In a new study, researchers developed a new AI tool to detect low glucose levels from ECG signals using non-invasive wearable sensors.
Tracking changes in blood sugar or glucose levels is vital for diabetic patients. Hyperglycemia (high blood glucose) results in long-term health complications that can damage the kidney, nerves, and blood vessels in the eye. Hypoglycemia (low blood glucose) leads to short-term problems with a person’s health such as confusion, irritability, and severe loss of attention that can be fatal under certain conditions as it affects the heart.
Current methods to measure blood glucose require needles and repeated finger pricks over the course of the day to draw blood needed for analysis. However, these methods are invasive and do not allow for continuous monitoring. An alternative to the finger prick test is Continuous Glucose Monitors (CGM). These devices measure the glucose levels in interstitial fluid, using a sensor with a small needle that sends back data to a display device. Most CGMs require finger pricking for calibration twice a day. The devices are expensive, which limits continuous monitoring of blood glucose levels and some studies suggest their reliability is limited in detecting hypoglycemia.
A recent study, published in Scientific Reports, aimed to detect low blood glucose levels in healthy individuals from raw ECG signals that were acquired using non-invasive wearable sensors. With the help of the latest artificial intelligence techniques such as deep learning, this technology can automatically detect hypoglycemia from a few heartbeats of raw ECG signals.
In the study, four healthy individuals were monitored for 24 hours over 14 consecutive days. Using wearable sensors, ECG, human rest/activity cycles, and continuous glucose levels were recorded for each participant. The researchers limited their analysis to data recorded during the night to minimize the influence of ECG circadian changes.
The study reported that the technology worked with an 82 percent reliability in detecting low glucose levels and is comparable to the performance of CGM devices, suggesting the new tool could eventually replace the finger pricking method. The researchers highlight how the ECG waveform changes in each participant when the blood glucose drops below the normal level and that the changes are different between the participants. Based on the results, the proposed AI model will help clinicians to visualize which part of the ECG signal is associated with low glucose levels in their patients.
The differences in results between the participants were significant, highlighting why previous studies using a cohort of individuals to detect low glucose levels from ECG signals were not successful in capturing the variability between subjects. The researchers used a personalized medicine approach to develop their technology in which the model was trained with each participant’s own data. According to the study author, Dr. Pecchia, this approach enabled the researchers to personalize tuning of the detection algorithms and emphasize the effect of low glucose levels on ECG in individuals, suggesting the proposed AI technology will help clinicians to adapt therapies for managing glucose levels to each individual patient. However, more clinical research is needed to confirm the results in wider populations.
Written by Ranjani Sabarinathan, MSc
Porumb M, Stranges S, Perscapè A, et al. (2020). Precision Medicine and Artificial Intelligence: A Pilot Study on Deep Learning for Hypoglycemic Events Detection based on ECG. Scientific Reports. doi: 10.1038/s41598-019-56927-5
AI can detect low-glucose levels via ECG without fingerprick test. (2020, January 13). Retrieved from https://www.eurekalert.org/pub_releases/2020-01/uow-acd011320.php
Image by Steve Buissinne from Pixabay