Average Joes challenge medical pros in telepathology game
Researchers in the Ozcan Research Group at UCLA have taken telepathology-which enables pathology in a remote location-into the video-gaming world.
Researchers in the Ozcan Research Group at the University of California, Los Angeles (UCLA) have taken telepathology—which enables pathology in a remote location—into the video-gaming world. Using malaria as their model, their approach, which involves average gamers with no medical training, reported diagnostic accuracy within 1.25% of a trained medical professional.1 The researchers' vision with this work is to enable remote pathological disease diagnosis without requiring a medical professional—a benefit to remote, resource-poor locations of the world without immediate access to hospitals or clinics.
The research team, led by Sam Mavandadi, a postdoc in the Bio-Photonics Laboratory at UCLA, and Aydogan Ozcan, Ph.D., associate professor in the Bioengineering and Electrical Engineering Departments at UCLA, designed a digital gaming platform that allows gamers from anywhere in the world to diagnose optical microscopy images of human red blood cells (RBCs) potentially infected with P. falciparum (the malaria parasite).1 The research recognizes that pathologists typically check between 100 and 300 different fields of view (FOVs) of a thin blood smear (at least 1,000 individual RBCs) using a light microscope with a 100X objective lens before declaring the sample negative, which takes a long time and is difficult to accomplish in remote locations.
|The BioGames gaming platform developed by researchers at UCLA uses the talent of gamers to diagnose malaria with accuracy with 1.25% of a medical professional. What's more, it can be played on a browser or Android device. (Image courtesy of the Ozcan Research Group at UCLA)|
Currently in proof-of-concept stage and called BioGames, the platform is designed for play on a Flash-enabled browser or Android smartphone or tablet, and comprises a "database" of digital RBC slides scanned with a brightfield optical microscope (see figure). At each FOV, the captured images were then digitally labeled with positive or negative labels, yielding a dataset of 7,116 unique images—1,603 of which were infected by the malaria parasite. To make it even more challenging, the researchers integrated another 118 infected and 595 uninfected images from the Center for Disease Control (CDC; Atlanta, GA).
The key to BioGames is that non-expert gamers have to play, as the platform relies on online crowd-sourcing. In the study, the gamer participants (as many as 30) were required to complete, with >99% accuracy, a training version of the game. Then, those who completed training successfully could begin the actual game, viewing multiple frames of RBC images and having different controls within the game to "kill" the infected cells one by one or designate the remaining cells in the current frame as "healthy." Once they completed a frame, the game assigned them a score based on their performance only on the control images (roughly 20% of all the images), allowing the game to provide feedback so that as the gamers continue to play, they can improve their diagnostic ability.
With more and more gamers, the results become even more accurate, notes Ozcan. For instance, with only 20 gamers, the accuracy of the results came within 1.25% of a medical pro.
So, what's next? Ozcan told BioOptics World that their platform can surely be extended to other diseases where medical imagers are used to look at the signatures of different types of cells and bacteria. "As an example, it can also be used to count CD4 or CD8 T cells for monitoring of HIV-positive patients," he says.
1. S. Mavandadi et al., PLoS ONE, 7, 5, e37245 (2012).