Wearable electronic gadgets could serve as valuable instruments for tracking blood glucose levels (BGLs) in people with diabetes. However, their compact size and limited power can result in significant measurement inaccuracies. In a recent study, researchers introduced a screening method that effectively filters out low-quality data during a preprocessing phase, improving the precision of BGL estimations. This advancement could lead to more convenient glucose monitoring solutions using everyday consumer devices, removing the necessity for finger pricks.
Diabetes has become a widespread health issue, impacting over 500 million adults globally. As there is currently no cure for type 1 or type 2 diabetes, managing blood glucose levels is crucial for patients. Traditionally, glucose monitoring through painful finger pricks has been the standard approach for years, but advancements in technology are gradually presenting better alternatives.
Numerous researchers have suggested noninvasive techniques to monitor BGLs with commonly available wearable technology, such as smartwatches. For instance, by placing the LEDs and photodetectors found in certain smartwatches against the skin, it is possible to measure pulse signals of oxyhemoglobin and hemoglobin to estimate a metabolic index from which BGLs can be derived. However, the compact size and limited battery life of smartwatches can lead to lower data quality in the captured signals. Additionally, as these devices are often worn on the limbs, daily movements can introduce inaccuracies. These challenges hamper the reliability and clinical usability of these wearables for diabetes care.
A research team from Hamamatsu Photonics K.K. in Japan has been exploring this issue in search of effective solutions. Led by Research and Development Engineer Tomoya Nakazawa, their recent study published in the Journal of Biomedical Optics (JBO) involved an extensive theoretical evaluation of error sources in the metabolic-index-based approach. Following their analysis, they developed a unique signal quality index to filter out poor-quality data before processing, thereby improving the precision of estimated BGLs.
“With smartwatches becoming increasingly popular among various age groups and the prevalence of diabetes on the rise globally, a technique to enhance signal quality that is easy to implement and adaptable to individuals’ differences is crucial for addressing the growing demand for noninvasive glucose monitoring solutions,” Nakazawa commented, highlighting the study’s purpose.
The researchers mathematically demonstrated how discrepancies between two types of phase delays in oxyhemoglobin and hemoglobin pulse signals, calculated by different methods, can effectively indicate noise impact. They identified two primary sources of phase error: background noise levels and estimation inaccuracies from discrete sampling intervals. After defining these error sources, they assessed their impact on the calculated metabolic index.
The proposed screening method involves setting thresholds for errors in phase estimation and metabolic index calculations. Any data that surpass these thresholds gets discarded, and missing values are estimated using alternative methods derived from remaining data.
To evaluate this approach, the researchers conducted an extensive experiment where the sensors of a commercial smartwatch were used to track the BGLs of a healthy participant during “oral challenges.” Over four months, the individual underwent 30 tests, fasting for two hours before consuming high-glucose foods. Their BGLs were measured using the smartwatch alongside a standard continuous glucose monitoring sensor, which provided reference measurements.
Significantly, applying the proposed screening method resulted in a notable improvement in accuracy. By utilizing the Parkes error grid technique to classify measurement inaccuracies, a considerably larger portion of the data points fell within Zone A when screening was utilized. This zone indicates clinically accurate values that would lead to appropriate treatment decisions. “The implementation of our screening process enhanced the accuracy of BGL estimations in our smartwatch prototype,” Nakazawa stated, emphasizing that “our technique could enable the integration of wearable and continuous BGL monitoring into devices like smartwatches and smart rings, which usually face limitations regarding size and signal quality.”
The research team also acknowledged some ongoing limitations of smartwatches compared to smartphone camera-based methods that affect their performance. Although the proposed technique could significantly boost smartwatches’ functionality, improvements in the photodetector and amplifier circuits could greatly enhance the appeal and clinical viability of wearable electronics for BGL monitoring.
Further investigations in this field promise to arm diabetes patients with powerful tools to manage their condition more effectively, ultimately improving their quality of life.
