Machine Learning increases the safety of diabetics while driving

Acute hypoglycaemia poses a risk for diabetic drivers, one which endangers not only their own lives but also those of other drivers. Scientists from the MTEC chairs of Stefan Feuerriegel and Elgar Fleisch are working with experts from the University of St. Gallen and the Inselspital Bern to develop a reliable system for the early detection of acute low blood sugar.

In persons suffering from diabetes mellitus, somatic cells cannot properly absorb the sugar extracted from foodstuffs. In order to counteract the resulting high blood sugar levels, diabetics take medication such as insulin. This can lead to serious complications such as hypoglycemia (a state in which blood sugar levels are acutely low). This, in turn, leads to difficulty concentrating and impaired perception, and can, in extreme cases, lead to a loss of consciousness. Hypoglycemia poses a particularly serious risk for diabetics while driving when quick reactions and decision-making are essential.

Increased road safety for diabetics

In order to reduce the increased risk of accidents for diabetics, the interdisciplinary research team is working on a method for the early detection of hypoglycemia. “The idea for the project originated two years ago as we were discussing the subject of digital health coaching with Christoph Stettler of the Inselspital,” says Tobias Kowatsch who is affiliated both at ETH Zurich and the University of St. Gallen (HSG). Kowatsch explains that the system is based on measuring the behavior of the driver. “On-board computers in modern cars already record a wide range of data. We use this data to detect behavioral changes in drivers that point to a hypoglycemic state.” For their project, which is called HEADWIND (Design and Evaluation of a Vehicle Hypoglycemia Warning System in Diabetes), the researchers were awarded a grant of 1.7 million francs by the Sinergia Programme of the Swiss National Science Foundation.

Medicine meets information technology

Sinergia grants are reserved for projects that are interdisciplinary, collaborative, and innovative. HEADWIND would not be possible without interdisciplinary cooperation. The data, which are retrieved live from the onboard computer, are analyzed in real-time via machine learning. The idea is to develop an adaptive algorithm, to integrate it into existing automobile systems, and to test the entire system under real-life conditions. This requires not only experts in information technology but also medical experts. While the first tests are planned for a driving simulator, the researchers intend for the final stage of testing to be carried out on the road.

The joint research group around Tobias Kowatsch, Stefan Feuerriegel, and Elgar Fleisch presented their initial results at the INFORMS annual meeting last October. “The results show a promising rate of detection”, explains Feuerriegel, who is responsible in the project for the machine learning algorithms. “We also found a positive indication that the system can detect critical driving patterns in previously unknown drivers.” The SNF Sinergia grant will allow the research team to continue testing the system, both in the lab and, eventually, in the field with real drivers.

Researchers collaborating on this study

  1. Professor Elgar Fleisch
  2. Professor Stefan Feuerriegel
  3. Professor Tobias Kowatsch
  4. Professor Felix Wortmann
  5. Professor Christoph Stettler
  6. PD Markus Laimer
  7. Dr. Thomas Züger

Author of this article: Sebastian Wagner, ETH Zurich.

Share this post