DICOM meets AI

“There’s still life in the old dog yet”: so say the British when something old has not yet outlived its usefulness. Although he expressed it in more detail and with respect to DICOM and related standards, this is precisely what Dr. Marc Kämmerer, innovation manager at VISUS, said in his talk entitled “Medical Data and Where to Find It,” which he gave at the Emerging Technology in Medicine (ETIM) event held in February of this year at the University Hospital in Essen, Germany.

This was the fourth time that Essen University Hospital had invited prominent experts from science and industry to offer a look at future methodologies and technologies in medicine. A focus of this year’s event was on artificial intelligence and its potential for diagnostics and treatments going forward.

Consolidation strengthens communication

The scenarios outlined here were impressive. If artificial intelligence and its algorithms deliver even half of what scientific studies currently suggest, then we are truly on the cusp of a revolution in diagnostics. That promise came with an appeal for a consolidated data platform based on the healthcare content management (HCM) concept—in other words on standards actually presumed to be antiquated, such as DICOM. The present challenge, however, is first to develop lovely new algorithms, feed medical data into them and train them. That may not sound exciting, but it is an essential foundation for any efforts surrounding artificial intelligence. On a side note, consolidating an institution’s medical data is not just relevant for the future—as Kämmerer emphasized, it has highly concrete applications in the present as well. Data need to be consolidated, for example, to make a concise overview of all treatment-related information available in one system and on a single screen. It also comes into play when optimizing communications with external parties such as patients or the medical services departments of health insurance providers. In scenarios like these, consolidating medical data on the basis of HCM principles would work like this: an HCM system would receive data, regardless of original format, from medical subsystems and convert these to the internationally recognized DICOM standard. Once converted to a standard format, these medical data could then be sorted, filtered and communicated through the application of additional standards such as HL7 and/or IHE integration profiles.

DICOM talks to everyone about everything

“In order for consolidation to enhance the value of data, the data have to meet the following requirements: you have to be able to use, access and communicate them, and they have to be reliable, consistent and secure,” Kämmerer noted in his talk. He then added, “And we have an established format that meets all of these prerequisites for smart medical data management: the DICOM standard. When it comes to data consolidation, the advantages of DICOM are that it can be used for sending a file’s metadata and, if desired, for saving the raw data as well.  Plus, DICOM has an equivalent for nearly every format. It has the DICOM Structured Report for text, for example, and the Native DICOM Object for biosignals.”
Consolidation requires categorizing data, which directly enhances their utility for health care institutions. Categorization allows users to apply smart filters to treatment data, for instance, showing only the data users actually need for the workflow at hand.

Standards are adapting to progress

In terms of developments surrounding artificial intelligence, however, Kämmerer’s presentation had a far more important message for his audience: DICOM—like other standards—is not a fixed, rigid structure. It is instead a dynamic object undergoing continuous development. “Sharing medical data is the ultimate aim of developing AI solutions. First, you have to communicate clinical data to an algorithm in order to train it. But then the results from the algorithm have to flow back into the process as well. And DICOM and other related systems support that communication process, just as they already support communication within health care institutions. They’re continuously updated and developed to that end, which, in practice, is an astonishingly fast process.” As an example of this kind of adaptation to new situations, he cited the publication of DICOM’s Supplement 219, which was available within one year: “This was what we call a structured report object, which defines how data are shared between two programs. The Supplement, in other words, supports an important use case for artificial intelligence. My message to AI solution developers—and to health care institutions—is therefore that, in the rush to AI, we shouldn’t lose sight of the standards! Using and adapting current standards will make data suitable for future applications and demands, where they will bring added value to the table going forward as well.”