Diagnostic Imaging
July 2004

Tech Watch

Grid storage adds intelligence to data archiving task

Distributing images over a network saves money and time while augmenting redundancy

Fixed content storage, a requirement of PACS and other medical information systems, can exact a heavy premium if not done correctly. Vancouver, BC-based Bycast wants to spare customers that premium with its management strategy that prioritizes data according to diagnostic value. Less important data are put on lower cost media, saving money and conserving resources.

Finding a lesion on a mammogram, for example, makes that image more important than one that has no such findings. This may seem like common sense, but data storage schemes generally do not apply common sense. Storage decisions are usually based on the amount of time that has elapsed since the image was made. As a result, many data are unnecessarily placed on expensive media for quick retrieval, even though there is little chance they will ever be called up.

"Just look at the cost difference between tape and fast-spinning media," said Deanne Farrar, Bycast vice president of marketing. "The high-performance media cost $50 to $90 per gigabyte. The same storage on tape costs pennies."

Maximizing the availability of high-performance media means that more important data can kept near at hand. Savings in time and money mount up quickly when large quantities of data must be archived. The Provincial Health Service Authority (PHSA) in British Columbia runs the province's four major cancer centers and their related clinics. It coordinates all tertiary-care services: pediatric, cardiac, and renal care, and transplants. It also runs the BC Centre for Disease Control and its screening clinics for tuberculosis, a large women's hospital specializing in complex obstetric cases, and all mammography screening clinics.

Much of the data supporting this care are in the form of images: 26 terabytes worth on the authority's StorageGRID. And it's expanding fast. Don Henkelman, chief information officer for the PHSA, estimates that physicians in British Columbia are adding 15 TB a year. The authority chose Bycast because it needed a storage system that could keep up.

"We needed a deep storage, but one that could expand over time with no disruption in the existing structure," Henkelman said.

The company has signed up several other large healthcare providers in the past year. Among them is Mercy Hospital in southern Florida, one of the nation's largest healthcare systems, with cardiovascular, cancer, and orthopedic centers. Another is the VA Eastern Colorado Health Care System, which encompasses a major referral medical center, nursing home units, and seven community outpatient clinics in southern Colorado.

StorageGRID manages the data at these facilities with interfaces that link to existing PACS, DICOM-compatible imaging modalities, viewing stations, and applications. The biggest advantage of StorageGRID is its transparency, Henkelman said. Users throughout the network can see where their data are and what equipment is connected.

"The imaging systems simply push the images onto the grid, and they are immediately available throughout the whole system," he said.

This feature has a strong impact on emergency and surgical care.

"As soon as the exam is done, the image is immediately available across a diverse geographic area. The advantages come in areas like neurosurgery, where you need teleradiology consults. With StorageGRID, we can immediately move a CT or MR image from a distant location where there is no expertise to one where there is a pediatric neurosurgeon, for example," he said.

The distributed nature of grid storage also provides a built-in fail-safe mechanism. No single node failure can cause a system crash, because StorageGRID replicates data according to rules that distribute the information across the grid. These rules allow hospitals to control the geographic location and number of data "replicas" to achieve the desired level of redundancy. If a node fails, image storage and access requests are rerouted to work around the failure.

The StorageGRID software adapts to changes in the network, assimilating new nodes and excluding those that have been eliminated or replaced. With each change, the grid reassesses the network and optimizes performance, setting the stage for storage capacity that can be expanded with new devices to meet new data demands. The software runs on any hardware platform that uses the Linux operating system.

"We provide the customer with the flexibility to choose the platform on which they want to deploy," said Moe Kermani, Ph.D., president and CEO of Bycast.

These solutions work with all types of storage hardware, so an institution can choose the types of media it wants to use. Expensive, fast-access media such as fast-spinning hard drives are conserved, while less valuable data are stored on the less expensive media. If certain data are in greater demand than anticipated, StorageGRID improves accessibility by automatically relocating those data to a different node featuring faster storage media.

This capability, called SMARTStorage, involves "meta-data" that describe the value of the imaging information. Some of these meta-data are available in DICOM headers, such as modality, body part, and time and date of exam. Others may be input specifically by hospitals using their own rules about where certain types of data should be stored and for how long. Rules might be written to store positive mammograms on fast-spinning disk media for 30 days, then move them to less expensive long-term media. Chest CRs, however, might be retained for 45 days on fast disk before being transferred to less costly storage.

These rules, whether DICOM-based or tailored to the institution, make StorageGRID aware of the value of the data, Kermani said. The need for this type of intelligent management will likely grow in the future with the proliferation of digital imaging and the use of postprocessing that creates 3D and 4D reconstructions.