Determining which treatment of many will help someone with depression is a job for AI. Aifred Health is using a rigorous approach to solving the problem as part of IBM’s Watson AI XPRIZE competition.
Depression is a silent killer, as the mental health condition affects millions of people worldwide. It’s also the subject of a research project competing for the IBM Watson AI XPRIZE, which will be awarded in April 2020. Aifred Health, which was named a leader among the XPRIZE teams, is working on using machine learning to improve the efficacy of mental health treatments.
Teams race to the top
In November 2018, the top 30 teams competing for the IBM Watson AI XPRIZE were named. Last month, the current top 10 teams out of 62 were announced.
“This was out of 147 initial teams, based on the judges’ evaluation of progress made and how they’re going to use AI to work with humans,” said Amir Banifatemi, general manager for innovation and growth at XPRIZE. “Every year, we recognize 10 teams — not necessarily finalists — for their effort.”
At the NeurIPS event in November, Aifred Health received a bonus milestone prize of $35,000. The Montreal-based team is developing AI-based tools for doctors to select the best treatments for each patient.
In second place was Choosito, which won $20,000. The Philadelphia-based team is using AI to automate the indexing, analysis, and searching of digital content.
Nectar Technologies won a People’s Choice Award of $5,000 for helping beekeepers use precision technology to protect and build their colonies.
“We’ve not yet asked teams to showcase prototypes,” Banifatemi told Robotics Business Review. “Some have gathered data, some are working on algorithm capabilities, and others on protocols and the capture of data.”
Diverse competitors over time
Using machine learning to treat asthma is different from monitoring bee populations or diagnosing responses to opioids.
“With the diversity of teams, they won’t have prototypes ready in the same time frame, so we didn’t require them for this milestone,” Banifatemi said. “Teams must set their own milestones for progress — they had to propose 3.5-year road map, and the judges evaluated them based on their own proposals.”
“When we designed this competition, we realized that technology matures over time,” he said. “Teams that enter [by the Dec. 14 deadline] may be different than those that entered the previous year.”
“We have a good balance between teams, with good finds entering later,” said Banifatemi. “If they think they have what it takes, they can take on teams that have had more time to develop.”
Aifred Health addresses depression
“Aifred Health has created a strong partnership with universities, enrolling 20 students,” Banifatemi said. “With this project, they’ve created a model that’s relatively pragmatic on how to manage doctor outcomes and understanding of depressive modes.”
“More than 320 million people in the world — or about 1 in 9 — suffer from depression,” said David Benrimoh, chief science officer at Aifred Health. “It’s the biggest set of disabilities on the planet, costing $1 trillion per year.”
“The biggest problem in depression treatment is that it takes a long time for people who do seek care take to get better,” he said. “It can take seven to nine months to get better with some combination of psychotherapy, exercise, and medication.”
“We’re working with a range of experts, from business and neuroscience to psychology,” Benrimoh told RBR. “We’re focusing on helping physicians make better decisions in mental health. There are particular decision points that need to be improved, and we’re doing something that’s not currently possible without AI.”
“Patients often go from one therapy to another, so the approach of helping doctors to better evaluate routes to treatment has huge potential,” Banifatemi said. “Working with multiple universities and researchers further validates the thinking that there’s not one single solution.”
“Standard therapies are roughly equally effective, but different people respond differently to different things,” Benrimoh noted. “We may have no idea why they work, and classical statistics have not been able to figure it out.”
Noninvasive data mining
“With AI predictions from more complex mining of data, doctors can decide which treatment to start on, depending on an individual’s characteristics,” he said. “Even before turning to genetics or EEGs, we can use demographic and clinical measures first.”
“We haven’t yet partnered with genetic testing companies, but there are established thresholds of trending,” said Benrimoh. “You don’t need AI; you can just trigger decision points, but AI will learn from that data.”
“We want to make a clinical decision-support system accessible,” he said. “The Vulcan AI tool is built on top of Pytorch, and we can use that data for training models.”
“So far, using deep learning and the data we have access to, we can not only see if a patient will get better,” Benrimoh said. “We can also select among treatments and predict what they’ll get better with.”
“We’ll package that into an app to help patients and physicians track their symptoms over time,” he added. “It’s difficult to know how we were feeling at a given moment, so this enables better communication.”
The XPRIZE experience
“We exist because of the XPRIZE,” Benrimoh said. “A few of us were students at McGill University and had a won a few competitions together. We were interested in AI and mental health, so when the prize was announced, we formed a team.”
“The judges really value the scientific and social value of what we’re working on, and it’s interesting working with a range of judges, because of the range of projects,” he said. “They are domain experts who know the technology and clinical practices well. In one report manuscript, one judge was quoting recent literature in the field back to us. That’s really appreciated.”
“This past year’s theme was evaluation — how will you evaluate your own technology and show benefits through project planning?” said Benrimoh. “This coming year, it’s implementation and real-world testing.”
“We have a study with just doctors, but we were proud to get ethics approval for human tests,” he said. “We want to conduct our initial studies within a year and prove its effectiveness by 2020.”
“We’re piloting just the data collection at a hospital in Montreal,” Benrimoh said. “Within a few months, we’ll have our first product — a rule-based, decision-support tool that will help us collect data, improve data sets, and train the model.”
Reasons for being named No. 1
“I don’t have any insight into the judging process, but here are a few reasons why I think we got the top spot this year,” said Benrimoh. “First, there’s the size and definition of our problem — depression affects everyone, directly or indirectly. It’s the biggest cost in terms of lost productivity and the impact on families. As other causes of death are falling, suicide is rising.”
“Also, the simplicity of articulating how AI can help make a specific decision better in a well-defined workflow is very straightforward,” he said. “We’re not making something cheaper or faster; we’re making it possible.”
“Finally, we have a great interdisciplinary team,” he said. “We have a five-person core team of two clinicians, two neuroscientists, and one machine-learning expert, plus about 15 other people. Some are programming machine learning, some volunteers are researching medical literature, and others on the business side are doing forecasting. That’s 22 people total, plus collaborators across multiple continents.”
“Our partners gave judges the confidence that we’ll be able to do what we said in time,” Benrimoh concluded. “We can have a positive impact.”
“Aifred’s project was well-organized; the way they approached the problem was really impressive to the judges,” said Banifatemi.
(Editor’s note: This is part one of a two-part profile of some of the top teams in the $5 million IBM Watson XPRIZE competition).
