Skip to main content

Pixels and Pathology

Shrinking vacation photos before emailing them is easy. Shrinking prostate pathology images while retaining the data crucial for making a cancer diagnosis is not so easy.

A UC San Diego team is building an automated system to do just this. Their computer vision and machine learning application analyzes low resolution images of prostate slides and selects just the parts of the images that are relevant for diagnosing cancer. These crucial regions are scanned at high resolution while the remaining 80 to 90 percent of the image is discarded, which decreases the space needed for storage and the time for network transmission by a factor of ten. Researchers from the computer science and electrical engineering departments at the Jacobs School and from the School of Medicine built the system, which benefits from the award-winning Adaboost algorithm invented by Jacobs School computer science professor Yoav Freund and Princeton's Robert Schapire.

Of the prostate pathology image above, only the pink areas in the image below are relevant for cancer diagnosis and should be scanned. Electrical engineering graduate student Mayank Kabra, computer science professor Yoav Freund and professor Stephen Baird from the UCSD Medical Center are collaborating on an automated system that makes these kinds of determinations.
Of the prostate pathology image above, only the pink areas in the image below are relevant for cancer diagnosis and should be scanned. Electrical engineering graduate student Mayank Kabra, computer science professor Yoav Freund and professor Stephen Baird from the UCSD Medical Center are collaborating on an automated system that makes these kinds of determinations.

Print Article