• E-mail
• Print
• Comment
• Font Size
• Digg
• del.icio.us
• Discuss article

Grid Technology AIDS Hospital's Diagnostics

Posted on: Wednesday, 6 April 2005, 03:00 CDT

In Site: Lessons from leading users

Doctors often need to balance the urgency of medical conditions with performing thorough analysis to arrive at an accurate diagnosis. Ideally, they'd like to do both.

Dr. Eric Bremer at Children's Memorial Research Center in Chicago turned to data mining and grid technology to help his organization garner speedy research results, while also performing thorough analysis.With nearly 3,000 children diagnosed with brain tumors in the U.S. each year and more than 12 brain tumor types, he says it's critical that doctors hit the right diagnosis - fast.

Divide and conquer

Using grid technology the hospital's director of the Brain Tumor Research Program was able to reduce the time it took a text-mining application to run from more than 24 hours to just less than an hour and 20 minutes. Aside from tapping internal resources, Bremersays he expanded his research capabilities and speeded results without adding significant cost, maintenance and IT burdens.

"Our overall goal is to define better diagnostics and therapies for kids with brain tumors. We wanted to lift the human barrier to working with such vast amounts of data within realistic time frames," he says.

Bremer last year rolled out LexiQuest Mine, a data-mining application from SPSS, to start pulling relevant information out of 125,000 document abstracts from 21 medical journals. With more than five years of research at his fingertips, he says he thought LexiQuest Mine would help him more quickly classify brain tumors in patients.

The software lets researchers ask questions, extract concepts and relate the analysis to known brain tumor diagnosis criteria such as gene lists. LexiQuest Mine helped the doctor query and retrieve documents, but more importantly mine the text for concepts. For example, the analytics software understands how groups of words work together such as clinical trial vs. failed clinical trial. Text mining would deliver more accurate results than, say a search program.

The problem with the software is that it took more than a day to generate some results.

"Initially, I thought the application had shut down, but that is just how long it took to process the request from a workstation,"he says."l quickly realized it wasn't humanly possible, even with this application, to realistically mine the amount of data I needed to mine for research."

Bremer then considered whether other technology could help advance his scientific endeavor.

"I am not an IT professional,but 1 was familiar with the notion of high-throughput and high-performance computing options," he says.

He considered grid computing an alternative to investing in high- end servers. Because of cost, space and potential managerial constraints, he wanted to better use the computing power he had in- house.

With the help of SPSS, Bremer became acquainted with grid computing products from United Devices.The application provider wanted to ensure its software met the specifications the doctor required and offered to grid-enable the application. SPSS had been a node on United Devices'worldwide grid since 2001.

"The application has different modes in which it can run, and each mode has a process that is a series of steps, and each step has a wrapper around it. The grid technology can grab the wrapped step and send it to any idle processors on the grid," explains Catherine DeSesa,system engineer at SPSS."Most applications that use command- line scripts can be grid-enabled."

Working with professional services folks from SPSS and United Devices, Children's Memorial rolled out its pilot of United Devices' GridMP Workstation product in October.

The software consists of a central Linux-based server and agents distributed to Windows machines designated to be on the grid. The software taps unused and idle processor time on doctors' and researchers' desktops and laptop computers. Upon a user request, the server software will parse out parts of the LexiQuest Mine application process to available machines. The machines perform their part of the job and return the results to the GridMP Server, which recompiles the data and delivers it to the user via a Web interface.

Policies defined in the software determine which machines can be used and eliminate the need for the researcher to assign the jobs to machines. Yet the grid has its limitations. It's restricted to internal PCs because of firewall and licensing issues, Bremer says.

Bremer started with about 18 desktops, expanded the pilot to more than 25 end-user machines and plans to increase it to more than 100 in the next few months.

"We get through a lot more data a lot more efficiently now," he says. His team was able to define data-mining workflows and automate about 80% of the process with the grid in place. Bremer's not sure the other 20% needs to be automated and plans to expand its use to more applications.

"The best result of the grid is that it opened our eyes to a lot more ways we could use it," he says."We lifted a barrier to being able to do more computationally intensive experiments."

While Bremer did not discuss specifically what his organization spent on the grid software,he says he was able to use an existing Linux server. United Devices says its GridMP Workstation product can start at $50,000 and scale according to the number of servers, agents and additional components purchased. Pricing for LexiQuest Mine starts at $60,000.

Copyright Network World Inc. Mar 21, 2005

Source: Network World

More News in this Category