The concept of machine learning or artificial intelligence (AI) assisting health care workers to make initial assessments from patient data is a new and growing field. In general, using AI could help create efficiencies in the health care system by having a software program identify results that may need further investigation by a doctor or other health professional. To develop this software, however, researchers are spending countless hours every day teaching artificial intelligence how to correctly identify patterns of diseases.
Consider one person spending 300,000 hours to complete this. Now consider 300,000 people being able to achieve the same result in one hour. This is the approach Dr. Ryan Brinkman, a bioinformatics scientist at BC Cancer, is taking. Through a citizen science project called Project Discovery, online gamers are helping scientists in the real world solve problems that require human input. Between June 2020 and June 2021, more than 23,000 online players have made over 15 million data submissions, helping AI detect patterns of cells and saving scientists approximately 240 years of research.
“This will change everything in my field,” says Dr. Brinkman. “Progress in the development of automated data analysis, especially in a clinical setting has stalled. We need to develop methods of leveraging artificial intelligence to take the next step and to do this we need training data. This project will give us a chance to significantly speed up the process of developing training data. Nothing like that exists today.”
This work, known as citizen science, is being conducted in a game within a game. In a digital, sci-fi universe where online players can mine fictional planets, engage other players in online warfare and soar through wormholes, there is an optional detour built into the game that allows players to take on a mini-challenge to unlock rewards to help them. The mini-challenge appears to the player as a task to identify patterns of unidentified cells, but in reality the cells in the game represent cells in the blood and this work, seamlessly embedded into the online sci-fi world, is teaching AI technology how to identify patterns of cells that could indicate diseases like COVID-19 or leukemia and lymphoma.
“Conventional data analysis is time consuming and not reproducible. This process of looking for patterns of cells in blood is known as flow cytometry and it’s really the only way to look at details of the immune system for signs of diseases like cancer or immunity. This will be incredibly helpful for diagnosis and to evaluate clinical trials – but only once a program knows which patterns to look for. That is where the gamers come in, to help the machines to learn which patterns need to be flagged.”
The goal is to build a pool of results that can be used to train software. The approach could save a significant amount of time and resources. It will also produce high quality training data.
Dr. Brinkman said the gamers’ performance has been impressive. His next step will be to hone this program to develop an algorithm that can pinpoint any disease from a blood sample.