Edge Intelligence for Connected In-home Healthcare: Challenges and Visions

Yuchen Zhao, Hamed Haddadi, and Payam Barnaghi
March 17, 2021

 

Over the coming decades, we will face an increasing number of challenges in the public health sector. Longer life expectancy brings a large aging population that requires more daily healthcare than before. Population living with long-term conditions such as dementia, which decrease their performance of daily activities, need not only extra medical resources but also social resources for their healthcare. Global-scale pandemics such as coronavirus disease 2019 (COVID-19) can cause unprecedented challenges that overwhelm the healthcare system that we currently have.

Technologies such as the Internet of Things (IoT) provide new opportunities to help meet the soaring demands in public healthcare. Morden commercial IoT devices can collect a variety of sensory data, ranging from simple ambient data to complex physiological data. By deploying millions of these IoT devices at people's homes and running machine learning (ML) algorithms on these data at the edge of the network, we can closely monitor people's health status and provide both short-term and long-term in-home healthcare services.

In-home Healthcare Powered by IoT and ML

Deploying IoT devices at home can help us collect different types of sensory data that indicate changes in patterns of people's daily activities associated with their health and well-being. For instance, passive infrared (PIR) sensors can monitor the presence and movement within the home. Smart plugs can record the way individuals use different appliances. Wearable devices can measure main vital signs such as body temperature and heart rate. All this environmental and physiological information can be used to analyze patterns of activities and analyze healthcare-related incidents with predictive models and adaptive algorithms. For example, training deep-learning models on time-series data [1] can extract patterns from the data and predict the potential risks. This will provide a new approach for proactive and predictive models of in-home care and health monitoring. Figure 1 depicts an intelligent healthcare system that runs ML algorithms both on sensory data on edge devices and on processed data on a Cloud server.

Figure 1: An intelligent healthcare system running ML algorithms both at the edge and on the Cloud.

Figure 1: An intelligent healthcare system running ML algorithms both at the edge and on the Cloud.

What Are the Challenges?

IoT systems in real-world deployment have heterogeneity at different layers of the network and system. For example, IoT devices made by different manufacturers may not use the same protocol or interface. The same types of devices may not use the same data format or sampling frequency, which causes issues when integrating them into the system. The set of deployed devices may vary from one household to another due to many factors such as budgets, demands, and layouts, which increases the complexity of designing and implementing analytical algorithms. These issues are from different system levels and exist during different stages in the system's lifecycle. Thus, different solutions, including standardization and deploying edge processing solutions are needed.

Another major challenge is the privacy issues of collecting data from people's homes and processing these data, especially when the data contain sensitive and personal information. Communicating the data to a Cloud server to train and use ML models leads to risks of exposing this information to untrusted parties. To enable privacy-aware processing, edge intelligence, and local processing is a preferred choice for applications that require a higher level of privacy and more control on the data. However, providing edge intelligence on an isolated node in healthcare applications will limit analytical models from accessing and learning from wider and population data in many use-case scenarios. There are several approaches to address this issue. One approach is to use federated learning (FL) [2], which allows different clients to jointly train a model without releasing their raw data and use the trained model locally.

Apart from technical challenges, people's ability to interact with IoT devices may also cause issues when running the system. Using and managing IoT devices with limited interfaces can be challenging. Therefore, it is likely that data quality may vary. Issues such as missing data or missing labels may happen regularly in the system. A data analysis system for such applications requires the analytical models to be robust to process data with varying qualities and missing values.

What Are the Opportunities?

The growing capability of sensors is one major factor that brings new opportunities. Although current commercial IoT devices mainly focus on ambient and physiological data, devices that can sense more complex data are being developed and available to deploy. For example, high-definition activity recognition with radio signals [3] can detect human activities through walls and may replace PIR sensors in the future. Medical checks and electroencephalogram (EEG) measurements are becoming available at home on low-cost IoT devices. These devices will broaden the definition of “sensory data” and enable several new healthcare services and applications.

The computational power of IoT edge units also keeps growing. Consequently, we can conduct complex machine learning tasks on IoT sensory data at the edge. These tasks use data with higher dimensionalities and bit rates, such as live videos, which can provide more accurate and fine-grained healthcare monitoring. With machine learning frameworks customized for IoT edge units, it is possible to conduct real-time monitoring and healthcare-related emergency detection.

With privacy-preserving data analytics running on a local node, a system can routinely extract different digital biomarkers from the raw data stored at the edge and use the biomarkers in different applications such as vital sign measurements and analysis, which traditionally can only be done in clinical settings.

Conclusions

In-home healthcare using IoT technologies and ML algorithms will play an important role in reshaping healthcare systems shortly. It can greatly improve the quality of life for people living with long-term conditions and provide continuous health and activity monitoring. Edge intelligence provides an opportunity to create more privacy-aware and efficient solutions for analyzing sensitive and personal information and potentially giving more control to the end-users on their data. However, to achieve this, we not only need robust technical solutions, and we also need participation from other related sectors, including clinicians, social and community services, user groups, to develop end-to-end solutions that are clinically safe and socially acceptable. At the Care Research & Technology (CR&T) center[1] of UK Dementia Research Institute (UKDRI), scientists, engineers, and doctors work together to utilize the cutting-edge technologies in artificial intelligence, engineering, robotics, and sleep science to build new healthcare systems [4] [5] [6] that will create dementia-friendly healthy homes.

[1] https://ukdri.ac.uk/centres/care-research-technology

References

  1. S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol. 9, no. 8, pp. 1735-1780, 1997.
  2. H. B. McMahan, E. Moore, D. Ramage, S. Hampson and A. Blaise Aguera y, “Communication-Efficient Learning of Deep Networks from Decentralized Data,” in Proceedings of the 20th International Conference on Artificial Intelligence and Statistics, 2017.
  3. M. Zhao, T. Li, M. Abu Alsheikh, Y. Tian, H. Zhao, A. Torralba and D. Katabi, “Through-Wall Human Pose Estimation Using Radio Signals,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018.
  4. S. Enshaeifar, P. Barnaghi, S. Skillman, A. Markides, T. Elsaleh, S. T. Acton, R. Nilforooshan and H. Rostill, “The Internet of Things for Dementia Care,” IEEE Internet Computing, vol. 22, no. 1, pp. 8-17, 2018.
  5. S. Enshaeifar, A. Zoha, A. Markides, S. Skillman, S. T. Acton, T. Elsaleh, M. Hassanpour, A. Ahrabian, M. Kenny, S. Klein, H. Rostill, R. Nilforooshan and P. Barnaghi, “Health Management and Pattern Analysis of Daily Living Activities of People with Dementia Using In-home Sensors and Machine Learning Techniques,” PLOS ONE, vol. 13, no. 5, 2018.
  6. Y. Zhao, H. Haddadi, S. Skillman, S. Enshaeifar and P. Barnaghi, “Privacy-Preserving Activity and Health Monitoring on Databox,” in Proceedings of the Third ACM International Workshop on Edge Systems, Analytics and Networking, Heraklion, Greece, 2020.

 

Yuchen ZhaoYuchen Zhao is a Research Associate in the Dyson School of Design Engineering at Imperial College London. He is also with the Care Research and Technology Centre at the UK Dementia Research Institute. He obtained an MSc and BEng both in Information Security from Wuhan University and Huazhong University of Science and Technology, respectively. He studied for Ph.D. in Computer Science at the University of St Andrews. His research revolves around privacy protection and includes topics such as usable privacy, user experience in Human-Data Interaction, and privacy metrics. Currently, his research focuses on privacy-preserving data analytics and federated machine learning on distributed IoT platforms.

 

Hamed HaddadiHamed Haddadi is a Reader in Human-Centred Systems and the Director of Postgraduate Studies at the Dyson School of Design Engineering at The Faculty of Engineering, Imperial College London. He leads the Systems and Algorithms Laboratory and is an Academic Fellow of the Data Science Institute. He is also a Visiting Professor at Brave Software where he works on developing privacy-preserving analytics protocols. He is interested in User-Centred Systems, IoT, Applied Machine Learning, and Data Security & Privacy. He enjoys designing and building systems that enable better use of our digital footprint while respecting users' privacy. He has spent time working and collaborating with Intel Research, Microsoft Research, AT&T Research, Telefonica, and Sony Europe. When not in the lab, he prefers to be on a ski slope or in a kayak.

 

Payam BarnaghiPayam Barnaghi is Chair in Machine Intelligence Applied to Medicine in the Department of Brain Sciences at Imperial College London. He is Deputy Director and Group Lead in the Care Research and Technology Centre at the UK Dementia Research Institute. He is an associate editor of the IEEE Transactions on Big Data and vice-chair of the IEEE SIG on Big Data Intelligent Networking. His main research goal is to develop AI and machine learning solutions for healthcare and create affordable and scalable digital systems that can be applied across a range of health conditions. He works on machine learning, the Internet of Things (IoT), semantic computing, adaptive algorithms, and computational neuroscience to solve problems and develop new technologies for future healthcare systems.