Tencent: ‘Medical AI is very different’
By Dzof Azmi June 11, 2019
- Plans to roll out a system across the world to detect Parkinson’s disease early on
- The current system is deployed in 15 cities in China to help with pre-screening
"MEDICAL AI is very different from AI," said Dr Wei Fan (pic), Tencent head of Medical Lab AI. "If somebody tells you that you can apply a black box model on AI for medical... I think this guy doesn't know what he's talking about"
At least Tencent is confident that Wei knows his stuff. Armed with a PhD in Computer Science from Columbia University, infused with a passion for medicine, Wei was invited in 2017 to establish the Tencent Medical Artificial Intelligence Laboratory.
Two years on, the team presented a progress report during the Tencent Global Digital Ecosystem Summit held recently in Kunming, China.
Two of their projects have now borne fruit: One is an image analysis software which will help in the fight against cervical cancer; And the other is a project that Wei is personally involved in, a system to observe and diagnose patients who have Parkinson's Disease.
Wei explained that diagnosis of Parkinson's is conventionally difficult because a doctor can only observe a patient at the clinic, when the symptoms may not be as distinct.
"For Parkinson's, almost 60% of medications are misdiagnosed," he clarified. On top of that, the doctor may not be sure how effective the drug he is giving is until he sees the patient at the hospital again.
Observation beyond a clinic’s walls
Wei's solution is to install a camera in the home and let AI analyse the footage. "It's based on your daily movements and (looks for) differences before and after," explained Wei. "For example you will know before and after taking a drug how the motion function improves, how quickly it improves and how long the improvement will last."
However, training the software to recognise the tell-tale erratic movement was challenging; There simply wasn't enough data available in footage from hospitals and clinics to create an effective learning set.
Instead, Wei had to somehow produce his own data. He created a computer model of the human hand, taking into account real-life physiological constraints, and then used that to help generate data. "So when we see a patient, we map the virtual glove on to the actual hand," he said.
This was not a trivial task. "You have to understand human anatomy and the human body, as well the physiology of the humans."
Medical AI is regulated
Another reason why Medical AI is difficult to implement is because it is regulated. Specifically, they have to conduct Random Control Tests (RCT) to satisfy requirements by the FDA in China, and solutions may need to undergo medical trails first.
"For medical AI to help patients and doctors we need to really to follow the traditional the regulation framework," said Wei. "We follow the very strict standard in medical field"
Wei is optimistic about the possibilities of analysing videos to detect not just Parkinson's but a whole range of ailments.
"Parkinson's is only the starting point we are also scaling up to other diseases as well, which we couldn't do because we don't have the technology (yet)". Wei confirmed the team is already looking at the early detection of multiple sclerosis, as well as other chronic diseases such as cerebral palsy, and acute coronary syndrome (ACS).
Wei envisages that such sensors can be put in place in, say, an old persons’ home, and help to mitigate other issues, such as common accidents affecting the elderly. "They have also talked about using our gait analysis for preventing seniors from falling," enthused Wei. "If you think they have a chance to fall, you (jolt them with) a little bit of electricity and they will freeze!"
A lot of this is still in the future. For the time being, Wei’s hands are full, with the current system deployed in 15 cities in China to help with pre-screening, which will hopefully roll out to the rest of the world.
"We're going to launch a multi-country multi-centre trial involving not only Asia but also other Western countries as well,” he announced, including London and Hong Kong. “That will give us a lot more confidence that what we're doing is not limited to only one region (in how it works).”
Dzof Azmi reports from the Tencent Global Digital Ecosystem summit in Kunming, China, at the invitation of Tencent. All editorials are independent.