If a radiology scan suggests a mass that can be concerning for something serious (of even if not,) what comes next is taking tissue from that mass and examining it. When a cancer is suspected by a radiologist, a biopsy is taken, and that lump of tissue is sent to a pathology lab. There, a team processes the sample into glass slides for examination by a pathologist. These scientists often sit hunched over microscopes for hours, relying on years of pattern recognition to see if the images before them contain cancerous cells.
The field of pathology is critical to both the diagnosis and treatment of many cancers, yet the tools of this trade haven’t been upgraded much in 100 years. We still turn biopsies into physical slides, have pathologists view them on microscopes for hours on end, and rely on memory to find problems. By comparison, radiology went through a digital revolution 20 years ago and now digital X-rays enable remote readings and tele-radiology (including overnight readings from across the globe.
If discrete digital imaging files are the initial frontier for AI in healthcare, pathology could be one of the first specialties to benefit. Think about it, you are looking at tissue samples to find subtle abnormalities that can indicate disease and the shape of those abnormalities can tell you what may be causing the disease or the nature of the severity or aggressiveness of the cause. It seems tailor-made for AI! That is correct and pathology, much like radiology, is about interpretation of pictures. However, there is a difference between those. While radiology images are pure pictures created by x-ray or MRI, pathology images are actual tissues samples taken from the patients. This tissue needs to be prepared with certain material and then viewed under the microscope. Well, that means that those slides are not necessarily photographed or digitized to create a structured data file for AI to examine.
One of the main reasons pathology lagged behind in the move to digital was the high cost of digitization of slides. Each “slide” has to be incredibly high resolution so that it can be zoomed in on and searched, like a detailed satellite image. So the technology required for pathology image storage has been one that’s needed to catch up with the market need.
