Posted on Diabetes

Enhancing the accuracy of wearables that measure blood glucose levels

Diabetes is an increasingly pervasive disease, currently affecting over 500 million adults worldwide. Since there is as yet no cure for type 1 or type 2 diabetes, patients must regularly monitor their BGLs to keep them in check. Though BGL-measuring devices relying on painful finger pricks have been the gold standard for decades, modern technology is slowly opening doors to better alternatives. Many researchers have proposed noninvasive methods to monitor BGLs using widely available wearable devices, such as smartwatches. For example, by placing the LEDs and photodetectors present in certain smartwatches against the skin, oxyhemoglobin and hemoglobin pulse signals can be measured to calculate a metabolic index, from which BGLs can be estimated. However, given the small size and limited power of smartwatches and similar wearables, the data quality of the measured signals tends to be quite low. Moreover, as these devices are worn on extremities, daily movements introduce measurement errors. These problems limit the accuracy and clinical applicability of such wearables for diabetes management. A team from Hamamatsu Photonics K.K., Japan, has been actively researching this issue, looking for effective solutions. In a recent study led by Research and Development Engineer Tomoya Nakazawa, published in the Journal of Biomedical Optics (JBO), they conducted an in-depth theoretical analysis of the sources of errors in the metabolic-index-based method. Based on this analysis, they implemented a novel signal quality index to filter out low-quality data as a preprocessing step and thereby enhance the accuracy of estimated BGLs. “As smartwatches are widely adopted across different regions and age groups, and with the global rise in diabetes cases, a signal quality enhancing method that is easy to implement and apply regardless of personal and individual differences is absolutely essential for meeting the increasing worldwide demand for noninvasive glucose monitoring devices,” remarks Nakazawa, explaining the motivation behind the study. First, the researchers mathematically showed that discrepancy between the two types of phase delays in the oxyhemoglobin and hemoglobin pulse signal calculated by different methods provides a good measure of the influence of noise. They then considered two main sources of phase error, namely, a background noise level and the estimation errors introduced via sampling at discrete intervals. After formalizing these sources of errors, they calculated the effect on the estimated metabolic index. The proposed screening approach involves implementing thresholds for the phase estimation and metabolic index errors. Data chunks that exceed the set thresholds are discarded, and the missing values are approximated using other means based on the rest of the data. To test this strategy, the researchers conducted a long-term experiment in which the sensors in a commercial smartwatch were used to monitor the BGLs of a healthy individual during “oral challenges.” In each of the 30 tests conducted over four months, the subject would fast for two hours before consuming high-glucose foods. Their BGLs were measured using the smartwatch and a commercial continuous glucose monitoring sensor, the latter of which was used to capture the reference values. Notably, preprocessing the data with the proposed screening method led to a notable increase in accuracy. Using the Parkes error grid technique to categorize measurement errors, a substantially higher percentage of data points ended up in Zone A when screening was applied. This refers to clinically accurate values that would lead to correct treatment decisions. “Adopting the screening process improved BGL estimation accuracy in our smartwatch-based prototype,” remarks Nakazawa, “Our technique could facilitate the integration of wearable and continuous BGL monitoring into devices such as smartwatches and smart rings, which are typically constrained in terms of size and signal quality,” he adds, highlighting the impact of the research work. The research team also noted some of the current limitations of smartwatches that lead to inferior performance compared to smartphone camera-based techniques. Though the proposed method could certainly help enhance the performance of the former, hardware improvements in the photodetector and amplifier circuits could go a long way to make wearable electronics a more attractive and clinically acceptable option to monitor BGLs.

Diabetes is a growing global issue, currently affecting over 500 million adults. As there is still no cure for either type 1 or type 2 diabetes, patients need to regularly monitor their blood glucose levels (BGLs) to manage their condition. While traditional BGL-measuring devices that require painful finger pricks have been the standard for many years, modern technology is beginning to offer better alternatives.

Many researchers have proposed noninvasive methods to monitor blood glucose levels (BGLs) using commonly available wearable devices, such as smartwatches. For instance, by positioning the LEDs and photodetectors found in certain smartwatches against the skin, it is possible to measure the pulse signals of oxyhemoglobin and haemoglobin. This data can then be used to calculate a metabolic index, which can help estimate BGLs. However, due to the small size and limited power of these smartwatches and similar wearables, the quality of the measured signals is often low. Additionally, daily movements can introduce measurement errors because these devices are typically worn on the extremities. These issues hinder the accuracy and clinical applicability of wearables for managing diabetes.

A team from Hamamatsu Photonics K.K. in Japan has been actively researching solutions to a pressing issue. In a recent study led by Research and Development Engineer Tomoya Nakazawa and published in the Journal of Biomedical Optics (JBO), they conducted a thorough theoretical analysis of the errors associated with the metabolic index-based method. Based on their findings, they developed a novel signal quality index to filter out low-quality data as a preprocessing step, which enhances the accuracy of estimated blood glucose levels (BGLs).

“As smartwatches are widely adopted across different regions and age groups, and with the global rise in diabetes cases, a signal quality enhancing method that is easy to implement and apply regardless of personal and individual differences is essential for meeting the increasing worldwide demand for noninvasive glucose monitoring devices,” remarks Nakazawa, explaining the motivation behind the study.

First, the researchers mathematically showed that the discrepancy between the two types of phase delays in the oxyhemoglobin and haemoglobin pulse signal calculated by different methods provides a good measure of the influence of noise. They then considered two primary sources of phase error: a background noise level and the estimation errors introduced via sampling at discrete intervals. After formalizing these sources of errors, they calculated the effect on the estimated metabolic index.

The proposed screening approach involves implementing thresholds for the phase estimation and metabolic index errors. Data chunks that exceed the set thresholds are discarded, and the missing values are approximated using other means based on the rest of the data.

To test this strategy, the researchers conducted a long-term experiment in which the sensors in a commercial smartwatch were used to monitor the BGLs of a healthy individual during “oral challenges.” In each of the 30 tests conducted over four months, the subject would fast for two hours before consuming high-glucose foods. Their BGLs were measured using the smartwatch and a commercial continuous glucose monitoring sensor, which was used to capture the reference values.

Notably, preprocessing the data with the proposed screening method led to a notable increase in accuracy. Using the Parkes error grid technique to categorize measurement errors, a substantially higher percentage of data points ended up in Zone A when screening was applied. This refers to clinically accurate values that would lead to correct treatment decisions. “Adopting the screening process improved BGL estimation accuracy in our smartwatch-based prototype,” remarks Nakazawa, “Our technique could facilitate the integration of wearable and continuous BGL monitoring into devices such as smartwatches and smart rings, which are typically constrained in terms of size and signal quality,” he adds, highlighting the impact of the research work.

The research team also noted some of the current limitations of smartwatches that lead to inferior performance compared to smartphone camera-based techniques. Though the proposed method could certainly help enhance the former’s performance, hardware improvements in the photodetector and amplifier circuits could go a long way toward making wearable electronics a more attractive and clinically acceptable option for monitoring BGLs.