Managing diabetes with artificial intelligence

Dr. Khalid Qaraqe, professor at Texas A&M University at Qatar, explains how a new Artificial Intelligence (AI) technology developed in Qatar will change the lives of people living with diabetes. 

Individuals living with diabetes are forced to contend with a common challenge: managing hypoglycaemia. It is a condition that occurs when glucose levels in the blood drop below the required level for people to sustainably function. In fact, individuals with Type 1 diabetes typically experience two episodes of mild hypoglycaemia a week. During these episodes, they experience symptoms such as sweating, tiredness, dizziness, and tremors. If they do not consume sugar immediately when this occurs, there is also a danger that their condition can worsen.

Managing hypoglycaemia has been difficult for patients to date. However, by using the latest AI technology we are finding new ways to lessen these episodes' impact on people's day-to-day lives.

Detecting hypoglycaemic events

I lead a team at Texas A&M University at Qatar that has been trying to tackle the issue of hypoglycaemia for diabetes patients. In collaboration with the Qatar National Research Fund, we have developed an AI-based model that detects hypoglycaemic states.

One of the main risks of hypoglycaemia is that people cannot realize what is happening to them. This happens particularly with children, who sense that something is wrong but are often unable to connect the symptoms they are experiencing with their diabetes, and to understand that these symptoms stem from low blood sugar. We therefore set out to build a machine learning structure that effectively identifies hypoglycaemic events.

Smart technology can be central to medical innovations like this. Based on the premise that physical tremors could be a sign of hypoglycaemia, our team collected data on tremors from patients wearing smartwatches and matched this against data collected from Continuous Glucose Monitoring Devices: small devices that many diabetes patients have implanted directly under their skin to continuously monitor glucose levels. We then built an AI model that correlated patterns between the frequency and magnitude of tremors, including whether each participant had low or high blood glucose levels.

The results were compelling: physical tremors are a clear sign of hypoglycaemia and by monitoring tremors more closely, we can help those living with diabetes inefficiently managing their condition.

Predicting episodes

While detecting hypoglycaemia is useful, the real potential of technological innovation is in improving our ability to predict when hypoglycaemic attacks will happen. If an episode is already occurring, we are already late. We can make the most significant difference to the lives of those with diabetes if we can enable them to anticipate when they are nearing an attack, giving them the opportunity to prevent it from occurring, rather than just managing it when it happens.

The next step of our research was therefore to develop the AI model from serving a primarily analytical function to a predictive one. By developing a tailored app that interprets the data gathered from each patient, we have been able to achieve this and can currently predict episodes happening between 30 seconds to one minute in advance. Our algorithms have proven to be 82% accurate in adults and 86% accurate in children.

We designed the app – named TREMOR – to be very user friendly, and we hope to see it available via app stores in the coming months, helping patients to take a more proactive approach towards efficiently managing their diabetes.

A low-cost wearable tech solution

To be effective, any technology that detects hypoglycaemic events must be able to always monitor the diabetic individual. Our data collection relied on smartwatches and, with the rapid rise of wearable technology in recent years, similar devices are a natural solution for patients to manage their diabetes on an ongoing basis. However, there are drawbacks as the cost of smart watches is a challenge.

To ensure innovations for managing diabetes are accessible to all, we need to find low-cost alternatives. Our researchers are currently progressing with developing a new wearable device for diabetes patients that will be much more affordable. The final output is likely to be either a bracelet or a ring that transmits tremor data to smartphones via Bluetooth.

Changing lives

The AI-based app system we have developed is the first of its kind, offering real-time hypoglycaemic monitoring and prevention. Designed to be non-invasive and with simple alerts that warn users of dangerous glucose levels, it is intended to make managing diabetes easier for anyone living with the condition. For parents especially, solutions like this offer a fast and effective way to monitor glucose levels in their child, which will undoubtedly give them peace of mind.

Significant progress has already been made in predicting hypoglycaemic episodes prior to this project. Using Continuous Glucose Monitoring Devices, episodes can currently be predicted with reasonable accuracy up to twenty minutes in advance. However, apps like TREMOR can enable people to self-monitor and manage their condition without the time-consuming and sometimes stressful experiences of processes like blood sample collection. We expect demand for this type of non-invasive technology to only increase in the coming years.

Improving the warning time for non-invasive technology will be essential however, which is the focus of our next phase of research. We are currently working towards developing our algorithms so that they can predict episodes up to 15 minutes prior to the event – approaching the time achieved by invasive options.

Focussing medical practitioners' efforts

If we can help diabetes patients to manage their condition more effectively, we can not only improve their quality of life but also relieve unnecessary pressures on healthcare systems. Mismanagement of hypoglycaemia can too easily lead to more serious symptoms, which then require practitioners' attention; but, with the right tools, most individuals could avoid this, freeing up clinicians to help patients with the most severe cases.

In the upcoming months and years, we will hopefully see similar AI-based wearable technology solutions targeting a range of health conditions. The difference this could make – to patient lifestyle, cost savings and capacity pressures – is monumental.