Managing diabetes with diafyt thanks to self-learning and smart technology

We all make mistakes from time to time or forget to do things. But people with diabetes can have serious health problems if they miscalculate or inject the wrong insulin dose. A research team from the Dresden University of Technology (TU Dresden) and a startup from Leipzig aim to make living with diabetes easier for patients thanks to a combination of self-learning AI and a high-tech insulin pen based on innovative semiconductor sensors to measure glucose levels.

In this MEDICA.de interview, D. Eng. René Richter, Institute of Precision Engineering and Electronic Design at TU Dresden, explains how the system works.

Dr. Richter, how do the individual components – meaning the insulin pen and algorithm – interact with each other?

Eng. René Richter: We all make mistakes and sometimes lose our keys or forget to turn off the iron. People with diabetes are no exception. Wouldn’t it be nice if there was a tool that discreetly monitors us and sends a message if we need to pick up an insulin pen that we left at a restaurant or missed an insulin injection?

A smart insulin pen can help people better manage their diabetes through monitoring. The device features high-tech sensors, a microcontroller and wireless connectivity. Software on your smartphone or smartwatch can calculate the injected dose, insulin temperature, and location, so it’s no longer a showstopper if you misplace your insulin pen. Simply open the Find My app or say, "Hey Siri, find my insulin pen."

The insulin pen improves the dosing accuracy and the quality of the input data for the algorithm, thus optimizing its work.

Insulin is administered based on a person’s respective blood glucose levels. How can the algorithm help prevent insulin dosing errors?

Richter: Even an experienced person with diabetes occasionally has trouble calculating the optimal insulin dose. Yet errors can lead to hyperglycemia or hypoglycemia, which can ideally still be corrected, though this entails risks. Patients often ride the blood sugar roller coaster.

Artificial intelligence in upstream operations ensures the algorithm observes and learns about the individual’s insulin demands. After a few short days, the algorithm has adapted and provides accurate dosing suggestions.

This prevents and regulates hyperglycemia and lowers the HbA1c level.

What factors does the algorithm use to calculate the accurate insulin dose?

Richter: The input data for the suggested dose includes the current time, the consumed number of carbohydrates and the recently taken insulin dosage. The algorithm also uses a personal user profile that is calculated via blood sugar levels, insulin doses, and entered meals.

Do users still have to adjust or check, or is this an autonomous diabetes management system?

Richter: Thanks to artificial intelligence, the algorithm is self-learning and automatically makes adjustments. The algorithm does a lot independently. It is fast and efficient, but it is not designed to fully replace the decision-making process on the part of the user. Diabetes is a complex multifactorial disease, which means there are other influences that are not being mapped.

What role do semiconductor sensors play in blood sugar monitoring?

Richter: Semiconductors are more advanced than conventional sensors. Apart from blood glucose, they can also track other components in the blood, such as ketones. This results in better treatment, especially for people with type 2 diabetes.

The algorithm was designed to calculate the optimal insulin dose for type 1 diabetics. Type 2 diabetes can be managed without insulin, but could people with type 2 diabetes still benefit from the pen and algorithm?

Richter: People with type 1 and type 2 diabetes can benefit from the diafyt smart insulin pen since the smart functions are equally medically effective in this setting. However, the algorithm and the AI ​​pertaining to the insulin dose calculation were specifically developed to address the intensive insulin treatment in type 1 diabetes. Our next step is to also build an algorithm to target type 2 diabetes.