A shaking mouse cursor, slow scrolling speed and repetitive online searches could allow computers to spot early signs of Parkinson's or Alzheimer's disease.
Research by Microsoft found a link between a trembling mouse cursor and online searches suggesting people have Parkinson's disease.
They say people's online data could one day be used to spot early warning signs of the brain diseases and lead to quicker diagnosis.
Scientists looked at data from more than 31 million Americans using the Bing search engine to try to work out the behaviour of people who might have the conditions.
They suggest the shaking caused by Parkinson's and memory loss of Alzheimer's patients are symptoms which could be noticed in people's online activity.
Software could even be developed to spot the diseases before another person notices it, and to warn people to visit a doctor, the researchers say.
Researchers from Microsoft and Duke University in North Carolina looked at how much people's mouses shake side-to-side or back-and-forth while they try to make normal movements.
They also used data from Microsoft's Bing search engine to show when people searched the same things online over and over again.
This could point to memory problems, which are a symptom of the brain degeneration in both Parkinson's and Alzheimer's.
Around 127,000 people in the UK are thought to have Parkinson's disease – approximately one in 500 people – and more than 520,000 people have Alzheimer's, which is the most common type of dementia.
Both are incurable degenerative brain diseases which tend to worsen over time and cause changes in people's behaviour and can make patients reliant on care.
One of the main symptoms of Parkinson's is uncontrollable shaking of parts of the body, often the fingers.
The researchers identified 703 people who potentially already have Parkinson's – those who search for things like 'just diagnosed with Parkinson's'.
They now plan to compare those people's online activity to that of volunteers who have a confirmed diagnosis.
It is hoped that, if the algorithms can correctly identify people who already have Parkinson's or Alzheimer's, they could pave the way for software that can one day spot it before it has been diagnosed.
Computers could notice symptoms earlier than a patient or family member then alert people and advise them to go to a doctor, the researchers suggest.
Because the computer algorithm would have been analysing the person's activity for a long time – people in the study were watched for 18 months – there could be extra information to help with a diagnosis.
Lead author of the study, Dr Murali Doraiswamy told the Wall Street Journal: 'Both of these conditions in their very early stages can be very hard to differentiate from a host of benign conditions, so the misdiagnosis rate is high.
'[With computer data] you can see how someone is changing over time, which might give you greater sensitivity and accuracy in making a diagnosis.'
Those who were monitored in the original study did not know they were part of it, but researchers say the data is anonymous and it was not a breach of privacy.
Addressing concerns about people's privacy being breached, Dr Doraiswamy added: 'This type of data is generally considered acceptable to analyze because it’s de-identified, so there’s no privacy issues.'
Microsoft and Duke University's research was published in the journal npj Digital Medicine.