Circle Trumps Square
Unique pattern-recognition software aids pathologists in identifying digital features
From music to medical records, digital technology is revolutionizing nearly every corner of society — and the field of pathology is no exception. Whole slide imaging, for example, can transform a single tissue slide into an image 200,000 pixels wide and 100,000 high — or about 50-100 times larger than a high-resolution beachscape one might set as a desktop background.
U-M pathologists Ulysses Balis, M.D., and Jason Hipp, M.D., Ph.D., were not only confronting questions about how to sift through such an overwhelming cascade of data, but how to harness its potential in ways that would be undreamed of using traditional methods.
Enter SIVQ. Spatially Invariant Vector Quantization is a unique pattern-matching software designed by Balis, Hipp and their collaborators. The program can quickly and accurately identify features within a digital picture, far exceeding what the human eye can do unaided.
With a few clicks, the algorithm can recognize microorganisms, separate tumors from background tissue, or identify cell types unique to a particular diagnosis — such as the cherry-red nucleoli of Reed-Sternberg cells associated with Hodgkin’s lymphoma. It can also be used to rapidly calculate the area of an irregularly shaped feature or to eliminate the slow and painstaking tallying of tiny elements. “Three things set SIVQ apart from other pattern-matching programs,” says Balis, director of the Division of Pathology Informatics at the U-M Medical School and associate professor of pathology. “It’s extremely flexible, requires minimal training to operate, and it’s based around the circle rather than the traditional square.”
While the difference between a circle and a square may not sound like much, when it comes to pattern matching, it’s fairly revolutionary. Circles have continuous symmetry. Unlike a square, if you rotate a circle 45 degrees, the shape remains the same. So as the program searches an image for a particular pattern, its rings also spin like a combination lock, checking every possible degree of rotation.
“What that means is that it will efficiently find the image no matter how it’s rotated or even flipped, like in a mirror,” says Hipp, a pathology informatics fellow. “That’s good because in pathology, human cells don’t line up all nice and neat. They can face any direction.”
The tool is flexible enough to be used across many disciplines of medicine — recently, a Harvard pathology fellow used SIVQ to analyze gunshot wounds in autopsy photos — but its potential doesn’t stop there. In their first article on SIVQ, published last year in the Journal of Pathology Informatics, the researchers showed how the software was able to home in on parked helicopters in a satellite photo of Baghdad, Iraq. (It can even find Waldo in a Where’s Waldo? picture.)
While Hipp and Balis believe SIVQ is a potential “game changer” for pathologists, it’s meant to augment rather than replace human capabilities.
“By eventually bringing a tool like this into the clinical workflow, we can provide a higher level of expertise that is distributed more widely, and lower the rate at which findings get overlooked by even the most skilled and diligent pathologists,” Balis says. —IAN DEMSKY