Advancing Diabetic Care Through Non-Invasive Glucose Monitoring Using Optical Sensors and IoT Technologies

A. Optical approaches for Non-invasive glucose level monitoring

The standard SMBG test has a limitation because the blood sample gives only a fragment of the real glucose level discontinuously, meaning that SMBG cannot indicate the ongoing glucose level changes during the day, even if this test is performed frequently. Changing symptomatic hypoglycemia or hyperglycemia for a patient could be uncomfortable and make it difficult to make a correct healthcare decision. For this reason, a recent study ensured that many glucometers for self-monitoring do not meet the level of accuracy; therefore, focusing on using Continuous Glucose Monitoring (CGM) is highly considered to help a patient decide on the medicine doses. CGM provides glucose concentration measurements based on a skin sensor that measures glucose concentration in the interstitial fluid and transmits the sensor values (usually at 5–15 min intervals) to a dedicated receiver, which can be seen in real time.¹⁶ A classification for glucose level monitoring is presented in reference¹⁷ based on different criteria, as shown in Figure 1. Non-invasive glucose level monitoring methods are based on measuring glucose concentration based on its chemical, thermal, electrical, or optical sensing properties.

Figure 1. Classification of glucose measurements according to different criteria.¹⁷

In 2021, a review of non-Invasive sensing systems for detecting glucose levels revealed a classification for the existing methods according to the sensor types. The classes include non-invasive glucose detection through the patient’s eye, finger, wrist, forearm, and abdomen.¹⁸ Light is focused on biological tissues and is affected by reflection, scattering, and transmission, according to the structure and chemical components of the tissue sample. This is the concept behind optical-based non-invasive glucose measurement methods. These properties are differentiated according to the analysis of the device used; thus, the interpretation of the glucose levels is based on the received spectrum. Many optical non-invasive methods have been analyzed to successfully test glucose levels and are classified in references.^17,19 Another classification based on optical approaches divides the classes into electromechanical, electrochemical, and electromagnetic (EM) methods. This study discusses the characteristics required for efficient sensors to predict a correct decision from the collected data. The authors clarified the optical approaches based on many features.²⁰

In 2021, a design was presented to test BG using optical non-invasive sensors based on finger tissue, including an infrared light-emitting diode (LED) with a wavelength of 950 nm, an ultrasound sender with 2 MHz, a sensitive IR detector, and two “Arduino” microcontrollers. The implementation of the system methodology depends on spreading the near-infrared light through the finger to obtain the glucose level. The authors did not provide the patient with the facility for auto use, as did other designs.²¹ Near-infrared NIR with LED is one of the technologies used with the associated sensor circuit to monitor the BG concentration. The experimental results of this approach and a mathematical model showed that the higher the glucose concentration, the lower the sensor output voltage. As the authors explained, this approach can be beneficial for individual patients by reducing the need for finger pricking and the pain it causes; however, further studies are needed to evaluate the accuracy and reliability of the designed system under different conditions.²² In 2023, a study considering IoT for non-invasive BG was published. The work idea is to build a system consisting of LED and NIR sensors to produce signals that are disseminated through the fingertip. A phototransistor is positioned next to the LED to detect any reflected signals and be stored in a “Thing-speak.” Mathematically analyzing the reflected signal intensity with an algorithm installed in the “Arduino” shows the relationship between glucose concentration and voltage. The experimental results of this study showed that the accuracy ranged from 1.13 to 16.41%. However, IoT is considered a promising tool owing to its continuous monitoring capabilities, data analysis, and remote access features.²³ Another study showed that LED light and biosensors provide good results for measuring individual glucose levels. More research is required to develop non-invasive BGL systems. There is still a need to improve the hardware of the devices because most IR technologies are software. Developing a design with low power and cost that is processed in real time would lead to a non-invasive system with high performance.¹⁹ A method of Fast IR spectroscopic glucose concentration measurement using a mid-infrared wavelength-swept pulsed quantum cascade laser (QCL) is adopted to measure the concentration of blood glucose using attenuated total reflection. The high power of the QCL with the proposed algorithm enables the BG measurements. The proposed method allows detection of the skin and enables the measurement of the tear film on the eyeball in the tangential direction. This has potential applications in the diagnosis and management of diabetes and minimizes the measurement time to 20 ms.²⁴

B. Commercial devices for non-invasive glucose level monitoring

“The standard ISO 15197 provides the quality guidelines, requirements, and specifications that glucose measuring devices should comply with to guarantee their suitability for human use”. The ISO guidelines evaluate each device and show whether it is suitable for commercialization.¹² Many commercial devices in the market are approved by the ISO standard, and almost all of them use spectroscopic techniques, especially NIR. A review of some of these devices, especially those using IoT technologies.

Figure 2. (a) The CoG device, (b) A cross sectional view of the CoG.²⁵

Figure 3. Invasive blood glucose meter GlucoTrack.²⁷

Figure 4. Eversion sensor insertion procedure.²⁸

Figure 5. Abbott lifestyle, and attached skin sensor.³¹

A. Materials

The Abbott Freestyle Library (CGM) was used to monitor the BG, where each sensor lifetime was approximately 14 days. This device can be used for patients aged ≥ 4 years because of its ease of use and can be interconnected with an available smartphone application.³⁵ The only restriction was the alarm sound every 15 min.

The suggested model is considered a new way to let the patient know his health situation only when it is critical and needs further attention by connecting the CGM device to a smart device to avoid unnecessary alarms. For more benefit from the stored BG results in the CGM device, a model is created to calculate the A1C for the patient to let him know if he is in the correct track of his health lifestyle. The personal smart device uses the cloud for extra storage space to collect BG data in a special database (DB) for GMI calculation using data from the past two weeks. The sketch in Figure 6 shows the devices and the connections between them, while the entire algorithm steps are explained in the solution approach subsection.

Figure 6. Abbott freestyle libre CGM and IoT technology.

B. Ambulatory Glucose Profile (AGP)

“AGP is considered as the standardized, practical one page report for graphically presenting a summary of glycemic control status in patients with diabetes who use continuous glucose monitoring (CGM) systems as part of their daily diabetes care.” The AGP report consists of three sections.

Figure 7. Glucose target ranges and categories.

Dx, diagnosis; DKA, diabetic ketoacidosis; Glu, glucose; Hypo, hypoglycemia; Maj, major; ER, emergency room; admit, admittance.³⁹

C. Methods

Initially, we tested the efficiency of the Abbott freestyle library device by comparing its results with the standard finger-prick BG test for all 10 patients for 24 h. The finger-prick test was performed three times a day (before breakfast, after breakfast, and at night before sleeping), while CGM tests the BG every 15 min. The averages of both tests based on Equation 1 for the finger pick test and Equation 2 for the CGM test are presented in Tables 2 and 3, respectively. The results in Table 3 show the average of the test results every 8 h.³⁴

where x_i is the test value every 15 min for a day. The Statistical Package for the Social Sciences software (version 24.0) was used to perform all statistical analyses, normality of data, and parametric statistical tests. Results are presented as (mean ± SD (standard deviation) SD). The independent samples t-test compares the Mean of two groups to assess the significance of differences between the groups P < 0.05. The findings of the study demonstrated that, significantly at P < 0.05, there were no differences between the means for three-time tests of patients in both methods based on the standard AGP report. Figure 8 shows the results of the comparison, where the data are almost similar for both tests, which means that the CGM device with the skin sensor that we used can measure blood glucose efficiently.

Figure 8. (a) Diagram (b) Scatter plot.

Both shows the mean and standard deviation for finger prick and CGM tests for 10 patients in 24 hours.³⁴

Usually, an A1C test is done every two or three months in the laboratory to check for improvement or worsening of BG; however, a significant change in BG can be known within two weeks. We used the stored data for 14 days to calculate A1C in our algorithm by using a standard formula to estimate A1C (eA1C) based on GMI, which gives an approximate value to the laboratory A1C.

The GMI is the average (mean) glucose value based on data collected by the CGM. The average glucose values from the CGM to obtain the GMI percentage can be calculated using Equation 3.

Both A1C (eA1C) and GMI provide information to maintain better control over blood sugar levels, and eA1C% can be obtained using Equation 4. The estimated A1C can then be calculated using Equation 5.⁴²

The term mean CGM data is the average glucose level over a specific period and can be found from Equation 6.

According to,^40,41 the GMI and A1C may not agree if the patient has acute hyperglycemia due to illness, steroid administration, or diabetic ketoacidosis; thus, GMI will be higher than the laboratory A1C measured at the same time and vice versa, where the GMI is lower than A1C if the patient suffers from hypoglycemia. This fact is used to alert the patient with a colored alarm (red, yellow, or green) to indicate hyperglycemia, hypoglycemia, or stability.

Another idea suggested to be added to this work is the use of a Markov chain to predict the GMI and A1C for the next 14 days. This can be done based on the stored data of past GMI results to allow the patient to find a healthy lifestyle. Because CGM provides glucose readings every few minutes, from this data, a calculation is made to estimate the GMI for each day using Equation 3.

The Markov chain concept was adopted from a previous study.⁴³ Their idea was to collect sequential blood sugar measurements and define discrete states: hypoglycemic, normal, and hyperglycemic. A transition probability matrix (TPM) was used to show the likelihood of moving from one state to another. The model is used to predict or detect anomalies in blood sugar behavior.

The idea of our suggestion is to build a three-state (S_i, where i = 0 to 2) model, where state S₀ indicates good control (stable in BG), while S₁ and S₂ indicate moderate and poor control, respectively. The intervals based on BG average are as: 7.3% < GMI < 7.8% for S₀, 7.8% < GMI < 8.0% for S₁, while GMI < 7.2% or GMI > 8.0% would be for S₂.

The state of transition is a change from one state to another based on the next period, where this transition state is a random process and is expressed in the form of probability. The probability of reaching state j from state i in one day is $P_{\mathit{ij}}$ . To calculate the state distribution after 14 days, we used Equation 7, where ${\pi }_0$ represents the initial state vector. This gives the probability of being in each state after 14 days.

In this way, an estimated result can be calculated to help the patient find if he is on the right track of his lifestyle or if the lifestyle needs some changes.

D. Solution approach

This study aims to benefit from IoT technologies by connecting a CGM device to a personal smart device to transfer data between them and to be stored in a cloud database. Another benefit of this connection is that it enhances the CGM device based on the three proposed algorithms.

Algorithm 1 calculates the patient’s BG level based on AGP standardization, so that a message appears every 8 h to the smart device instead of the 15 min alarm during a stable health situation; otherwise, an emergency message will appear.

Algorithm 2 uses the data stored to calculate A1C every 14 days, based on the standard in Table 4. Then, a comparison between GMI and A1C will start, and the result is sent to the patient as a colored alarm either red, yellow, or green to indicate hyperglycemia, hypoglycemia, or being stable, respectively.

Algorithm 3 calculates the GMI every two weeks depending on the stored data, and a calculation is also performed to estimate the result for the next 14 days to help the patient to control his lifestyle. A suggested idea may be added to the algorithm to estimate A1C every 3 months for more accurate results based on the stored outputs.

The solution is based on Equation (2) to calculate the mean glucose of one day and on Equation (6) to obtain the mean CGM data, while Equations (4) and (5) are used to calculate the predicted GMI for the next 14 days based on the Markov chain model using three states, as mentioned before.