researchers have "taught" computers to analyze microscopic breast cancer
cells, and an early study shows they are slightly more accurate than
human pathologists in predicting patient survival. The computer analysis
lends support to the idea that cancer is an ecosystem and that the
entire microenvironment of a tumor needs evaluation.
A Stanford website story by Andrew Myers
explains that computer scientists at the Stanford School of Engineering
and pathologists at the Stanford School of Medicine collaborated to
train to analyze images of malignant tumors. Their model is called
C-Path, for Computational Pathologist, and the findings are published in
the Nov. 9 issue of Science Translational Medicine. Daphne Koller,
professor of computer science and an author of the paper, said the
computer "learned" by poring over images of tissue samples taken from
patients with known prognoses. Tissue sample images from two groups of
breast cancer patients, 248 patients from the Netherlands and 328
patients from Canada were analyzed.
story points out pathological analysis of breast cancer hasn't changed
much since it was developed in 1928. Pathologists have been trained to
look at what percent of the tumor is made up of tube-like cells, the
diversity of the outermost cells of the tumor, and the frequency with
which the cells divide. These factors are then scored to assign patients
into three groups that predict survival rates. C-Path looked at 6,642
cellular factors, and found that characteristics in the cells
surrounding the tumor were important in predicting survival.
built a model based on features of the stroma - the microenvironment
between cancer cells - that was a stronger predictor of outcome than the
one built exclusively from features of the epithelial (outermost)
cells," said Andrew Beck,
an assistant professor of pathology at Harvard Medical School who
participated in the Stanford research when he was a doctoral student.
learning, we are coming to think of cancer more holistically, as a
complex system rather than a bunch of bad cells in a tumor," said Matt van de Rijn,
pathology professor and co-author of the study. "We're looking at a
future where computers and humans collaborate to improve results for
patients across the world."
Koller, described in Newsweek
as "on the bleeding edge of the leading edge" in artificial
intelligence, thinks that eventually, through human-computer
collaboration, computers will learn to identify the best course of
treatment for an individual patient; they might also predict whether
nonmalignant cell samples might turn cancerous, so that many patients
could avoid cancer.