When a group of researchers considered creating a library of brain scans that would be freely available to peers around the world, they weren't sure they could make it happen. They thought other researchers would balk at sharing their work. And even if scientists did contribute to the database, the technological hurdles were daunting. They needed systems to process and archive the data, and then make it available securely while protecting patients' privacy. A nice idea, the more charitable skeptics said, but not workable in the real world.

Happily, the doubters were wrong. Dartmouth College's Functional Magnetic Resonance Imaging Data Center (fMRIDC), an electronic archive of human brain scans, went live in 2001 and is meeting its goal of providing free access to a broad collection of brain scan data to researchers, thanks to the perseverance of Dartmouth's Department of Psychological and Brain Sciences and technology solutions from Sun Microsystems.

Now, an upgrade to the Sun Solaris 10 Operating System is enabling the expansion and enhancing the performance of fMRIDC, allowing for greater data storage capacity and sharing with fewer administration headaches.

Setting the Standard

Dartmouth College, a member of the Ivy League, is a private, four-year undergraduate university with graduate schools of business, engineering, and medicine and 16 graduate programs in the arts and sciences. Founded in 1769, Dartmouth is the ninth-oldest college in the United States and blends a long tradition of educational excellence with a focus on leading-edge research.

Functional Magnetic Resonance Imaging, or fMRI, is a relatively new, noninvasive technology that allows researchers to observe brain function directly in real time. It provides great opportunities to advance our understanding of how the brain works, and may become a standard tool for assessing neurological status and neurosurgical risks. But fRMI technology is expensive, which puts it out of reach for many institutions, and the lack of a common data standard means that researchers often cannot work with data collected by those who do have access to the technology.

Dartmouth researchers tackled that issue by creating the fMRIDC, which gives researchers worldwide access to peer-reviewed data from fMRI experiments, says James Dobson, system architect at Dartmouth College. The data is distributed both on CDs and through high-speed networking on Internet2, an alternate worldwide computer network used primarily by scientific and educational institutions.

Dobson beta-tested the Solaris 10 OS last summer, and he says its new features will enable expansion and easier management of the fMRIDC and will also boost performance and simplify administration of the department's entire computer network, all while cutting costs.

Protecting Privacy

The brain scan images stored in fMRIDC come from fMRI results published in peer-reviewed journals. Distributing the results of such diagnostic tests requires that patients' privacy be rigorously protected.

"When data arrives, we remove all traces of personal identity information, and the image files are converted into a standard format so that they can be shared with the entire research community," says Dobson. "To protect data which have not yet been anonymized, we utilize several Solaris features, including strong access control lists (ACLs), firewalls, and Role Based Access Control (RBAC)."

The Solaris 10 OS offers what Paul Sangster, senior Solaris security architect at Sun, refers to as "extra layers of strong security." These layers include:

• A high-performance firewall based on popular IP filter technology that limits which machines can communicate with which medical systems
• "Containers" (aka "zones") that partition the system, keeping patient data separate from system access
• Advanced user and process rights management capabilities
• A framework that allows encrypted standards-based security protection when sharing sensitive medical data

Once processed, the fMRI data is stored together with the necessary technical descriptions so that researchers can study and replicate the original results.

The Brain Behind the Brain Database

The fMRIDC is powered by a high-performance computing cluster environment of Sun Fire V880, Sun Fire 6800, Sun Enterprise 420R, and Sun Enterprise 5500 servers, in addition to a Sun Blade 2000 Anniversary Edition personal visualization workstation.

Software components include Sun Java System Grid Engine software for distributed resource management, Sun HPC Cluster Tools for scalable parallel applications, Sun Studio for application development, and Sun StorEdge Performance Suite software for optimized access to the data, which is stored in Sun StorEdge 3910 and SunStorEdge L700 tape libraries with 15 TB of data storage capacity.

Dobson says the Sun Solaris 10 OS provides a cost-effective, integrated, and interoperable software system that will simplify the task of managing a complex and rapidly expanding scientific environment. He cites the operating system's new DTrace utility, which allows Dartmouth IT staff to perform the equivalent of a brain scan on their computer network.

"DTrace easily diagnoses system problems and reduces the costs and time associated with determining root causes of intermittent problems," says Dobson. "We have used it to diagnose some system problems. It is a neat tool for looking in on the entire system."

The Solaris 10 OS's enhanced architecture will also streamline resource utilization and reduce downtime, Dobson says. And Dartmouth researchers also appreciate the x86 integration with open source software.

"Many of the toolkits and applications used in medical imaging are based on open implementations of various standards," Dobson explains. "These libraries, which, for example, provide support for PNG or JPEG image formats, are often included in each release of an application. Having support for these file types built into the OS means less software to install and manage."

Dobson adds that new features in the Solaris 10 OS such as directory services and tools for storage management have eliminated the need for additional solutions such as the Veritas Volume Manager, an online storage solution, saving the department at least $10,000 a year.

"The integration of production-level storage software is critical for data-intensive science. We collect gigabytes of data a day and need to safely store and process this—often creating an order of magnitude more data," says Dobson. "We needed a simpler architecture of systems and software to reduce the risk of data loss."

Preparing for Expansion

Dobson says the addition of Solaris 10 to the fMRIDC's proven Sun hardware and software infrastructure will also enable Dartmouth's Psychological and Brain Sciences Department to expand and improve the database.

"The data processing paradigm is changing as experiments grow in size. These changes have pushed us to adopt high-performance computing techniques and workflow systems to quickly run pipelines of data through a series of applications," he says.

"We have been using Solaris to manage and store the data and for some of our visualization tasks. With the release of Solaris 10 and the Janus project (providing Linux compatibility) we expect to be able to run more of the processing tools on Solaris."