Science Translational Medicine: Computer Software That Spots Cancer
New imaging software may rival the eyes of a pathologist, according to a study in the 9 November issue of the journal Science Translational Medicine.
Researchers created the computer program, called Computational Pathologist or “C-Path,” to scan microscopic images of breast tissue for over 6000 characteristics of cancer. The software helped predict breast cancer severity in two groups of women, and could be a useful tool for gauging a patient’s chance of survival.
Since the 1920s, pathologists have mostly relied on the same small set of features to spot abnormalities in tissue samples. Daphne Koller of Stanford University and colleagues developed C-Path with the goal of identifying additional features of cancer tissue that could help paint a more accurate picture of survival outcome.
This image shows 18 breast cancer images adjacent to matching images that have been automatically labeled by image processing software. This region and object detection step is an important component of the process that allows the computer to provide quantitative interpretations of breast cancer microscopic images. | Image © Science/AAAS
Identifying tissue types as epithelial (tissue that lines the outer surfaces of the body and organs, like skin) or stromal (the connective tissue that supports epithelial tissue) is one important part of cancer diagnosis. This process required a bit more work for the C-Path researchers. The team had to teach the computer program how to spot each tissue type using hand-marked samples.
When they tested C-Path on tissue samples from a group of patients in the Netherlands, the software found a set of brand-new features associated with a poor chance of survival, including some stromal features that had not been used previously to predict survival.
“Specifically, the significance of the stromal, non-cancer tissue that surrounds the cancer cells supports the emerging view of cancer as more than a few cells gone awry, but as an entire ecosystem involving interactions between multiple cells,” Koller said.
In a separate group of patients from Vancouver, Canada, C-Path predicted the chance of survival in women based on a comprehensive set of known and new cancer tissue features.
“C-Path is capable of providing more accurate predictions of survival than a human pathologist, using the same images. This can be of great value in providing high-quality medical care, especially in parts of the world where trained pathologists are scarce,” Koller said.
The results make C-Path one of the first potentially usable computerized pathology systems, but there are significant limitations in the software that may prevent its immediate use in medical facilities, David Rimm writes in a related Perspective article in the journal. The software’s inability to classify epithelial or stromal tissue without manual assistance, for instance, means that pathologists will have to “retrain” the software for use in new institutions.
Read the abstract, “Systematic Analysis of Breast Cancer Morphology Uncovers Stromal Features Associated with Survival,” by Daphne Koller and colleagues.