AI and diabetes: a smart alliance with potential

Diabetes is one of the greatest health challenges worldwide. Despite significant advances in care, the number of cases is increasing. According to forecasts, the number of diabetics could rise to around 1.3 billion by 2050. In Germany alone, over eight million people currently suffer from diabetes. To stop this worrying trend, scientists are increasingly relying on the latest technologies. They are developing preventive and innovative approaches to early detection and therapy. In the following, we show how AI can positively influence prognosis and what new possibilities are emerging.

Diabetes challenges those affected every day. Sudden blood sugar fluctuations and the fear of hypoglycaemia require constant attention, planning and usually also a subtle insecurity in everyday life. The latest developments based on artificial intelligence (AI) offer promising approaches to daily diabetes management – whether it's smart applications that calculate the optimal insulin dose or monitoring systems that recognise symptoms before they have a dramatic effect, such as the onset of hypoglycaemia while driving.

Early diagnosis using deep learning models

The first signs of diabetes are often subtle and vary depending on the type. In type 1 diabetes, symptoms such as extreme thirst or frequent urination occur suddenly. In contrast, type 2 diabetes develops gradually. Early warning signs such as fatigue, visual disturbances or poorly healing wounds often go unnoticed.

AI-based methods for analysing health data could identify the risk of disease at an early stage and take countermeasures. This is suggested by a study from 2023.

While traditional machine learning models often only look at individual values at a specific point in time, deep learning approaches can incorporate continuous changes – such as long-term elevated blood sugar levels. This makes it possible to assess risks more accurately.

Diagnosing diabetes using voice or skin analysis

Canadian researchers have developed an AI that can recognise diabetes from a person's voice rather than from blood tests. The disease causes changes in the sound at an early stage. These changes are not perceived by our hearing, but they are by the AI. Researchers suspect that blood sugar concentration is linked to the elastic properties of the vocal cords.

Another AI-based method for the early detection of diabetes has been developed by researchers at the Technical University of Munich (TUM) and Helmholtz Munich. It combines optical and acoustic principles to create a detailed image of the skin. The RSOM (raster-scan optoacoustic mesoscopy) method shows that diabetes affects blood vessels in different skin layers in different ways. 32 specific features identified with the help of AI allow for an accurate determination of disease progression. One advantage of this method is that diagnosis is quick and monitoring can be done without blood samples.

Valuable support in the daily lives of people with type 1 diabetes

Close monitoring using data-driven systems is crucial for patients with type 1 diabetes. It enables real-time adjustments to be made and helps to avoid complications. Diabetes management app Daily Dose is an example of this. It helps patients with diabetes to manage their daily insulin intake and offers more control over their condition 24/7. The app also provides nutritional tips and helps to avoid hyperglycaemia. If there are indications of a sharp drop in blood sugar overnight, those affected can react in time. This provides enormous relief for those affected, who previously feared hypoglycaemia during the night. The necessary data is collected by a smart insulin pen and compared in the system. In total, three machine learning algorithms are integrated into the app, which together ensure that users can make the right and important decisions for their insulin intake. The Diafyt app offers a similarly personalised health management service. It also uses a connected pen to determine individual insulin requirements and provides a dosage recommendation. The app not only combines various activity and vital data, but also ‘learns’ from past values, helping to maintain stable blood sugar levels.

Lifestyle and prevention of type 2 diabetes

Genetic factors, unhealthy nutrition and a poor lifestyle are among the main causes of type 2 diabetes. Often, a combination of these factors leads to the onset of the disease. Obesity, lack of exercise, hereditary factors and smoking have a negative impact on insulin levels. A healthier lifestyle and weight reduction can improve or even reverse the disease. People with type 2 diabetes are already benefiting from apps. One example is the digital health application Glucura, a kind of diabetes coach for your pocket. It offers personalised tips on nutrition, exercise and weight loss. With these approaches, sufferers can actively work to restore their health and stabilise their blood sugar levels in the long term. During the introductory phase, you wear a glucose sensor on your arm and keep a diary of your eating and exercise behaviour in the app. The AI behind the app then provides continuous feedback as a personal coach.

Will AI help to cure diabetes?

AI-supported diabetes research is already much more than a mere pipe dream. One example is the recently awarded project for researchers at Helmholtz Munich. It aims to find new ways of treating type 1 diabetes through the interaction of AI and stem cell biology.

In fact, stem cell research has already produced a promising approach for type 1 diabetes. This could potentially cure patients in the future. The VX-880 therapy is based on insulin-producing islet cells that are grown from stem cells in the laboratory and infused directly into the hepatic portal vein of the participants. This is how the natural production of insulin can be restored. The study results so far are promising: after 180 days, seven out of a total of twelve participants no longer needed daily insulin. Two participants were able to reduce their insulin requirements by 70%, and one participant needs 24% less insulin. However, it remains to be seen how the upcoming tests turn out. Although there is no explicit evidence that artificial intelligence is being used in the project, AI-supported procedures in stem cell research are likely to increase. They help to precisely identify and isolate cell types.

Digital twins in diabetes treatment

Digital twins have long been used in urban planning, in construction and in the preservation of cultural heritage. They could also gain relevance in medicine, especially in diabetes care. The digital twin of a person virtually simulates individual health data. This could help to identify risks such as hypo- or hyperglycaemia even faster and to adapt therapies even more individually. According to a PwC study, as many as 72 per cent of diabetes patients surveyed would support the use of a digital double.

Despite the high level of acceptance in the survey, data security remains a key concern. 89 per cent of respondents want clear data protection rules before the technology is used. In addition, the question of who will bear the costs remains to be answered.

One thing is certain, however: improved diabetes management through advanced AI technology could be beneficial in the long term for both those affected and our currently vulnerable healthcare system.

Text: Alexa Brandt