Early medical image archiving was focused on one main goal: the display of images. No easy feat, especially as storage requirements grew and physicians began to expect images on-the-go through their tablets or phones, but the mission of sharing images remained clear.
Today, however, some are looking ahead at the next generation of archiving and getting beyond the display, to take a deeper dive into image data—and that means more than simply reading imaging reports.
“Today when people are talking about big data and data mining, they are still talking about text,” says Gary J. Wendt, MD, vice chair of informatics, professor of radiology, and enterprise director of medical imaging at the University of Wisconsin-Madison. “They’re not talking about actually mining content out of images. I think that’s probably the next generation, actually processing image data, not just text data.”
Wendt sees a future PACS that doesn’t only find and present images, but automatically performs more complex tasks such as plotting volumetrics of various nodules or describing other characteristics of disease.
“Even in Imaging 3.0, there’s not a lot of automation and aggregation of data,” says Wendt. “There’s been some work on aggregation of text data—digging stuff out of the EMR and past reports—but there’s really not been a lot done with automated image processing.”
Wendt’s comments echo those of Eliot Siegel, MD, with the University of Maryland School of Medicine, who is another champion of bringing big data to medical imaging. In a January webinar hosted by the Society for Imaging Informatics in Medicine, Siegel likened the amount of data embedded within an image to that of dark matter in the universe, which lingers unseen even though we know it’s there.
“There is so much in our images that we just aren’t aware of because the images are untagged and not mineable. All of that information we’re not using is like universal dark matter—it’s vast,” he says. “This must change if medical imaging is going to play a substantial role in this era of big data, medical guidelines, decision support and personalized medicine.”
Siegel says one goal should be identification of molecular pathways for cancer rather than simply its diagnosis or appearance in one pathology. Likewise, Wendt looks forward to the day where he can query the system for a similar cohort of patients—say, patients with brain tumors who are three to six months post-diagnosis, for instance.
“Give me a relevant cohort of priors, not just of the same tumor type, but the same tumor type at the same time post-diagnosis,” says Wendt, adding that this method would allow for better comparisons.
Ultimately, the clinical impact of such next-generation image archiving would come from the creation of more relevant reports. This would be especially beneficial in oncology, where treatments can be modified based on tumor progression, and comparisons to similar cohorts of patients at an oncologist’s fingertips would be useful.
“That’s going to take a whole new type of analytics,” says Wendt. “It’s going to be a whole new world.”