A group of scientists from the National Institute of Technology (NIT) Rourkela led by Prof. Mirza Khalid Baig, assistant professor in the Department of Biotechnology and Medical Engineering, has developed a novel AI-based method intended to enhance predictions of blood sugar for effective control of diabetes. This innovation can potentially revolutionize the way diabetes is treated by patients and healthcare professionals, allowing for easier informed treatment choices and compatibility with intelligent health devices such as mobile phones.

The study, co-authored by Prof. Baig and his research scholar Deepjyoti Kalita, was published in the IEEE Journal of Biomedical and Health Informatics. Their research offers a new machine-learning model intended to improve the precision of predictions of blood glucose levels, which is essential in diabetes management. The creation of this model seeks to overcome the difficulties of those with diabetes and their doctors through improved tools to make effective and timely decisions on treatment.

Care for diabetes, especially in nations such as India, is increasingly becoming significant. As per the ICMR INDIAB study published in 2023, the total prevalence of diabetes in India is 11.4%, whereas prediabetes is present in 15.3% of the population. With diabetes levels so high, it is essential to create new-age solutions that can better control the disease and make it more efficient. Prof. Baig and his research team provide one such solution through the use of advanced digital health technologies, especially Artificial Intelligence (AI).

Machine learning has been applied to diabetes research with a variety of purposes, some of which aim to create predictive tools to facilitate better decisions from doctors and patients. Still, most of these AI-based models have certain shortfalls, such as when dealing with predictability and interpretability. Among the main concerns related to most of the predictive AI models is the fact that these models act like a “black box,” the predictions of which are not readable or interpretable. This opacity makes it challenging for both the doctors and the patients to entirely trust these tools.

Also, conventional models like statistical forecast techniques or simple neural networks lack the capability to handle long-term glucose fluctuations and need fine-tuning of considerable complexity in order to function efficiently, thereby compromising on their efficacy in actual implementation. The proposed AI model designed by researchers at NIT Rourkela aims to get around such problems by enhancing glucose forecasting accuracy.

The researchers’ method involves deep learning methods, namely a dedicated AI model that learns from past blood sugar data and forecasts future glucose levels with higher accuracy. The model automatically processes glucose data, recognizing important patterns and making precise predictions without the necessity of manual tuning. In contrast to traditional models that have difficulty with long-term trends, the NIT Rourkela team’s model can recognize and predict long-term glucose variations, offering more accurate and trustworthy predictions.

Prof. Baig describes that the innovation is in employing multi-head attention layers in a neural basis expansion network. This architecture enables the AI model to concentrate on the most pertinent data points and ignore irrelevant noise, resulting in better performance. Additionally, the model accomplishes these outcomes without needing large datasets or high computing power, making it both accurate and efficient. The aim is to develop an accessible tool that can be easily incorporated into digital health solutions to enable enhanced diabetes management.

One of the biggest benefits of this new AI system is that it makes predictions individualized to align with a person’s own patterns of blood sugar. This more personalized approach helps make the model more accurate, which is critical to being able to make quick changes to insulin doses, meals, and exercise. Patients are able to use the model for a more individualized and knowledgeable way of dealing with their diabetes, enhancing overall health results.

The model is also made to function seamlessly on devices like smartphones and insulin pumps, and it is thus extremely accessible for daily use. It can be incorporated into mobile health apps for real-time tracking of glucose, providing patients and clinicians with real-time information on blood sugar levels. It can also be used in clinics to assist physicians in creating individualized treatment plans using accurate predictions.

As a part of their continuous endeavor, the researchers intend to test the technology developed through clinical trials at hospitals in Odisha. These will be carried out in collaboration with senior diabetologists to ensure that the technology is tested extensively before being put on the market. The initiative has been supported by the Department of Science and Technology (DST) and the Department of Biotechnology (DBT), which have supported it financially to make this innovative solution a reality.

In summary, the AI-based model of predicting blood sugar levels created by the team at NIT Rourkela is a great leap towards diabetes management. By enhancing the precision of glucose predictions and offering a more individualized method of diabetes management, this model can potentially assist millions of individuals with diabetes in making improved treatment choices and enhancing their quality of life. By integrating this technology with intelligent health devices, it can be easily implemented into daily diabetes care, rendering it a convenient tool for patients and medical professionals alike.