Computer Trained By Scientists To Evaluate Breast Cancer
Since the 20’s the way in which breast cancer is evaluated in terms of characteristics and breast cancer classification have remained mostly unchanged. A doctor analyzes a tissue sample under a microscope and writes down the diagnostic. Pathologists examine the breast tissue sample and the scores the tumor according to a classic scale that has remained largely unchanged since 1928. These scales are used to asses cancer severity leading therefore to a certain treatment option and also establish the prognosis.
Computer scientists and pathologists from Standford School published a paper yesterday in Science Translational Medicine to report their results in training a computer evaluate and analyze cancer microscopic images tissue samples . The computer results were more accurate than the average doctor analysis.
The computer model, named Computational Pathologist (C-Path), can learn how to analyze microscopic images of breast cancer tissues and determine the patient prognosis.
For C-path training, the research team used already known prognosis tissue samples, and trained the computer to measure and recognize different tumor structures and correlate them with patient survival. Next the computer correlated its results with the known data and adapted its algorithms to better predict survival. In time the computer figured out what cancer features influence more or less the patient’s prognosis.
In simple terms, this computer is a learning student according to Daphne Kolle, PhD, paper senior author and computer science professor.
The medical world uses three cancer cell key features to evaluate breast cancer:
- Tube-like cells percentage of the tumor
- The abnormal nuceli diversity in the periferic epithelial cells
- Cell division rates
These three features are determined subjectively using a microscope and a score is calculated that classifies the patient into a certain prognosis group.
Pathologists are usually trained to search, evaluate and score certain cellular abnormalities that are known for their clinical significance, but tumors present numerous additional characteristics, whose clinical importance has not been evaluated and correlated to a certain prognosis according to Andrew Beck, MD, paper’s first author.
The computer does exactly that and considers several factors in order to determine which can lead to a good or bad prognosis. In fact C-Path, examines and correlates about 6,642 cellular factors. After training the computer on one group of patients, C-Path evaluated breast cancer samples it had not seen before from new patients. The C-Path results was then compared with the known data. The conclusion was that C-Path managed to statistically surpass acreage pathologist evaluation.
The results of the Standofrd scientists team are of major importance as having a “smart and continuously learning” computer that can evaluate tumors will bring pathology accurate diagnosis to undeveloped countries, improving therefore the cancer treatment and overall survival rate.