AI Can Put an End to Finger Prick Blood Glucose Testing

2020-01-19

Artificial intelligence is being used to develop new technologies that can detect low blood sugar levels using non-invasive sensors through electrocardiogram (ECG) testing. This technology, developed by Dr. Leandro Pecchia and his research team at the University of Warwick, can detect hypoglycemic events from raw ECG signals. In terms of blood sugar testing, the commonly used method is continuous glucose monitoring (CGM), which is invasive and requires calibration through finger pricks. However, Dr. Pecchia's team has published a pilot study titled "Precision Medicine and Artificial Intelligence: Deep Learning for Hypoglycemic Event Detection from ECG" that demonstrates the latest development in artificial intelligence can be used to detect hypoglycemic events from raw ECG signals obtained through non-invasive wearable sensors. Two pilot studies conducted on healthy volunteers found that the average sensitivity and specificity of the hypoglycemic detection using this method is comparable to CGM. However, the non-invasive nature of this method is a significant breakthrough, as finger pricks can be painful, uncomfortable, and inconvenient, especially for nocturnal and pediatric hypoglycemic patients. Dr. Pecchia stated, "Our innovation lies in using artificial intelligence to automatically detect hypoglycemia through a few ECG beats. This is relevant because ECG can be detected in any situation, including during sleep." The model used by the researchers is called the Warwick model, which emphasizes the changes in ECG during hypoglycemic periods for each participant. The artificial intelligence model is trained by the researchers using each participant's own data. Due to the significant inter-individual differences, training the system using pooled data would not yield the same results. A more effective approach is personalized treatment based on the new system. The success of the University of Warwick scientists' approach may be attributed to the fact that the artificial intelligence algorithm is trained using each participant's own data. Pecchia said, "Training the artificial intelligence algorithm on pooled ECG data would be hindered by these inter-individual differences. Our approach allows for personalized adjustments to the detection algorithm and highlights the impact of hypoglycemic events on individual ECGs. Based on this information, clinicians can tailor treatment to each individual. Clearly, more clinical research is needed to validate these results for widespread adoption. That is why we are seeking partners."

RECENT AI NEWS

RECENT AI TOOLS