Dr. Jason Hipp and his colleagues recently were selecting images from digital prostate cancer slides for an upcoming article in a scientific journal. The conversation turned to just how subjective the process can be.
By adjusting the level of magnification and the field of view, a researcher can present an image in a light most favorable to her work, while a different view might open the possibilities for alternative interpretations.
A pilot project at the U-M Health System aims to avoid this type of “selection bias” by making entire digital slides available for review in a public repository.
Having full images during the peer review process would make the review of new scientific studies more rigorous, because reviewers could see more than just the images pre-selected by the authors, says Hipp, an informatics fellow and clinical lecturer in the Department of Pathology at the Medical School.
“This type of transparency benefits everyone: researchers, pathologists and, ultimately, the people who come to the hospital seeking treatment for cancer and other illnesses,” he adds.
Digital imaging is becoming more widespread and the vision is for the repository to also be used in medical training and for side-by-side testing of new software tools designed to analyze digital pathology images.
Earlier this month, the group’s proposal about the need for such a repository was presented at the annual American Society for Clinical Pathology conference in Las Vegas. They also published an editorial on the topic this summer in the American Journal of Surgical Pathology, a distinguished journal that publishes relatively few articles about digital imaging issues.
The repository is a place where researchers can upload digital slides known as “whole slide images” (WSI), the product of scanning an entire glass slide into a digital image. These generally are very large files, ranging from 200MB to 1GB and contain about 100,000 x 200,000 pixels — about 50-100 times larger than the high-resolution images of beaches and mountainscapes many people use as backgrounds for their desktop computers.
Along with slides, the repository website includes a section for researchers to share and test different software tools that are being developed to analyze the massive amount of data contained in WSI sets.
“You could run two programs for calculating the area of a certain feature on the same cancer slide and compare the results,” Hipp says.
Although the U-M pilot site is operational, the researchers believe a truly open and useful repository would require the support of large pathology organizations, as well as the endorsement of scientific journals, editors and from the pathologists who would ultimately be using it.
“Many of the agencies funding scientific research are already requiring researchers to make their primary data publicly available; it’s probably only a matter of time before they start making similar demands regarding images,” says Ulysses Balis, director of the division of Pathology Informatics at U-M and the senior researcher on Hipp’s team.
