Online crowd-sourcing — in which a task is
presented to the public, who respond, for free, with various solutions and
suggestions — has been used to evaluate potential consumer products, develop
software algorithms and solve vexing research-and-development challenges. But
diagnosing infectious diseases?
Working
on the assumption that large groups of public non-experts can be trained to
recognize infectious diseases with the accuracy of trained pathologists,
researchers from the UCLA Henry Samueli School of Engineering and Applied
Science and the David Geffen School of Medicine at UCLA have created a
crowd-sourced online gaming system in which players distinguish
malaria-infected red blood cells from healthy ones by viewing digital images
obtained from microscopes.
The
UCLA team found that a small group of non-experts playing the game (mostly
undergraduate student volunteers) was collectively able to diagnosis
malaria-infected red blood cells with an accuracy that was within 1.25 percent
of the diagnostic decisions made by a trained medical professional.
The
game, which can be accessed on cell phones and personal computers, can be
played by anyone around the world, including children.
"The
idea is, if you carefully combine the decisions of people — even non-experts —
they become very competitive," said Aydogan Ozcan, an associate professor
of electrical engineering and bioengineering and the corresponding author of
the crowd-sourcing research. "Also, if you just look at one person's
response, it may be OK, but that one person will inevitably make some mistakes.
But if you combine 10 to 20, maybe 50 non-expert gamers together, you improve
your accuracy greatly in terms of analysis."
Crowd-sourcing,
the UCLA researchers say, could potentially help overcome limitations in the
diagnosis of malaria, which affects some 210 million people annually worldwide
and accounts for 20 percent of all childhood deaths in sub-Saharan Africa and
almost 40 percent of all hospitalizations throughout that continent.
The
current gold standard for malaria diagnosis involves a trained pathologist
using a conventional light microscope to view images of cells and count the
number of malaria-causing parasites. The process is very time-consuming, and
given the large number of cases in resource-poor countries, the sheer volume
presents a big challenge. In addition, a significant portion of cases reported
in sub-Sahara Africa are actually false positives, leading to unnecessary and costly
treatments and hospitalizations.
By
training hundreds, and perhaps thousands, of members of the public to identify
malaria through UCLA's crowd-sourced game, a much greater number of diagnoses
could be made more quickly — at no cost and with a high degree of collective
accuracy.
"The
idea is to use crowds to get collectively better in pathologic analysis of
microscopic images, which could be applicable to various telemedicine
problems," said Sam Mavandadi, a postdoctoral scholar in Ozcan's research
group and the study's first author.
Ozcan
and Mavandadi emphasized that the same platform could be applied to combine the
decisions of minimally trained health care workers to significantly boost the
accuracy of diagnosis, which is especially promising for telepathology, among
other telemedicine fields.
The new
UCLA study, "Distributed Medical Image Analysis and Diagnosis Through
Crowd-Sourced Games," has been accepted for publication in the journal PLoS
ONE.
In
addition to developing the crowd-sourced gaming platform that allows players to
assist in identifying malaria in cells imaged under a light microscope, Ozcan's
research group created an automated algorithm for diagnosing the same images
using computer vision, as well as a novel hybrid platform for combining human
and machine resources toward efficient, accurate and remote diagnosis of
malaria.
"The
most exciting aspect is that this is an entirely novel approach in the area of
visual diagnostics, which really challenges diagnostic algorithms used to date,"
said Karin Nielsen, a professor of infectious diseases in the department of
pediatrics at the Geffen School of Medicine. "It is diagnostics outside
the box — that is, the study introduces an entirely new concept in diagnostics
with the use of games for this purpose. The potential applications of this new
approach are immense."
How
the game works:
Before
playing the game, each player is given a brief online tutorial and an
explanation of what malaria-infected red blood cells typically look like using
sample images. After completing a short training phase, players go through the
actual game, in which they are presented with multiple frames of red blood cell
images and can use a "syringe" tool to "kill" the infected
cells one-by-one and use a "collect-all" tool to designate the
remaining cells in the frame as "healthy."
Within
each frame, there are a certain number of cells whose status (i.e., infected or
not) is known by the game but not by the players. These control cell images
allow Ozcan's team to dynamically estimate the performance of gamers as they go
through each frame and also helps the team assign a score for every frame the
gamer passes through.
"I
believe that, similar to other very innovative ideas, one of the major
challenges will be the skepticism of traditional microscopists, pathologists
and clinical laboratory personnel, not to mention malaria experts, who will
initially view with suspicion a gaming approach in malaria diagnostics,"
said Nielsen, also an author of the study. "It is a very revolutionary
proposal and it might take a few clinical studies in the field to document the
efficacy of this platform in order to convince traditional microbiologists and
other infectious disease colleagues."
"Scaling
up accurate, automated and remote diagnosis of malaria through a crowd-sourced
gaming platform may lead to significant changes for developing countries,"
Ozcan said.
"It
could eliminate the current overuse and misuse of anti-malarial drugs, improve
management of non-malaria fevers by ruling malaria out, lead to
better use of existing funds, and reduce risks due to long-term side-effects of
anti-malarial drugs on patients who don't need treatment," Mavandadi
added.
Ozcan's
team hopes to bring the platform into the field through clinical trials to help
validate its use and facilitate implementation of the technology worldwide.
Nielsen and Ozcan plan to implement it at clinical sites in countries such as
Mozambique, Malawi and Brazil.
In
addition, the same crowd-sourcing and gaming-based micro-analysis and medical
diagnosis platform could be further scaled up for a variety of other biomedical
and environmental applications in which microscopic images need to be examined
by experts, the researchers said.
No comments:
Post a Comment