Teaching Computers to Tell Cancer Cells Apart
Prostate cancer is notoriously difficult for pathologists to reproducibly read and interpret. It is a jumble of cell types in the tissue – normal cells mixed in with not just cancer cells, but several different kinds of cancer cells. Prostate cancer cells have different Gleason pattern numbers assigned based on their morphology – how their size, shape and orientation appear under the microscope. Scientist Robert Veltri, Ph.D., working with a biomedical engineer from Case Western Reserve University, Anant Madabhushi, Ph.D., has come up with a way for a computer to help. Veltri has been studying the structure of the nucleus of prostate cancer cells for years, and that expertise is now part of an automated tissue-studying program that uses seven key structural features to analyze Gleason grade. "The computer program needs only three out of seven features to distinguish with 90 percent accuracy Gleason grade patterns 3 and 4, and to differentiate aggressive from non-aggressive prostate cancer," he says.