Paging Dr Watson: IBM’s supercomputer in patient care

When someone complains about too much information, or TMI, it can sound like one of those first-world problems easily solved by switching off the smartphone and stepping away from Twitter. Unfortunately, TMI is an all too real and urgent problem for those in the medical community.

Everyone from doctors developing cancer care programmes to pharmaceutical researchers working on new drug discoveries are drowning under the weight of a growing body of medical literature that doubles in volume every five years.

By 2020, there will be 200 times more medical information and data out there than a physician could possibly process in her lifetime. But that’s okay because IBM’s Watson “supercomputer” has gone to medical school and is ready to see you now.

Watson is most famous for beating a human at the American television gameshow Jeopardy!, which might seem like a far cry from the complex task of helping doctors read through medical research. However, as vice-president and chief technology officer for Watson, Rob High, puts it: "It represented a really grand challenge."

"[Winning at Jeopardy! involved] dealing with enormous quantities of information – Watson has to read through 200 million pages in three seconds to find answers – but it was also dealing with a game that was designed to bring out the worst of human language in terms of subtlety, innuendo, and idiosyncrasy, while misdirecting the contestant.

“To have a computer that could see through all of that and understand all of those subtleties of language and normal forms of human expression and still be able to compete with humans at that task, while at the same time reading though enormous quantities of information, that was an enormous challenge. It was a challenge in its own right.”

This worked out in favour of more complex applications because doctors approached IBM after seeing the gameshow: "It had the benefit of informing a very broad swathe of the public in areas that we would not have imagined previously as having this problem and recognising potential for this problem solving."

‘Cognitive computer’

As

Jeopardy!

demonstrated, Watson is smarter than most other computers. This is because it is not technically a supercomputer; it is a “cognitive computer”. This means it can think a little bit like a human does and make sense of the messy, unstructured data in the world around it, using it to answer important questions.

Watson can do this in any field. Feed in all available information on say, cooking, and Watson can learn how ingredients pair with each other and suggest recipes based on this. In fact, it already does this on the Chef Watson project.

High says that cognitive computing is partly about “whatever it takes to help amplify human cognition”. Watson isn’t a form of AI gunning for world domination; it is more like a smart and diligent personal assistant that consistently writes well-researched reports for you.

In the case of medical research Watson can “read” through hundreds of thousands of medical journal articles and recommend what ones the doctor should be reading, given the problem they’re currently working on. With cancer care, Watson can help a medical professional develop a personalised care plan based on case notes, doctors’ notes, and medical literature.

“It’s a fairly sophisticated set of activities that are being performed, that starts with doing patient segmentation.

“We try to find the ten out of a million patients that are just like this patient, and not just based on age, race, sex but based on 100, 1,000 or even 5,000 variables that define who that person is because that then becomes a good basis for making predictions about how that patient is going to respond to the cancer, and to the treatment.”

Read prose

This isn’t as easy as stuffing text into Watson and waiting for it to spit back data patterns. Watson has learned to read prose (clinical notes), and is versed in the nuances of seven languages, including inflection and intonation, so it can watch videos and listen to audio as well.

Eighty percent of data that exists is in this form – prose, audio, video – and the beauty of cognitive computing, says High, is that it taps into this 80 per cent unlike traditional computing methods that rely on structured data being fed into a programme or interface.

“There is so much data being produced. There is 2.5EB (1 Exabyte is equal to 1 billion GB) of data being produced on a daily basis and nobody has time to read all of that,” observes High, who thinks that these data demands will soon see Watson emerge from specialist scenarios to tackle the big data of everyday life.

And I've finally hit on the buzzword. The term "big data" is overused and not very well understood. But the less said about it as a generic, catch-all term, the better. Big data should only ever be discussed in the context of how it is used to make life easier, and in the case of a research centre at University College Cork (UCC) it is helping premature babies survive.

Ceaseless chugging

At the INFANT (Irish Centre for Fetal and Neonatal Translational Research) Centre based in UCC the ceaseless chugging of machinery, humming of sensors, and whirring of equipment is the soundtrack to the constant monitoring of at risk newborns, capturing data 24/7.

"Our research is about interrogating that data in a clever way that can't be done if you're just monitoring a baby at the bedside, and trying to turn that massive data into relevant information that is clinically usable," says Geraldine Boylan, professor of neonatal physiology and director of INFANT, which works with Cork University Maternity Hospital.

A lot of people don’t think about babies when they think about data, says Boylan, but this is a huge part of what INFANT does. Clinical scientists work side by side with data scientists and electrical engineers to turn data from these neonatal sensors into predicting and preventing, for example, an infection in a premature baby, based on the oxygen levels, heart rate and brain activity of babies who have been previously monitored, with consent, at the centre. One project, Babylink, involved working with IBM’s data analytics tools. It involved developing an algorithm to detect when a baby might have a seizure. This involves using an electroencephalogram (EEG), which is a bit like a heart rate lead but is placed directly onto the scalp.

“A baby can’t tell you if it’s not feeling well and there isn’t a person standing by the baby the entire time, so the only indicator we have is by monitoring signals in the brain. And of course you need people to interpret those signals that are being collected 24 hours a day, seven days a week. That isn’t something a human can do very well but an algorithm can.”

This new algorithm has been developed by taking vast amounts of data recorded over the past ten years from babies in Cork, London and other international partners of INFANT. It can detect, and thus help prevent or treat, 90 per cent of seizures occurring in the babies monitored.

Detect problems earlier

“Just as Watson learns, we are giving our algorithms the data to learn what’s good, what’s bad, what we should be worried about. It is only by giving it lots of data that we can actually train them to detect problems earlier, treat them faster, and improve the outcome for the baby,” she explains.This really is a big data problem, says Boylan: “”There are approximately 20,000 NICU cots in the US, for example, and a pre-term baby could spend up to three months in the hospital. Meanwhile data is constantly being recorded from 16 different sensors on each baby. That equates to nearly 700MB per day per baby. When you add in video monitoring and analysis, that’s serious data – it goes up to 240GB per day.”

There is now ten years’ worth of this kind of data from INFANT waiting for the right kind of data analytics to make meaning of it and allow it to help future at risk babies. If we were to feed all of this data to Watson it would help hugely, reflects Boylan.

“There are patterns there that we haven’t even seen yet and our data scientists are working on discovering these at the moment. The possibilities are endless.”