Linux helps medical research: a case study

Dr. Bob Hollebeek, Professor of Physics at the University of Pennsylvania and the Director of NSCP explains IBM's strategy for Linux in this environment to Rich Berman.

Berman: What were the challenges you were facing at the start of this project?

Dr. Hollebeek: Very simply, we needed to acquire digital patient records and make them available. We focused, initially on mammography.

Mammography is an area that is just converting over from analog to digital methodologies. Interestingly enough, the volume of information generated by these new, digital, mammographic machines is really quite large.

Typically, we measure their output in terabytes, per hospital, per year. Another challenge has been trying to build something that would allow hospitals to securely capture that information from those digital machines without having to deploy really serious databases with the hardware on site.

We have solved that problem with the constraints are that you have to handle really high volumes, and you have to do it at relatively high security levels.

Berman: And how was Linux a solution to the challenges of high volume and high security?

Dr. Hollebeek: We needed a vendor that could deliver all the basic components of a large-scale, on demand solution. The Linux cluster includes software, servers and storage equipment that enables leading edge research and data sharing between healthcare facilities.

We basically built systems that were able to move medical content from one hospital to another while simultaneously satisfying all of the government regulations for security. We also had to build in the work flow requirements within -hospitals such as when you can and cannot move medical information.

This work resulted in what is called NDMA, which stands for National Digital Mammography Archive. The development of NDMA was needed because mammography has an extremely large volume of data files that were often unavailable, leaving radiologists without comparative records for making diagnosis.

Berman: How did the University of Pennsylvania get involved in this project?

Dr. Hollebeek: The University of Pennsylvania helped start NDMA and later decided that this was sufficiently interesting to consider commercializing as a technology transfer. So, over the past year, I have been working with a new company called i3ARCHIVE.

Berman: How does this all come together?

Dr. Hollebeek: i3ARCHIVE now has real hospital clients so we are beginning to bring up hospital systems and provide commercial services to them. The way this is done is to put a very high security but lightweight footprint inside the hospital network.

This "WallPlug system" is capable of interfacing between things that have to happen inside of the hospital and things that need to be transferred to and from the outside. These are the kinds of services that we need in the back-end to build these serious databases. That device is called a wall plug just to evoke the idea that this is something that you can plug into and you get services from.

Berman: Where does Linux play a part in this?

Dr. Hollebeek: Linux is used in two places in the system. The wall plugs are special security devices. Part of that runs Linux. Linux is used there to manage network protocols and do very efficient network transport.

The files we use are fairly large and we want to move them efficiently and rapidly. For programming networked systems, Linux is really quite a good choice.

Secondly, the back end databases which are very big, they can scale up to 1,000 terabytes, exist on IBM hosting floors. They run entirely on Linux. So they are running big parallel databases and multi-node systems and clusters.

Berman: Was it because of the clusters that you decided on Linux?

Dr. Hollebeek: In order to get the large scale that we need we had to build cluster systems that are scalable i.e. they can be made very large. That has been a research area of mine for a long time.

It is just more efficient and easier to build clusters in a Linux environment. There is a lot of cluster experience in Linux and a lot of tools. I think if you are going to build a big cluster, you are going to build on a Linux system.

Berman: Would the wall plug would be a good example of IBM's on demand computing?

Dr. Hollebeek: Yes, this is an example of on demand computing. It is particularly difficult to do that kind of thing for medical records, as I said, because of all the security issues. I have been involved in high end data management, high speed networks, data mining, etc. for some time, and in this medical arena,

The data security issues make everything more difficult. What we are doing now, is bringing all the results of the research into a clinical environment and to deploying it in hospitals. It is not just a research toy, it is real honest-to-goodness (it actually does work) kind of stuff.

Berman: And now what happens?

Dr. Hollebeek: Well, once you can interconnect hospital enterprises and can make services of this kind available to hospitals, it facilitates a whole bunch of things. It opens up the possibility of moving records to obtain second opinions for example.

The second thing that it facilitates is being able to transfer records for a patient from one hospital to another. For instance a patient may get a screening exam at one hospital but go to a different hospital for diagnostics, we could move that data.

A third area is retrieving prior exams. It turns out that old film studies are often difficult to find. Many women carry film around with them as they go from one place to another, especially if they have had any sort of problem. Locating all of the records is difficult, at best, and often fails. One in three times you cannot find the records. They are not necessarily lost, but perhaps are misplaced.

Add in the high shipping costs and the inconvenience of mailing records to someone. Once those records are in digital form, they can be moved around, they can be sent as long as you are sure that you have satisfied all of the proper regulations.

Women with their records in a system like this, don't have to worry about them getting lost or misplaced, she can get them delivered anywhere she wants. She can even request a copy of her records on a CD.

Availability of records has greatly improved. The fourth thing is that delivering other services related to the image is made possible by these systems including new an sophisticated aids to detection.

Berman: Can you give an example of this?

Dr. Hollebeek: Yes. There is a standard digital procedure of the CAD, Computer System Detection, which is a computer procedure that goes through the digital image, or the scanned film image looking for abnormalities and helping the physician make sure they don't miss anything and look at problem spots, and things like that.

Typically, in order for a hospital to acquire a service like that, they have to buy some expensive equipment, they have to hook it up, they have to maintain it, they have to operate it, and the have to pay for it. Once all of that is accomplished we can gets these kinds of reports from the physicians that are in their hospital.

However, with the WallPlugs that we deploy, we get the digital records in our system, and that diagnostic software can run on our systems, and we are able to return the results back to the hospital.

So, lets say a new hospital decides they want this, what is the barrier? The previous barrier was all kinds of equipment, and lots of other things. Now the barrier becomes to just sign up for the service.

Berman: This sounds like a great service.

Dr. Hollebeek: We think it's important to deploy those kinds of services. This is a mechanism by which we can propagate sophisticated digital services to hospitals much more quickly than if you have to deploy hardware on each site.

When hospitals go through the transition from film to digital it requires a new digital machine, and many hospitals are being asked to buy these new machines for a lot of reasons including cost savings. For the hospitals who acquire a new machine there is a pretty big barrier to full operation because the new machines are the ones that are prolific in their output.

They double the digital load inside the hospital. Unlike broken bones, their output is information that you want to re-access year after year as the same person comes in for annual screening. There again, the procedure that we are following is really on-demand provided by these local WallPlugs.

This is a good model and it allows the hospital to easily transition into the digital mode. These systems are all externally managed so it doesn't require expertise on the part of the hospital, so small clinics or mid-sized hospitals that otherwise might find these transitions really difficult, can just go ahead and do it.

Berman: What do you see developing in the future?

Dr. Hollebeek: We're very enthusiastic about what the future holds. What we are planning in the future is to build out a significant infrastructure on the IBM Global Structure. IBM Global Hosting provides the networks, the floors, the location, and the communication links between sites.

Then we are putting clustered systems and storage systems on these sites. That is the backbone of the distributed database. Where we are headed is building fairly large clusters which very large storage capacity, and parallel database locaters in those clusters in Linux environments. Linux will continue to play a dominant role in the way the architecture runs.

Berman: What are the benefits to using Linux?

Dr. Hollebeek: We are concentrating on building very large scale structures that do a lot of complicated things together, many nodes working together, and Linux is really quite ideal for that kind of a problem.

Linux is also relatively high security. It is easy to lock the system down and know exactly how secure it is. So those are all pluses. The main thing that Linux brings here is a lot of expertise in clustering, and a lot of tools in networking systems together that are distributed.

I think that this is the first large scale attempt to bring hospitals together in a multi-vendor environment so we can handle data from different providers of different medical instrumentation to provide all of the ability to track and move of the information from hospital A to hospital B.

That is going to be something that is a leading issue in the next decade. This is because we are trying to become more efficient in medical systems. President Bush was correct when he said that we are using 18th century methodologies in some of these areas.

Berman: Your clients are able to take advantage of 21st Century technology while avoid up-front infrastructure costs by using the Linux and IBM's grid-based architecture that you've invested in.

Dr. Hollebeek: We are trying to help the computer revolution to arrive in health care. There are so many issues there that we chose the path of making something that is very easy to install in the hospital and yet through the wall plug, gives them all of the services.

If you are thinking of putting electricity in your house, you don't have to build a generator, or a dam, or a line to the power plant, you just put the plug in the wall. In the hospital they want digital services, they want digital storage, they want digital imaging, and they want databases.

But they don't want to build all of that themselves, nor is it appropriate to put that level of infrastructure inside the hospital. That is where we are headed on the grand scale. I think it is working quite well. The first clients found it quite painless. The information goes into the wall, and they are confident it is taken care of.

When they need it back, we give it back to them. That allows them to move on and do a better good of medicine without trying to concentrate on building a serious database in their hospital.