There's no cure for dementia, but diagnosing it and treating it can go a long way toward improved quality for life those affected, as well as their caregivers.
Since 2012, IBM Research, in partnership with ten academic and industrial partners have been working on a way to identify those at risk, faster, and with greater ease.
Of course, these days that means using a smartphone or tablet.
The project started off at DemCare, said IBM scientist and head of the project, Aharon Satt.
"Everybody loses some of the cognitive capabilities in normal aging, but in dementia, which is far more rapid than the normal decline and eventually, people lose some of their cognitive capabilities and there is a lot of interest to detect it very early," Satt said.
Dementia is fairly common, and the number of cases are expected to triple by 2050. So, identifying it is one of those "the sooner the better" situations. But at the same time, it's impractical to march every person older than 60 in to the doctor for an MRI.
"The question is how to do it, and typically, dementia is detected quite late because by the time the family puts enough pressure to get to the memory clinic, the patient might already be in the first stage," he said.
What the team came up with is essentially an application that would work by asking the user questions and showing them picture prompts, and recording the user's voice. The questions ask the user to repeat a sentence they just heard, count down from a number, and describe pictures, then there's half a minute to name all the animals the person can remember, along with a few other prompts, like describing something sad, or the events of the previous morning.
While those questions may seem random, they're actually an adaptation of a battery of neuropsychological tests, which would normally take about half an hour to perform, and require a trained clinician to administer. Each question aims to assess a different capability in the brain. The project shortens and automates the process down to about five minutes.
The recordings get uploaded to IBM's Watson. Watson runs algorithms on the recordings. Over the years, they've collected hundreds of recordings of patients, which were analyzed and put into three groups — the control group (older folks who complained of memory loss, but did not actually suffer from anything abnormal), pre-dementia (those exhibiting mild cognitive impairment), and early dementia including early Alzheimer's. It obviously requires patients who want to be actively involved in managing their own health.
Watson doesn't analyze the actual content from the recordings, but rather, the tone of voice, the pauses between words, the amount of hesitation, or the continuity of the speech. The chief reason is because they wanted the technology to be language agnostic, Satt said. That was actually one of the challenges in creating the technology.
"When a human judges it, clinicians usually refer to what is being said, and we proved algorithmically that we don't need it," Satt said.
The algorithms can kick out results with 85% accuracy and place the user in one of the three groups, Satt said.
It's not a formal diagnosis, but it's meant to be a tool to help clinicians.
"Even if it says you probably have dementia, it's not the end of the game. You need to go do some careful medical evaluation, but it is risk assessment," Satt said.
He also said they are still trying to determine how much information the user should get back in the results, if it's less detailed than what might go back to the physician or clinician.
At the moment, the technology is only available to clinicians who are taking part in the research. They've focused on Europe, but are hoping to also expand to the United States. So, it's not a product just yet.
"Between research and product, there is a very big gap of regulation and ethical committees, etc.," Satt said.
Erin Carson has nothing to disclose. She doesn't hold investments in the technology companies she covers.
Erin Carson is a Staff Reporter for CNET and a former Multimedia Editor for TechRepublic.