells. Dr. Beck’s team placed first in two separate categories in the competition, where they challenged private companies and academic research institutions from around the globe.

During an evaluation, researchers were provided with slides of lymph node cells and urged to diagnose whether or not they contained cancer. The automated diagnostic method developed by the team proved accurate almost 92% of the time matching the success rate of a human pathologist, whose results were accurate 96% of the time.

"But the truly exciting thing was when we combined the pathologist's analysis with our automated computational diagnostic method, the result improved to 99.5 percent accuracy," said Dr. Beck. "Combining these two methods yielded a major reduction in errors."

One of the organizers of the competition, Dr. Jeroen van der Laak, who leads a digital pathology group at Radboud University Medical Center in the Netherlands, says the findings is a clear indication that AI is going to change the way pathologists use images in the years to come.