By Jim Massey | Dr. Yasir Khan
Join Dr Yasir Khan and Jim Massey from Cerner Middle East in a conversation of the use and challenges to the adoption of Healthcare AI in the Middle East. Jim is a data science evangelist and project lead for AI, data, deep learning and use case exploration at Cerner Middle East. He is a thought leader in practical application of data science for healthcare in the region. Dr. Yasir is a Physician Executive for Cerner Middle East with special interest in big data analytics, machine learning and cognitive computing applications. Together they are trying to explore AI applications for point of care clinical decision support and detecting unidentified patterns and correlations between clinical processes and outcomes.
JM: AI is a big field and covers a lot of parts, Big Data Analytics, NLP, Cognitive Computing, Signal and Image processing, Robotics and so on, but it’s been around for a long time, back in the late 70’s in Scotland, I was working with Stoyan Kableshkov[i] at Burroughs Machines. Just as we were going through the first big revolution in computing, that of raw processing power and the steep walk up the slope of Moore’s Law. We were using all this new found computing power to build rule based ‘AI’ engines, very statistics driven approach to AI, but also looking at things like NLP and voice recognition. Stoyan had a very different view of machine intelligence, way ahead of his time, wanting to simulate the way we think rather than just apply more power to logical linear processing to imitate intelligence. Since then there has been a lot of work in different fields of AI, and after the CPU revolution we went through the communication and storage computing revolutions so our ability to capture and store data is so much greater. Healthcare, although a high-tech industry, is behind the curve on technology such as this but there have been some examples of its use in improving the quality of outcomes, it hasn’t really had the health impact people thought it would, until now.
YK: What do you think are the main challenges in application of AI in healthcare.
JM: There are 3 areas that challenge me, ethics, data and adoption. I think these are issues that the industry must tackle, we are going to be on the leading edge and have to help sort this out. Ethics is around the ownership and use of the data, we are at best the custodians of the data which belongs to the patients themselves, how do we safeguard it, how do we manage consent, and how do we ensure that we are doing things for good?
Most data sets are not large, in Big Data terms in the EMR and the quality of the data is not always good so we have to spend a lot of time cleaning the data and working on techniques to use the data efficiently. Then there are a variety of tools and we should choose the right one for the right job, the old saying that “if all you have is a hammer then everything looks like a nail”, is so true in this field. This sometimes makes the results of studies hard to replicate, so we need to do “good science” here and make sure we are not finding patterns that don’t exist.
The other challenge is adoption, the clinicians are less conservative than they were when I first started out, and we have learned a lot about how to maximise adoption, however, in some cases the algorithms and results are a bit black boxish and clinicians need to understand how results are derived and that there is evidence behind them, “Trust Me” doesn’t work with your spouse or your doctor.
YK: So why has CME decided to invest into Artificial Intelligence?
JM: Well, Cerner corporately is already investing in it big time, and it is becoming the way of the world, our Middle East clients want to be part of this change and it is part of our CME vision to help them, it all reflects our core values of Community, Happiness, Integrity, Proactivity and Passion. Until they sort out data protection and some of the ethical issues it’s up to us to lead. Also, the genome and phenome of this region are different to those in other parts of the world, so what works in one place does not necessarily work here the same way, and what is a priority in US and Europe may not be the pressing issue here, so we need a local capability to focus on our client needs.
YK: What do you think about the role of artificial intelligence vs. clinical judgement? How do you see them working together?
JM: Phew! That’s a tough one and I really would defer to the clinicians here to answer it! We all know that the human body, and its physiological and pathological processes are quite complex, consequently I don’t think that care provision will become such a measured problem to which data alone can provide a solution. Probabilistic reasoning and clinical inferences combined with the process of elimination are central to clinical decision making. They allow the doctor to focus on the most likely cause of illness and determine the most effective treatments. The doctor, through years of training and experience can make decisions that the machine after a few days training on a limited data set cannot. I see artificial intelligence as the supporting tool which will greatly enhance the accuracy and relevance of clinical decision making, just like stethoscope, ECG and Echocardiogram has done in the past. It will lower the cognitive load and open new dimensions for care providers.
In some areas such as image analysis, AI can be better than humans, for instance I’ve seen studies where the examination of images such as mammograms for indications of problems can be done much more accurately and consistently by machine. AI can also make connections on our social behaviours and networking that were not possible before and this will become more useful in behavioural health.
However, I would still like some human involvement in the interpretation and diagnosis if it were my family being examined. It raises those darned ethical issues again, if you go to a doctor it is usually a very personal trust relationship, mostly because they wear a white coat with a stethoscope around their neck, letting a machine do the hard work does not engender that trust. There is also the question that has been brought to the forefront by Google’s “Duplex” announcement at its developer conference, should AI software that’s smart enough to make humans think they interacting with humans be forced to disclose itself? How does this affect the doctor patient relationship? Duplex has emerged at a sensitive time for technology companies, and the feature hasn’t helped alleviate questions about the industry’s growing power over data, AI and the consequences for privacy and work let alone its effect on healthcare.
Real challenge will be to train our clinicians to open their minds to the potential offered by AI and embrace it while maintaining the centrality of the doctor-patient relationship and importance of human touch. AI will not be competing with the humans but augmenting what they do best.
YK: What would be the initial focus of CME AI and why?
JM: We have two programs that I am working on, disease specific re-admission algorithms and Atrial Fibrillation risk factors. The diseases we have picked are Acute Myocardial Infarction and Congestive Heart Failure. There are a couple of reasons for this, firstly we are in a region where cardiovascular disease is a major problem, so if we get this right we will be helping our clients and their patients, secondly Cerner already has done a lot of work on these algorithms and although some of the factors may be different they are tried and tested, so we have the results with US data. If we can adjust and retrain on Middle Eastern data we can see if we really have something useful for the region. Working with the rest of Cerner, speeds up our learning curve and adds to the overall Cerner offering.
The second one on Atrial Fibrillation risk assessment came from an article I read in JAMA, and it looked like something we could improve on with supervised ML techniques and an EMR size data set. We are working with a local university to get a research grant for some of this work. Other clients have come up with ideas as well and we need to see which ones are viable. We are focusing on the big data, machine learning approach as it matches our data and our skills, the areas like Chatbots for diagnosis, robotics for surgery, image analysis etc., I will leave for others who have experience in these areas.
YK: What are the future areas of possible expansion?
JM: Well we can do similar sorts of things with our other clients, or take some more of the good stuff being done in the rest of Cerner and modify it for the region. But this area is limitless, and we need the good ideas from our associates and clients to drive the work. If we have success here I think it will be possible to get more clients involved and lots of ideas for the work. The other area is working with universities and other companies, already one has come to us to work on a genetics project if we get to work at this one it will be great and I’ll tell you more when I can.
YK: I heard you had opened an AI Lab in Dubai and wondered what it is all about.
JM: The rumours are that I’m mining crypto currencies, that would be nice, but unfortunately it’s much more mundane.
YK: OK so what are you really up to, I can’t believe it’s mundane?
JM: The idea is to work with our clients to help them get some value out of the data they have been collecting in EMRs over the years. Our clients prefer that their data doesn’t travel, so we must bring the machine to the data rather than the data to the machine!
YK: Tell us about the AI lab in Cerner ME, the specification of the equipment and its capabilities.
JM: Last year we were trying to see how we could help our clients get through the AI hype and use the techniques to mine their data for insights into health in the region, as I said earlier we have the issue of the data not being readily transportable, but there was so much Cerner was doing in this area that we didn’t want our clients left behind. So I put together my idea of a low cost entry into this area with a locally based machine….. basically, we have a Dell workstation which I gave steroids to. Twin 8 core Xeon CPUs at 3.2 GHz Clock speed, a 5.2 TFLOP GPU with 8GB memory and 1792 Cores, 192 GB of RAM (yes that’s not a typo!) 8TB of Hard Disk and 512 GB Solid State Disk. We are running Linux on it and it is quick.
The development environment is built to support R and Python 3 and currently the ML tools such as XGBoost, Keras, Tensorflow, Theano etc. I’m using Anaconda to run Jupyter notebooks as the IDE so people can develop on mere mortal workstations and laptops and we can then use the AI Lab to run on very large data sets. Using Anaconda allows us to create different software environments simultaneously so we don’t get as many problems with the sensitivity of the tools to different code levels and such.
YK: How will the AI Lab contribute to associate skill development in CME?
JM: We do a lot of work with our associates on keeping their skills up to date and relevant. I will be putting together a number of sessions for our associates to help increase awareness of AI tools and techniques and to share the experience of how we can use the tools and data to create informed decisions and to design usable interventions. This is the future in healthcare and it is important that we all have the new skills we need to support our clients.
YK: How does the UAE AI lab intend to take advantage and connect to AI initiatives and experiences at Cerner globally?
JM: As I’ve mentioned we are working closely with them, letting each other know what we are up to and collaborating on the initial pilots, I intend to keep deepening that relationship as we to ensure that the results of this work is able to be shared as far and wide as we can.
Digital Health Buzz! aims to be the destination of choice when it comes to what’s happening in the digital health world. We are not about news and views, but informative articles and thoughts to apply in your business.