AI can detect low glucose levels through ECG without digital puncture test


Blood sugar monitoring is crucial for both healthy individuals and diabetic patients. Current methods for measuring glucose require needles and repeated punctures during the day. Fingertips can often be painful and deter patient compliance.

A new technique developed by researchers from the University of Warwick uses the latest findings of Artificial Intelligence to detect hypoglycemic events from unprocessed ECG signals, through portable sensors. The technology works with a 82% reliability and could replace the need for an invasive digital puncture test with a needle, which could be particularly useful for pediatric patients

Researchers at the University of Warwick have developed a new technology to detect low glucose levels through the ECG using a non-invasive portable sensor, which with the latest artificial intelligence can detect hypoglycemic events from raw ECG signals. Dr. Leandro Pecchia of the University of Warwick.

Currently, the NHS has continuous glucose monitors (MCG) for the detection of hypoglycemia (blood sugar levels or derma). They measure glucose in the interstitial fluid using an invasive sensor with a small needle, which sends alarms and data to a display device. In many cases, they require calibration twice a day with invasive blood glucose level tests with digital puncture.

However, Dr. Leandro Pecchia’s team at the University of Warwick published today, on January 13, 2020, the results in an article entitled ‘Precision medicine and artificial intelligence: a pilot study on deep learning for event detection ECG-based hypoglycemic in the journal Nature Springer Scientific reports proving that using the latest findings of Artificial Intelligence (i.e. deep learning), they can detect hypoglycemic events from unprocessed ECG signals acquired with non-invasive portable sensors.

Two pilot studies with healthy volunteers found that the average sensitivity and specificity of approximately 82% for the detection of hypoglycemia, which is comparable to the current CGM performance, although not invasive.

Dr. Leandro Pecchia, from the University of Warwick School of Engineering, comments: “Finger bites are never pleasant and in some circumstances they are particularly cumbersome. Taking a finger during the night is certainly unpleasant, especially for pediatric patients. Our innovation consisted in the use of artificial intelligence for the automatic detection of hypoglycemia through a few beats of ECG. This is relevant because the ECG can be detected in any circumstance, including sleep. “

The figure shows the output of the algorithms over time: the green line represents normal glucose levels, while the red line represents low glucose levels. The horizontal line represents the glucose value of 4 mmol / L, which is considered the significant threshold for hypoglycemic events. The gray area surrounding the solid line reflects the measurement error bar.

The Warwick model highlights how the ECG changes in each subject during a hypoglycemia event. The following figure is an example. The solid lines represent the average beats of two different subjects when the glucose level is normal (green line) or low (red line). The red and green shadows represent the standard deviation of the beats around the average.

A comparison highlights that these two subjects have different changes in the ECG waveform during hypo events. In particular, Subject 1 has a visibly longer QT interval during hiccups, while Subject 2 does not.

The vertical bars represent the relative importance of each ECG wave to determine if a beat is classified as hypo or normal.

From these bars, a trained clinician sees that for Subject 1, the displacement of the T wave influences the classification, reflecting that when the subject is in hiccups, the repolarization of the ventricles is slower.

In Subject 2, the most important components of the ECG are the P wave and the increase in the T wave, which suggests that when this subject is in hiccups, the depolarization of the atria and the threshold for ventricular activation are particularly affected . This could influence subsequent clinical interventions.

This result is possible because the Warwick AI model is trained with each subject’s own data. The intersubjective differences are so significant that training the system using cohort data would not give the same results. Similarly, personalized therapy based on our system could be more effective than current approaches.

Dr. Leandro Pecchia comments:

“The differences highlighted above could explain why previous studies that used ECG to detect hypoglycemic events failed. The performance of trained AI algorithms on the ECG data of the cohort would be hampered by these differences between subjects. “

“Our approach allows personalized tuning of detection algorithms and emphasizes how hypoglycemic events affect ECG in individuals. Based on this information, doctors can adapt the therapy to each individual. Clearly more clinical research is required to confirm these results in larger populations. That’s why we are looking for partners. “

If you have an interesting article / experience / case study to share, contact us at [email protected]


Please enter your comment!
Please enter your name here