Researchers from Qatar Foundation (QF) and Texas A&M University in the US have developed an AI-based model to make a non-invasive gadget to detect low blood sugar in people with Type1 diabetes (T1D), Gulf Times has learnt.
“The idea was to detect the frequency and magnitude at which signs and symptoms of hypoglycemia or low blood sugar occur in people with T1D. To do this, we developed an app that would help us collect relevant data via a smart watch,” said Dr Khalid Qaraqe, professor, Electrical and Computer Engineering Programme, Texas A&M University at Qatar, a QF partner and the lead of the programme.
Signs and symptoms of hypoglycemia can include sweating, feeling tired, dizziness and tremors. These physiological tremors were the focus of the project led by Dr Qaraqe and funded by QF’s Qatar National Research Fund.
Dr Qaraqe explained, “We recruited 77 participants with T1D in two cohorts: one consisting of 45 adults located in the US, and 32 children, from 10-17 years located in Qatar. All members of both cohorts were users of Apple watch and continuous glucose monitor (CGM). The built-in motion sensor in the Apple watch and blood sugar data collected from the CGM were crucial for the study.”
To collect data, the app was installed on each of the participants’ Apple watch. Through the watch’s built-in motion sensor and the app, the group was able to determine the frequency at which tremors were occurring in participants with T1D.
"Hypoglycemia happens when the blood sugar drops to a level that's too low to sustain normal functioning- below 70 milligrams per deciliter. It is common in people with T1D to experience at least one or two episodes of mild hypoglycemia a week," said, Prof Goran Petrovski, diabetes consultant at QF’s Sidra Medicine who served as a consultant on the project.
Existing solutions for glucose monitoring such as continuous glucose monitors can prevent these events but are very costly. The need to come up with a simple, wearable and inexpensive solution is what started this research.
After obtaining the relevant data, the team used AI to develop a machine learning (ML) algorithm to correlate tremor frequency and hypoglycemia.
Prof Qaraqe said, “When the algorithm detects a tremor within the frequency range that is indicative of hypoglycemia, it will send a message to the app user which, in the case of children, is parents to alert them of possible hypoglycemia.”
The team has gone a step forward and developed another ML algorithm for prediction of hypoglycemia which can alert the user of hypoglycemia in advance, allowing them to prevent it from occurring rather than having to manage it. The algorithm is also able to predict and estimate the sugar level based on the tremor data.
“What we want to do now is to create a wearable device, in the form of a bracelet or a ring which will use compact high-precision accelerometers to capture low frequency physiologic tremors and predict the blood sugar level,” Prof Qaraqe pointed out.
Dr Petrovski said that the device will be a low cost one to cater to every segment of the society. “The device we envision will be a very low-cost device, which doesn’t require a screen making it low-maintenance and won’t need to be charged daily,” he noted.
The developed technology has an accuracy rate of between 85-89% and “the accuracy range of our technology is actually similar to that of commercially available medical devices,” according to Prof Qaraqe.
“Our technology is cloud-based which means it is low latency – that means it processes data with minimal delay. As of now, the total time between the tremor occurring, sending the data to cloud-based AI, performing the calculation in the cloud and getting it back is between 60-90 seconds, which is good, but we are working to reduce this further to a matter of a few seconds and have it almost in real time,” added, Prof Qaraqe.
“The idea was to detect the frequency and magnitude at which signs and symptoms of hypoglycemia or low blood sugar occur in people with T1D. To do this, we developed an app that would help us collect relevant data via a smart watch,” said Dr Khalid Qaraqe, professor, Electrical and Computer Engineering Programme, Texas A&M University at Qatar, a QF partner and the lead of the programme.
Signs and symptoms of hypoglycemia can include sweating, feeling tired, dizziness and tremors. These physiological tremors were the focus of the project led by Dr Qaraqe and funded by QF’s Qatar National Research Fund.
Dr Qaraqe explained, “We recruited 77 participants with T1D in two cohorts: one consisting of 45 adults located in the US, and 32 children, from 10-17 years located in Qatar. All members of both cohorts were users of Apple watch and continuous glucose monitor (CGM). The built-in motion sensor in the Apple watch and blood sugar data collected from the CGM were crucial for the study.”
To collect data, the app was installed on each of the participants’ Apple watch. Through the watch’s built-in motion sensor and the app, the group was able to determine the frequency at which tremors were occurring in participants with T1D.
"Hypoglycemia happens when the blood sugar drops to a level that's too low to sustain normal functioning- below 70 milligrams per deciliter. It is common in people with T1D to experience at least one or two episodes of mild hypoglycemia a week," said, Prof Goran Petrovski, diabetes consultant at QF’s Sidra Medicine who served as a consultant on the project.
Existing solutions for glucose monitoring such as continuous glucose monitors can prevent these events but are very costly. The need to come up with a simple, wearable and inexpensive solution is what started this research.
After obtaining the relevant data, the team used AI to develop a machine learning (ML) algorithm to correlate tremor frequency and hypoglycemia.
Prof Qaraqe said, “When the algorithm detects a tremor within the frequency range that is indicative of hypoglycemia, it will send a message to the app user which, in the case of children, is parents to alert them of possible hypoglycemia.”
The team has gone a step forward and developed another ML algorithm for prediction of hypoglycemia which can alert the user of hypoglycemia in advance, allowing them to prevent it from occurring rather than having to manage it. The algorithm is also able to predict and estimate the sugar level based on the tremor data.
“What we want to do now is to create a wearable device, in the form of a bracelet or a ring which will use compact high-precision accelerometers to capture low frequency physiologic tremors and predict the blood sugar level,” Prof Qaraqe pointed out.
Dr Petrovski said that the device will be a low cost one to cater to every segment of the society. “The device we envision will be a very low-cost device, which doesn’t require a screen making it low-maintenance and won’t need to be charged daily,” he noted.
The developed technology has an accuracy rate of between 85-89% and “the accuracy range of our technology is actually similar to that of commercially available medical devices,” according to Prof Qaraqe.
“Our technology is cloud-based which means it is low latency – that means it processes data with minimal delay. As of now, the total time between the tremor occurring, sending the data to cloud-based AI, performing the calculation in the cloud and getting it back is between 60-90 seconds, which is good, but we are working to reduce this further to a matter of a few seconds and have it almost in real time,” added, Prof Qaraqe.