>> Man: Live from Washington DC it's the Cube. Covering .conf2017, brought to you by splunk. >> Welcome back to Washington DC, Nations capital. Here for dotconf2017 as the Cube continues our coverage. The flagship broadcast of silicon idol tv. Along with Dave Alonte, I am John Walls. Glad to have you with us after we've had a little lunch break. Feeling good? >> Feel great, good conversation with customers, dug into the pricing model, got some good information. >> What did you learn at lunch? >> Well talk about it at the end of the day. >> Alright, good, look forward to it. Let's talk healthcare right now. Derek Merck is with us right now. He is the director of computer vision and imaging analytics at the Rhode Island Hospital. Which is the teaching hospital for Brown University. Derek thanks for joining us here on the Cube. Good to see ya. >> Absolutely, very excited to be here. >> So, well and as are we to have you. Director of computer vision and image analytics, so let's talk about that. What falls under your portfolio, and tell us where does Splunk come into that picture? >> It's been an interesting journey, Rhode Island hospital is a huge clinical service. Takes really good care of the people of Rhode Island. I'm in diagnostic imaging, so I work with all the CT scans, the MR's, radiography, ultrasonography, and what I try to do is automate the data that is coming off all of these machines as much as possible. So, you know typically the patient will come in, they'll get imaged for some reason, the physician will take a look at that image and make a diagnosis, and then that image goes into an archive. It may be used again later if the patient comes back but other than that it is not really used at all. With these sort of emergence of computer vision access to training images, sets of data, has become really critical. Diagnostic imaging has become really interested in taking better account of what imaging they have so that they can try to answer questions like what's alike about these images. What is different about these images, and automate diagnosis. What's similar about all the images of patients who have cancer, versus patients who don't have cancer. Which is basically what a radiologist job is, is to go and look at this patients image and figure out does this patient have cancer or not. SO that is the way you would teach a computer how to do it in an automated fashion. SO I spent a lot of time trying to figure out how do you keep, how do you take, keep better track of what is available and be able to ask these sort of population based questions about what we have in our portfolio of data, our data portfolio. I spent a lot of time writing systems by hand in python, or other kinds of scripting tools. I spent a lot of time trying to interface with the hospital informatics systems, the electronic medical record. The electronic medical record again really meant for taking care of patients it is not meant for population analytics. We ended up basically building our own health care analytic system just to keep track of what we had. What were the doctors saying about different cases. Show me all the cases where the doctors think that some particular thing happened. And be able to ask these questions in real time, generate huge data sets, anonymize them, run them through computer vision algorithms, train classifiers. Diagnostic imaging is really excited about this kind of technology. There has been a lot of interesting side projects as well. One of the most, one of the things that administration is the most interested is because of these kinds of systems we are keeping a lot better track of radiation exposure, per image, so the CT scanners will tell you how much radiation was used for an individual study. But again our analytic systems historically you have no way of saying what's the average? What's high, what's low? Its months of latency, six months of latency between when you run a scan and when American College of Radiology comes back and says some of your scans were a little high in radiation exposure. Whereas now because we keep track of all this data we have this real time dashboards and that is the kind of thing we use Splunk for. WE keep track of all the data we are collecting and then we create these dashboards and give them to people who haven't had access to this kind of analytics before. For looking at utilization, optimizing work flow, things like that. >> I am just kind of curious when you mention like x-rays and maybe Dave you know more about this than I do. But it seems like it is kind of a standard practice you have a certain amount of exposure for a certain amount of test, and that data I don't know how but it sounds like it is more critical to have that kind of data than someone a layman might think. I was curious of the analytics of that. What are you using to determine there in terms of that exposure? >> There's always a trade off with radiation based imaging. There is a lot of non radiation based imaging. Like you may have heard of magnetic resonance imaging, or MR. Those are thought to be perfectly safe. You can get MR's all day long. If fact they do give MR's to people all day long for research purposes sometimes. >> You climb in the tube, I don't want to climb in the tube. >> You get a little claustrophobic >> They are expensive >> That is the thing, we don't have very many of them. They are very slow but they're safe. Ultrasounds very safe, we give ultrasounds to pregnant women all the time very safe, but they don't give you very quality images back. They give you a very small field of view and things are wiggling around. A CT scan is super fast and it gives a physician all the information they need in a snap shot. CT scanners are so fast now they can freeze your beating heart. They can make a revolution around your body of thickness so they can capture your heart while it is in motion. You know like with anything if you have a camera and you take a picture of someone running across the screen you don't see the person you just see this sort of blur, right? Now with modern fast aperture cameras you can take a picture of nutrinos and things that are impossibly fast. I don't know that that's actually true. You might wand to edit that out. (laughing) >> But conceptually >> A CT scan is the same sort of thing. Your heart is beat all the time, your lungs are moving all the time. Your bowls are moving all the time. Your blood is coursing through your veins all the time. It is so fast it can freeze it and give you this volumetric data back. They use that for all kinds of different things. They're not able to do with other kinds of imaging modalities The downside is that they're potentially somewhat dangerous, right? People have known since the 1890's when x-rays were first discovered by Wilhome Rankin that if you put somebody under an x-ray beam for too long, your hair will fall out, you'll get skin burns, all kinds of things that these early pioneers of x-ray did to themselves without realizing it. Documenting all of these problems that can happen, and a CT can uses ionizing radiation if you get too many CT scans you'll get skin reactions, or other kinds of things. It is really important to keep track of the risk to benefit ratio there. People give you a CT scan if you fall down and you hurt your head. They give you a CT scan cause they're worried that you are going to die if you don't get the CT scan. Along with that is this idea of how do you track how many CT scans an individual patient gets in a year. Right now the hospital has a hard time keeping track if somebody comes into the emergency room of automatically identifying oh this patients already had six CT's should we put them in line for a MR instead of another CT. Again these are the kinds of things that we are able to get at through using, through better management of our data and organization of our data. >> You mentioned that you're doing more of this real time analysis, Splunk is obviously a tool that helps do that. Other tooling, are you using cloud based tools? >> We have to be really careful about cloud based stuff. There is this protected health information that everyone's really concerned about. Working with data at the hospital is really walking a fine line you need to be very conscious of security. There really reluctant to let non anonymized data out to cloud sources for storage. There are some ways of getting around that, but basically we run all of our servers in house. There's a couple of big data centers down in the basement of the hospital. Mostly they have clinical duties but we have a number of research servers that are installed down there as well. They're managed by the same IT staff in this sort of hardened architecture. I actually can't do any work from home which is an unusual kind of experience, I am used to being able to log in remotely. >> Oh darn (laughing) >> Or you spend too much time on the job. >> Some times you'd like to >> I'm ambivalent about it, there's goods and bads about it. >> So how do you deal with that streaming infrastructure and real time analysis. Do you guys sort of build your own? Any kind of resource tools, or >> I use a lot of open source tools. Traditionally the hospital wants to pay for everything. They feel like if they pay for things then it comes with uptime guarantees. When I build my systems though, because I'm working on shoestring budgets, And because I believe in open source. I use open source where ever I can. I wanted to mention we're actually for a lot of the work that we do supported through Splunk for good. So I don't pay for a full Splunk license, Cory Marshal who runs Splunk for good, has sort of recognized the value of some of the stuff that we're doing with dealing with non traditional data. It's not the sort of standard things that the other people who are working in the healthcare space with splunk are working with. We are working with imaging data. We are working with patient bedside telemetry data, you know the EKG signals and the heart rate signals. And aggregating all this stuff in to one place to make more sensible alerts and alarms. Oh this patient set off an alarm three times in the last hour I should send a page to the nurse who is taking care of this person. It's different that the kind of business optimism that I think a lot of people in the healthcare space are using splunk for. >> SO you have your core mission around diagnostic imaging. As we sort of touched on you have all these other peripheral factors in your industry. The affordable care act, obviously there's HIPPA, there's EMR, there's meaningful use. How much does that affect your mission? Does it get in the way? Is it something you have to be cognizant of like constantly, obviously HIPPA. Other factors? >> I try to just be cognoscente, I try not to let anything get in my way. Almost all of these things that you talk about they're really meant to protect the patient. I make sure that everything that I do is working with data is that we are anonymizing things, were using data securely, and we are trying to help the patients. I think I just have this moral check in my head of what is what I am doing right now good for my department, good for my institution, good for my patient. Then because I am aware of all these other rules they are very complicated and hard to navigate. At the end of the day I can say I understood that rule, I followed that rule, and what I did was the appropriate thing to do. >> It's like house rules. >> Yeah >> Okay, talk a little bit more about splunk, how are you using it, what it does for your mission, for your operation. >> What I came to the conference this year to talk about is this dose management system that we built that I think is really important. We've had vendors coming in and telling us that medicare isn't going to pay hospitals, or is going to reduce reimbursement to hospitals who can't prove that they're using ionizing radiation imaging appropriately. So what does that mean? No body quite knows exactly what that means. How do I tell whether my hospital is adhering to these rules that are ill defined and these vendors are coming in and they're trying to sell us solutions that are like a hundred thousand dollar a year licenses. Administration is taking this seriously, they're trying to figure out which of these vendors are we going to give money to. In the mean time a bunch of the CT technology staff and I basically put together a system that answers all these questions for them using Splunk. We use splunk to collect meta information about how all the scanners system wide are being use. We have 12 CT scanners, they shoot 90,000 different studies every year. Each one of those studies may be hundreds or even thousands of slices of data in these volumetric data sets. It's a huge amount of data to keep track of. Your not using Splunk to keep track of the imaging per se. Your using splunk to keep track of what imaging you collected. So it is a small fraction, it is just the metadata about each one of the studies. That metadata comes with a bunch of interesting information about what the radiation exposure for each one of those studies was. Splunk has these wonderfully adaptable easy to use tools. That once we covert our strange dicom, device independent communications in medicine data, we flatten it, normalize it, turn it into generic data, it is Json, it's dictionary files. Then splunk has these great tools that can be applied instead of to business analytics and optimization to image analytics and optimization. We build our dashboards on top of splunk to show per institution what was the average dose? Per protocol, per body type, you can track which technologist have the lower doses and higher doses. We found all kinds of interesting things. My favorite story the chief technologist was just telling me. I was putting together my slides for this presentation that I did here about this. I said we need an example of a does outlier. Some time when we had a higher than expected radiation event. We never have dangerously high radiation events. >> Good caveat, thank you. >> All the machines care about is whether you're harming some one and we never harm anyone. The machines don't track, this one is a little higher than you would expect it so that you can say why is that, what happened there? But now we do using our splunk dashboards. So I asked him can you get me an example for my slide deck. He literally just looked over to the monitor that he had open and he says oh right here. Here is a patient who had a 69. These numbers are irrelevant, they're supposed to be 50. He knows what the numbers are supposed to be, to me numbers are just numbers. This patient had a 69 and he picks up the phone, this was 5 minutes ago, he calls down to the control room. He says I'm not blaming anyone but why did Mrs So and So have a little bit higher radiation dose? 69 is not dangerous by the way, alarms don't go off until like 75 or 80 or something like that. So he just called and he asked what was going on with this patient. She had a dislocated arm. Okay I understand. This was a head scan, I was like Scott what does a dislocated arm have to do with a head scan? He said well she went through the CT bore with her arm up over her head which is not the way but it was the only way she would tolerate. So the CT thought she was this big and it had to raise the amount of radiation that it was putting into her to go through a larger object. So he documented that, he put it down, and again we used splunk for ticketing for outlier identification. So he put this one into the outlier identification database that we have, he picked other for the reason because we don't have a drop down menu with dislocated arm. Marked it as closed and it is justified, so when the JCO Joint commission on hospital accreditation comes trough and they say well what do you do to manage your higher than expected radiation exposures? We can both say well we never have unsafe radiation exposures it is all documented right here. When it is higher than usual this is the way we document it, and here are examples of ten or twenty of these odd instances where something happened. Either it was completely justified like this lady where the machines were used appropriately, that was appropriate. Or very occasionally we'll find something strange like an improper head holder was being used at one site for a while. It was resulting in these head CT's should usually be around 45 or 50 and instead they were 55 or 60. They went and they took the metal head holder and replaced it with a carbon fiber head holder that they should have been using and then all of a sudden our doses came down, and we documented it. >> It was a dislocated arm, let's leave it at that alright and we are happy with that. Derek thanks for being with us >> Oh absolutely >> Appreciate the time here on the cube and glad to have you here. Continued good luck with your work at Rhode Island. >> Thank you very much, you guys have a good day. >> Very good thank you. Derek Merck joining us here on the cube. We'll continue live from Washington DC right after this. (upbeat music)