Playing House: How IBM's Watson is helping doctors diagnose the most rare and elusive illnesses...
Adam Rifkin stashed this in Medicine
IBM's cognitive computing system can be trained by 1200-question questionnaires.
Not long after its headline-making Jeopardy! win, Big Blue began retooling Watson for the healthcare industry, and it's since been put to work on a number of healthcare applications, from cancer diagnosis to diabetes analytics.
Now, Marburg's Centre for Rare and Undiagnosed Diseases have begun examining if Watson has a role to play in helping identify conditions that doctors may only come across once in their careers.
Typically, the patients arriving at the centre may turn up with inch-thick medical records replete with unstructured data -- X-rays, lab results, doctors' notes, scans, and more -- and have already passed through the hands of tens of medical professionals before reaching Marburg. It could take the centre's doctors days to trawl through such records, trying to digest all the information they contain in order to arrive at a diagnosis.
"Once patients show up [at the centre], they are highly complex to get a handle on and we are busy for a couple of days just reviewing the paperwork," Schäfer said. "It is clear to us that either we keep that old way of huge amounts of information on paper and work through it for days and days or weeks, or we set up a new system, where IT support screens the findings and gets out the information."
The aim in future is to feed all that information into Watson, and let the system crunch the data before arriving at a differential diagnosis -- a list of potential illnesses that all fit the patient's symptoms -- which will be handed to human medical staff to work out which of those conditions the patient has. And, if Watson spots a seemingly anomalous result amongst all the data, it can highlight it so that doctors can follow it up (or disregard it and carry out the test again).
While Watson could theoretically aid further in the diagnosis by weighting each of the illnesses that make up the differential diagnosis according to their likelihood, Schäfer says there may be pitfalls to such an approach.
"We are currently discussing if we are going to use this, but this is a risky thing: that if you say this or that is the most likely [condition the patient has], you will focus on this first. But that doesn't help the patient if he is the one percent of patients that have the other [condition]. We need to be sure that if Watson would supply the likelihood, there is a concern that at the end the physician gives up thinking. 'It's 90 percent that it's a myocardial infarction, that's good enough," but if it's a one percent chance that it's an aortic dissection, which is also a deadly disease, you take the risk that no one thinks about this. We are discussing currently if we want to use this feature."
The centre aims to take the Watson system live over the coming months. To help facilitate its diagnoses, the centre will introduce a 1,200-question digital questionnaire that patients fill in. The information they give will be fed into the IBM system to give it an extra insight into factors that could affect the diagnosis, from their symptoms and family history, to the environment they live in, and their jobs. The data will be anonymised before being fed into the IBM Bluemix cloud for processing. Watson will also be able to translate the information provided from German to English to cross reference it against medical information in the language. In future, the system may also be used to make suggestions of further tests that could be carried out to pinpoint the diagnosis by ruling in or out certain conditions.