Siegel at SIIM: Big data + personalized medicine = big opportunity for imaging

DALLAS—Personalized medicine will completely revolutionize the way we do diagnostic imaging, Eliot L. Siegel, MD, chief of radiology and nuclear imaging at VA Maryland Healthcare System in Baltimore, predicted at the Society for Imaging Informatics in Medicine annual meeting. Despite the potential, imaging is not even close to ready for the era of personalized medicine, he continued.

Siegel painted a harsh picture of the future. “In the era of big data and personalized medicine, diagnostic imaging runs the risk of becoming a relatively invisible non-participant in medicine if it cannot be delivered in a format that can be consumed as part of big data.”

However, he also detailed a different course. Rather than bordering on the irrelevant, imaging can play a central role in medicine’s future, with each imaging study considered a physical examination that generates critical data related not only to the current clinical question but also to other aspects of the patient’s health.

In the future, a CT of the thorax may be performed to determine if a patient has a pulmonary embolism; the imaging report will answer that question. The study also generates a host of additional data about the patient such as bone density of spine and coronary artery calcification. These data will be discoverable as needed via algorithms.

Personalized medicine may spell possibility for imaging in other ways, predicted Siegel. Take, for example, oncology imaging. One challenge is to determine the molecular pathway for cancer rather than its diagnosis or appearance at pathology/histology. The current protocol relies on biopsy, but it may be possible to use diagnostic imaging to predict the molecular pathway noninvasively. “Biopsies for DNA tumor information will result in an increased number of image-guided interventional cases.”

Siegel also foresees an increased number of screening studies based on clinical, imaging and genomic/proteomic risk factors with emphasis on wellness.

The practice of personalized medicine

Personalized medicine requires massive amounts of data about millions of individuals so physicians can locate information on a small handful of similar patients. These data originate in multiple sources, creating a tremendous challenge and the need for an incredible amount of computing power, said Siegel. Hence, artificial intelligence is a prerequisite. Siegel shared a few stats for skeptics:

  • Medical information doubles every five years;
  • 90 percent of current data was created in the last two years;
  • 80 percent of these data are unstructured; and
  • The shortage of physicians will hit 95,000 by 2020.

Enter Watson

Unlike a human physician, Watson, IBM’s artificial intelligence technology of Jeopardy fame, can process 500 gigabytes, approximately one million books, per second.

However, informatics challenges continue to haunt even Watson. Unstructured data, fragmented EMRs and the lack of integration of changing signs and symptoms over time in clinical information systems all pose obstacles.

Imaging informatics stakeholders can help set the stage in multiple ways, said Siegel. They can facilitate the translation of medical guidelines into machine-intelligible form, generate radiology data in a format that can be consumed by guidelines algorithms and they can lobby for structures to share large datasets, such as the Alzheimer’s Disease Neuroimaging Initiative and Pediatric Brian Tumor Consortium.

Siegel suggested radiology can prepare for the era of personalized medicine and decision support by adopting a hybrid approach to reporting that combines structured reporting and natural language processing. Structured reporting carries the advantage of creating structured data, while natural language processing offers more freedom. The hybrid model may allow radiologists to leverage the best of both approaches.  

Siegel concluded with five challenges to be overcome if radiology is to seize the big opportunity of personalized medicine. These are:

  • Lack of acquisition standards;
  • Lack of image quantification tools, which makes it difficult to measure lesion size and change over time;
  • Lack of an imaging tagging standard to translate image markups and annotations among workstation;
  • Lack of mechanisms to share images and data from clinical trials; and
  • The patient-centric configuration of the EMR, which makes data mining difficult.