Fei-Fei Li, Stanford computer science professor and co-director of Stanford’s Human-Centered AI Institute (HAI), shared her thoughts on possible AI technologies that could help care for the seniors during the coronavirus pandemic at today’s COVID-19 and AI: A Virtual Conference.
“This is a very good example of basic science research ramping into real-world applications in a time of crisis,” Li said. “We’ve spent the last several years looking at how a set of AI technology could help ageing seniors to live more independently and cope with chronic disease management. But recently, we realize the same technology for longer term care might also be helpful for seniors in this acute pandemic.”
The HAI hosted and livestreamed virtual conference was organized to explore the impact of COVID-19 on society and how AI can be leveraged to increase understanding of the virus and its spread. Respected researchers from both academia and industry discussed AI applications in diagnostics and treatment, epidemiological tracking and forecasting of the spread of the virus, information and disinformation, and the broader human impact of COVID-19 and pandemics in general on economies, culture, government, and human behaviour.
The USA’s Centers for Disease Control and Prevention (CDC) says that based on currently available information and clinical expertise, older adults and people who have serious underlying medical conditions are at higher risk for severe illness from COVID-19.
Li pointed to a number of contributing factors — more seniors have existing health conditions that increase vulnerability, and seniors’ community living may contribute to spread of the highly infectious virus. Seniors also interact with caregivers who tend to be younger and may carry the virus without symptoms. Li also referenced challenges in medical triage or interruptions in chronic disease treatment due to overwhelmed healthcare systems.
Li identified AI-powered smart home sensor technology as a possible way to help families and clinicians remotely monitor housebound seniors for infection symptoms or symptom progression or regression and potentially help manage their chronic health issues.
The first step is collecting data by putting sensors in the home. There are various kinds of smart sensors, which complement each other in modality and the kind of data they capture. Li says the considerations with respect to privacy and security when handling data from various sensors are front and centre for her team.
Cameras are versatile sensors that can capture a lot of detailed and useful information on a person’s activity. But cameras pose greater privacy risks than depth sensors, thermal sensors, wearable sensors, etc.
Sensor data is then transferred to secure central servers, where machine learning models are trained to recognize clinically relevant patterns, including respiratory, sleep, dietary and other behaviours. When deployed, these models can be installed directly on edge devices, so that personal data need not leave the home. The last step is to set up an interface for caretakers and families to access detection results.
Li also presented a few examples of the kind of information the sensors can gather, such as fever detection or respiratory pattern recognition using thermal sensors, which can be useful for early detection of infection. Understanding human movement is also an important clinical indicator for many relevant illnesses, and Li said the team’s preliminary work showed high accuracy to discern different types of movements. Other examples she gave are sleep pattern analysis and dietary patterns from fluid intake to pill consumption.
Li underlined that the solutions she presented are not meant to perform diagnostic decision-making or replace clinicians, but rather to help keep an eye on ageing populations living at home and send timely alerts to clinicians and families as appropriate. This can be of critical assistance in efforts to maintain physical distancing for a vulnerable population while mitigating hospital overload.
“It’s really important that we don’t overhype any technical tools, and it is important to recognize the limits of machine learning and AI in today’s capabilities,” said Li.
Her talk was part of the conference’s Treatments & Vaccines session, which also touched on machine learning enabled systems for delivering care to critically ill patients, identifying COVID-19 vaccine candidates with machine learning, repurposing existing drugs to fight COVID-19, and other topics.
Kaggle founder and CEO Anthony Goldbloom also spoke in this session, on applying ML technologies to COVID-19 related research. He said while ML may be useful in certain areas, it can be “pretty much useless” in others. He said he has observed that even a small number of domain experts are having a huge influence on the ML community in the ongoing COVID-19 challenges Kaggle is hosting.
Goldbloom said a project where machine learning could help is in fine-grained analysis of differences in transmission rates across cities. He noted however that it is unreasonable to expect ML to predict the impact of a policy that is only going to be applied next week.
A video recap of the virtual conference is here.
Journalist: Yuan Yuan | Editor: Michael Sarazen