University Radiology Group: A Common Archive for a Distributed-reading Solution

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon

Alberto Goldszal, PhD, MBAArchiving and distributing the large data sets associated with images can be a challenge for many radiology practices. University Radiology in New Brunswick, New Jersey, would encounter a larger-than-usual share of obstacles on this front if it were not for a carefully thought-out image-archiving and -distribution strategy built upon the unique attributes of its PACS technology.

University Radiology employs more than 90 radiologists and reads more than 950,000 exams per year—about 650,000 for six hospitals, with the remaining exams done at its 10 imaging centers. The group’s clients are independent, unaffiliated locations that are geographically distributed throughout New York, New Jersey (its primary market), and Pennsylvania.

University Radiology also has an in-house night-coverage service, with radiologists reading from an even more geographically disparate area that extends from Illinois to Washington to California—and across the world to Germany and Israel. About a million studies are added to the group’s image archive annually.

Avoiding Failure

Alberto Goldszal, PhD, MBA, is University Radiology’s CIO. He notes that in deciding how to configure the archive, several years ago, he and his colleagues had one clear objective in mind—to avoid any single point of failure and ensure access to images 24 hours a day, 365 days a year, without breaking the bank. Implementing a redundant image archive was deemed the best way to attain the required failover capability.

The archive is a component of University Radiology’s Synapse® PACS (FUJIFILM Medical Systems USA, Stamford, Connecticut). Instead of being stored in only one repository, replicated images reside in three locations: online spinning disks at University Radiology’s primary data center in East Brunswick, at a second facility in North Brunswick (about 10 miles away), and on tape (offline).

Ensuring that radiologists and referring clinicians would be able to obtain images rapidly from at least one source (despite any technological difficulties elsewhere) was critical, but University Radiology did not want to shoulder the financial and logistical burden of adding hardware and grappling with different data silos to support the multisite structure.

“With the Synapse application, this was not a concern; there was ample flexibility to set up the archive for multiple sites,” Goldszal states. He adds that because the archive component is vendor neutral and operates independently of the PACS application itself, University Radiology will avoid cost- and-time-intensive data migration, should it deploy a different PACS in the future.

University Radiology needed to structure the system so that it would meet physicians’ demand for access to all relevant prior studies from the sites that it covers. “We have a lot of site-to-site overlap,” Goldszal observes. “Sometimes, it’s because of insurance coverage; sometimes, it is to see a particular specialist; and sometimes, it’s a matter of proximity, but for us, it is not unusual to find images of a certain person at a few different institutions that may be just a few miles from each other. It was imperative to address that disparity.”

This is where probabilistic matching, which is embedded in Synapse under the umbrella of a feature called CommonView, comes into use. “Instead of having a universal medical number assigned to them alone, patients in the United States have a medical-record number from every health-care provider they see,” Goldszal explains. “It always complicates the process of aggregating and retrieving information, including images, but it is even more of a challenge when that information needs to be retrieved from unaffiliated sites.”

To overcome this problem, he explains, CommonView determines the probability that the records for John Smith from site A or hospital B belong to the same John Smith who underwent another imaging procedure at site C or hospital D. Calculations are based on weights assigned to various attributes, such as a patient’s home address or age.

“We can, for example, take the first four letters of the last name, the date of birth, and the gender, and give that a matching probability expressed in percentages,” Goldszal elaborates. He adds that a wizard offers feedback on how to assess the viability of each match—given the selected attributes—and reports that very few (if any) PACS feature a comparable capability.

Playing Fetch

University Radiology’s overall image-retrieval and