Login to Your Account



Eyes on Alzheimer's: Finding answers via crowdsourcing

web_pic_sept._11_2017.jpg

By Nuala Moran
Staff Writer

LONDON – An online computer game is beginning to deliver results in a project that is crowdsourcing analyses of live video images of blood flow in Alzheimer's disease model mouse brains, in a bid to discover new drug targets.

For the first stage of the project, Eyesonalz, Cornell University's Human Computation Institute developed Stallcatchers, a game in which players study movies of mouse brains to assess if highlighted capillaries are clogged.

To date, a crowd of several thousands, consisting mainly of middle school students and seniors, have analyzed 40,000 images. The gamers compete for points by identifying clots, or stalls, which reduce the blood flow in the brains of Alzheimer's disease patients and in the transgenic mouse model, by 30 percent to 50 percent.

The first results, presented at the British Science Festival at Brighton University on Friday, show, against expectations, that stalls are not concentrated around the amyloid plaques that are the hallmark of Alzheimer's disease.

However, said Pietro Michelucci, executive director of the Human Computation Institute, "if plaques are not causing stalls, this could be important information for developing treatments."

Michelucci was describing the project and launching the next phase, in which links between dietary fat and blood flow in the mouse brains will be investigated.

Eyesonalz was initiated on the back of a breakthrough by Michelucci's colleague, Chris Schaffer, at Cornell's biomedical engineering department, in using fluorescence microscopy to look into the brains of live mice and capture videos of blood flow in vivo.

"The problem was that answering research questions using the imaging data took one year of manual lab analysis. Machines are not good at this, whereas humans are good, but are slow," Michelucci said.

One hour's worth of collected data requires a week's worth of annotation by laboratory personnel. "It would take decades using paid experts to get to a treatment target. Instead, we have developed an online game anyone can play, which at the same time analyses the data. [Gamers] are doing the same job as paid analysts," said Michelucci.

The magic number

Since its launch in October 2016, Eyesonalz has compressed one year of Alzheimer's research at Cornell University into just two weeks.

To score points, gamers must make a binary decision: Is the blood flowing or is it stalled? Moving white dots on a black background indicate flowing blood; a black gap is a stall.

"We designed the game so that even early stage Alzheimer's patients can contribute directly to their own potential treatment," Michelucci said.

Although the link between reduced blood flow and Alzheimer's disease was established some time ago, its role in disease etiology is not understood. However, reversing the stalls restores cognitive function and reduces other Alzheimer's symptoms in mice.

In the brain of Alzheimer's mice, 2 percent of capillaries are stalled vs. 0.5 percent in healthy brains. Because of downstream effects, the stalls in Alzheimer's mice reduce blood flow by up to 50 percent.

As Michelucci noted, while gamers bring eye power, that creates a new problem, of how to be sure the crowd can be trusted. The answer lies in using validated data to work out how many people's answers are needed to come up with the same result as a pathologist.

Michelucci originally concluded 15 percent 20 non-expert crowd answers are equivalent to one expert. Subsequently, by introducing dynamic assessments of crowd answers, the "magic number" has been cut to seven, speeding up analysis.

Another problem is how to control for random inputs such as web bots, or a cat climbing on a keyboard. The solution is a blue bar, or gauge, on the screen, that shows gamers how many right answers they have given.

"We build trust over time and throw examples at people where we know the answer. If they get it wrong, they lose trust very quickly and we know not to take any notice," Michelucci said.

The techniques for analyzing crowdsourced data were validated in two studies using images that had been assessed by experts. The experiment Michelucci reported at the British Science Festival, assessing the proximity of stalls to amyloid plaques, is the first time the crowd has analyzed de novo data.

"Cornell researchers are still checking [the crowdsourced findings], but it looks negative and that the plaques are not directly causing the stalls," said Michelucci. "We have answered an Alzheimer's research question for the first time by using crowdsourcing."

The next stage of the Eyesonalz project, looking at the possible role of a high fat-diet in the development of stalls, is expected to take two to four months to complete. All findings will be checked by the Cornell research team.

It is intended to include the names of all volunteers as co-authors when the results are published in a scientific journal.

These are the early stages of the whole project, which will move on to evaluate the role of cardiovascular risk factors in Alzheimer's; examine the impact on the pathology of reducing the number of stalls; attempt to identify the molecular and cellular mechanisms upstream of stall formation; and screen 300,000 drugs that have cleared safety trials.

Funding for the Eyesonalz, which also involved scientists at the University of California, Berkeley and Princeton, comes from the Brighteye Foundation.