Harvard, MIT researchers harness deep learning for more accurate breast cancer pathology

In an evaluation of slides of lymph node cells, the automated diagnostic method was accurate about 92% of the time.  This was almost as accurate as human pathologists–who are about 96% correct. But when combined–the results were even better.  “The truly exciting thing was when we combined the pathologist’s analysis with our automated computational diagnostic method, the result improved to 99.5% accuracy,” said pathologist Dr. Andrew Beck, Director of Bioinformatics at the Cancer Research Institute at Beth Israel Deaconess Medical Center and an associate professor at Harvard Medical School. “Combining these two methods yielded a major reduction in errors.”

Beck and team member Aditya Khosla of the MIT Computer Science and Artificial Intelligence Laboratory have formed a startup company aimed at the application of artificial intelligence to pathology.  Increasingly, the industry expectation is that deep learning can be useful to aid human analysis–becoming a first step to pre-analyze and learn from massive quantities of data to offer insights that are subsequently reviewed by people.  “Our AI method is based on deep learning, a machine-learning algorithm used for a range of applications including speech recognition and image recognition,” said Beck.  “This approach teaches machines to interpret the complex patterns and structure observed in real-life data by building multi-layer artificial neural networks, in a process which is thought to show similarities with the learning process that occurs in layers of neurons in the brain’s neocortex.”

The team used a deep, multilayer convolutional network approach to train the computer to distinguish between cancerous tumor regions and normal regions.  They started with hundreds of pathologist training slides that are labeled with regions of cancer and regions of normal cells.

The next step was to extract millions of these training examples and use deep learning to build a computational model to classify them.  The team then went back and identified the particular kinds of mistakes to which the computer was prone–and then retrain it with more difficult training examples to continue to refine its performance.  “There have been many reasons to think that digitizing images and using machine learning could help pathologists be faster, more accurate and make more accurate diagnoses for patients,” Beck added.  “This has been a big mission in the field of pathology for more than 30 years. But it’s been only recently that improved scanning, storage, processing and algorithms have made it possible to pursue this mission effectively.”

REFERENCE:  Fierce Medical Devices; 21 JUN; Stacy Lawrence

Leave a Comment