Storage solutions must meet high-volume need

Medical imaging storage solutions need to work smarter and faster, to meet the rapidly expanding needs of high-volume radiology departments. That's the assessment of Robert Cecil, network director of radiology and cardiology at the Cleveland Clinic.

Prefetch algorithms waste time and bandwidth. What's needed are archive systems that can classify exams based on the likelihood that they will be accessed in future, he said, at the Society of Imaging Informatics in Medicine meeting in Austin.

For every prefetched prior exam opened at a workstation for review, 10 will never be utilized. Cecil predicted that the current trend of small and mid-sized PACS to retain all procedures online will someday be the norm, but it isn't yet, especially with large, high-volume hospitals. He disagrees with vendors who suggest that archival data is rarely accessed.

"The vendor assumption that maintaining one to three years of images online is adequate is an inaccurate one. Old data is not equally valued," he said.

For example, he said, for the average inpatient, diagnostic procedures performed on the first day will be referenced more often than those conducted on the fifth day of a hospital stay. By differentiating the value of old data, it's possible to achieve more cost-effective storage without compromising retrieval of images that are less likely to be regularly reviewed.

It should be possible to differentiate among the types of cases that yield images most frequently accessed once archived. These include mammograms, radiology oncology procedures, and heart patients with stents. A doctor might not access these images for years, but they should be readily accessible at a predictable time. Heart patients with stents tend to visit their cardiologist three years after the procedure, he said.

Intelligent algorithms of the future should be designed to weight the value of old archival images and retrieve those most likely needed for review. This capability is especially applicable for procedures with large data sets because only a handful of images may be needed.

While innovations in storage technology have enabled disk capacity to double every 18 months, medical imaging demands are moving faster, he said. The data storage industry is primarily driven by the consumer market. In most enterprise IT scenarios, the number of transactions that can take place in an hour is more important than the number of megabytes that transfer in an hour. Megabyte-intensive procedures of large image data sets have the opposite requirements.

"If you really need to move large data sets out of a storage archive, how to move them and where they are going to be put may become an increasingly difficult problem," Cecil predicted.

In addition, he said, archive retrieval speed may not have anything to do with the transfer speed of the media, he said, noting that multiple variables contribute to slow image transfer.

"Delays in the pipeline all contribute to delays in retrieving data. The key is to know what these are, and to make data available in a consistent manner," he said. "If a radiologist knows that retrieving an archived prior will take 10 minutes, he will probably do other work during that time. What is aggravating is inconsistency--whether an archived image will arrive in 10 minutes, 30 minutes, or sometime in between."

In addition to smart archive algorithms, a multivariate storage solution is needed, he said. Such a solution should be based on a balanced mix of disk and tape magnetic technologies managed by appropriate hierarchical storage management software.