AI-supported smart home technology for in-home care

Interview with Prof. Thorsten Jungeblut, Department of Industrial Internet of Things, Bielefeld University of Applied Sciences (HSBI)

The use of artificial intelligence (AI) is intended to ensure greater safety and independence in home care. In an interview with MEDICA-tradefair.com, Prof. Thorsten Jungeblut describes how data from numerous sensors in the KogniHome research apartment in Bielefeld records the situation at home and identifies any need for action.

Prof. Jungeblut, what data is collected?

Prof. Thorsten Jungeblut: Essentially, we collect movement information and create movement profiles. We use minimal sensor technology instead of cameras and microphones, which collect sensitive data. We try to generate added value by combining different sensors. The very simple sensors such as motion detectors or door contact sensors are data-saving and only collect a single piece of information. However, pieces of furniture or household appliances such as a refrigerator can also be equipped in this way. This allows us to recognize with what a person is interacting.

We can detect the power consumption of appliances using adapter plugs and very precise radar-based motion detectors tell us where people are. We can track walking routes and even recognize whether someone is lying in bed and breathing. All of this is done while maintaining privacy.

Do you also collect data with wearables?

Jungeblut: We use smartwatches. While these have to be on at all times, the smart home sensors are always active, even when going to the toilet at night. The more sensors we have, the more accurate the information is. Part of our research also focuses on the question of how much data we need. The aim is to detect activity and anomalies. For activity detection, we use a smartwatch to obtain additional physiological parameters such as heart rate.

What is the most important data for eHealth?

Jungeblut: We pay particular attention to how behavior changes compared to the past. The focus is on major changes, apart from seasonal trends or changes on the weekends.

We are currently developing an app for relatives that uses a smiley face - sad, medium or happy - to indicate whether everything is OK. This saves check-up calls and leaves more room for social interaction. People don't want to feel like they are being watched, which is also the feedback we get in conversations with older people.

Do you need large amounts of data to train the AI?

Jungeblut: The amount of data collected is not very large, but the longer you record data, the better. We use simple sensor kits with up to 15 sensors. Motion detectors and door contact sensors don't generate a lot of data. But they are particularly worth protecting, especially when used over a long period of time. The aim is to only process the data locally; personal data should not even leave the home.

At best, monitoring starts very early on when people are doing well. If you notice a decline in activity after a few years, this can be an indication of neurodegenerative diseases such as Alzheimer's, Parkinson's or even depression. Computing resources to process the data locally are more expensive, but safer than centralized storage.

That is why we check whether low-performance devices are sufficient. Alternatively, there are special encryption methods (e.g. homomorphic encryption) for smart home users to transfer data to external service providers for analysis.

Shouldn't the data for telemedicine and better informative value be linked to diagnoses?

Jungeblut: We are at the beginning of research in this area, and doctors need to be on board. For example, there is a promising Alzheimer's drug (lecanemab), but it can only be used in the very early stages. The hope in medicine is that our data will provide very early indications of neurodegenerative diseases.