The constructed system is presented in
, where one may see a schematic sketch of the system () as well as an image of the camera with its optics and a laser illuminating a hand of a subject being fixed by gypsum to allow more accurate measurement (). The setup is very simple and includes only a green laser (at 532nm) to illuminate the inspected object and a camera (having slightly defocused optics) that is connected to a computer. The camera captures images of the secondary speckle pattern being reflected from the hand of the subject at rate of 350 frames per second (fps). The focal length of the optics that had been used in our experiments was 50mm and the distance from the laser to the subject’s hand was about 50cm. The laser output power was about 10mW.
The implemented optical configuration for remote measuring of glucose level in blood from subject’s hand: (a). Sketch of the optical system (b). Subject’s hand under laser illumination as viewed by the camera.
After extracting the speckle pattern in each frame we perform correlation and obtain the change in the 2-D position of the peak versus time. In
we show a typical system output with high signal to noise ratio. It includes only several pulses related to heart beating, while in the experiment we take into account the average of six of them. Every pulse is shaped similarly to ECG PQRST [29
]. It contains a P pulse, QRS complex, and a T pulse. However, this is a mechanical vibration profile, rather than an electrical signal (as ECG) and therefore it provides us with additional temporal information about vessel’s vibration due to blood flux pulsation.
Temporal plot of the outcome from the system used in the clinical tests with the graphical description of the observed parameters.
In this research, we have chosen to monitor two types of parameters: the first is proportional to the amplitude change and the second to the temporal change of the relative speckle shift. The parameters of the extracted pulse profile that are proportional to amplitude change are: main peak amplitude (positive and negative), pulse profile energy (positive and negative separately), positive to negative pulse peak ratio, secondary peak amplitude and main to secondary peak amplitude ratio. The parameters of the extracted pulse profile that are proportional to the temporal change are: pulse width (positive and negative), mean distance between peaks (gap or pulse rate), distance from positive to negative peak. These parameters are listed in
, while the graphical meaning of them is presented in .Amplitude ratio based parameters (parameters #9 and #12) were chosen in order to perform the experiment with reduced sensitivity to vibrations of the optical setup itself. Those parameters are dimensionless quantities and therefore they are not sensitive to a change in the laser’s projection angle.
Summary of the Observed Parameters
In order to make the measurements reliable, we needed to be sure that the same spot on the hand is illuminated and examined at every moment. Therefore, an individual hand template was constructed using gypsum, while a hole was drilled into it for each one of the subjects to allow the illumination of the wrist (). The diameter of the hole is slightly larger than the laser beam's diameter (approximately 1cm). We have tested four healthy subjects from 22 to 35 years old with different gender and weight. The summary of the subjects’ personal information is listed in
. All measurements were repeated several times to assure repeatability and correctness.
Summary of the Subjects’ Personal Information
3.1 Calibration tests
In order to authenticate the required accuracy of 10-15% variation (as per standard glucometer) for reliable experiment results, we needed to be able to illuminate the same spot on the wrist over time. We built individual fixation devices for subject’s hand using gypsum and executed several check tests. We inserted the hand of the subject, then marked the pulse's spot on the hand, drilled a hole through the gypsum in the position of the chosen spot, pulled the hand out and re-inserted it. The marked spot was at the exact position.
Another test was aimed to check the stability of the gypsum fixation over time. Each subject inserted his hand into the device and stayed fixed for approximately 30 minutes, while he was monitored by the system. In
one can see the stability of the system, while the results do not vary more than 15%. Glucose concentration is shown in units of [ml/dl] divided by 10 (representing a constant level of 100 [ml/dl]), while the units of parameter #6 (please refer to ) are counted in samples.
Fig. 3 Stability of the system: constant glucose level in blood (denoted by blue line with triangles) and the estimated parameter 6 (denoted by magenta line with rectangles). Glucose level is given in units of 0.1[ml/dl] (representing a constant level of 100 (more ...)
3.2 Main test
To ensure that the glucose blood level would rise only as consequence of drinking of a sweetened beverage during the experiment, each examined subject preserved a fast for about 12 hours before the measurement took place. The expected values of blood glucose level for non-diabetic person [1
] after fasting falls to values range between 90 to 110 [mg/dl]. At the beginning of every experiment we indeed checked that the subject's blood glucose level was at this range, while later the subject received a sweetened drink and the level was changed.
The rate at which the concentration of glucose increases is different for each individual and depends on many personal parameters, like: body weight, metabolic rate, level of insulin in blood etc. The blood glucose level as we obtained after drinking of about 400ml of sweetened beverage (40K Cal) by the subjects was from 150 to190 [mg/dL]. Each experiment lasted for 50-80 minutes, during it the measurements were carried out repeatedly every 5 minutes. Each 5 minutes sampling included capturing six subsequent video files of the illuminated spot and taking an accurate blood sample with a glucometer (“Accu-check”) and manual blood pressure measurement using standard sphygmomanometer. All experiments showed that blood pressure have not been changed over the time of the experiment. It is important to check this point in order to ensure that the expected change in the pulse profile is indeed caused by glucose intake, rather than by blood pressure change as was shown previously in Ref. [26
We wrote a MATLAB program that analyzed the videos and extract the observed parameters from the files. Each file contained about 5 seconds of video samples at rate of 350 fps (frames per second), usually containing 6 pulse peaks. Each peak is processed separately and the chosen parameters are extracted and averaged, therefore representing the average of approximately 30 peaks of pulse profile per each 5 minutes. For each parameter the final graph of the estimated glucose level was produced. Joint graphs of the estimated and the reference glucose level for each one of the parameters and for each one of the subjects were created.
Usually we took into an account only the first samples of the estimated values; the ones when the glucose level is still rising (the positive slope of the graph). Those samples were more reliable due to two main reasons: First is that glucose metabolism mediates changes in biochemistry levels: insulinotropic second messengers, including cyclic nucleotides, inositol phosphates, diacylglycerol and Ca2+
]. Moreover, biochemistry metabolism in limb is not linear [3
], therefore the change in blood fluid viscosity due to biochemistry metabolism is not linear as well. We believe that those metabolic processes initiated by glucose intake in the body cause accumulative change in blood viscosity. With some delay after the initial glucose intake, the additional metabolic processes affect the change in pulse profile which is not directly connected to the actual glucose level change. The second reason is an “exhaustion” of the subject. Although gypsum makes reliable fixation, it is not attached “strongly” enough to the hand and after approximately half an hour of the experiment the subject can produce spontaneous movement which may cause a change in the vibration profile which is not connected to the actual glucose change.
In addition to correlation coefficient, we used root mean square error (RMSE) estimation to quantify the relation between the reference and the estimated data:
is an i-th sample of the estimated values, ri
is an i-th sample of the reference values and N is the number of samples. The calculated samples were normalized before applying RMSE estimator in order to obtain the common estimation scale for all parameters.
Dozens of experiments were executed with four subjects in order to present a proof of principle validation. Initial results show a good correspondence of the estimated parameters with the positive slop of glucose level change in blood. However the extensive characterization of the parameters is in progress. There was not found any good correlation between the parameters related to temporal change of the pulse profile and the actual change in the glucose level. However, some of the amplitude related parameters showed a quite good matching with the actual glucose change. Some of the obtained results are presented in the following figures.
we present the temporal evolution of the chosen parameters versus the reference measurement of glucose level taken by glucometer while glucose concentration in blood is denoted by blue line with triangles and the optically measured parameters from the pulse profile is denoted by magenta line with squares. The graph of the reference (glucose level) was obtained by using a conventional glucose meter device (“Accu-check”). Error bars refer to standard deviation of positive and negative deviations separately, calculated over each 30 peak samples (per each point on the graph). Four different graphs on each figure refer to four different experiments taken with relevant subject on different days, during the morning hours while each subject preserved a fast of 12 hours. Estimated values were linearly transformed to glucose level units according to the calibration done per each subject at the first measurement (time 0). Correlation and RMSE coefficients are shown below each graph.
Data of subject #1: Glucose level in blood and amplitude of positive peak (parameter #1). Glucose level is denoted by blue line with triangles and the optically measured parameter is denoted by magenta line with rectangles.
Data of subject #4: Glucose level in blood and amplitude of positive peak (parameter #1). Glucose level is denoted by blue line with triangles and the optically measured parameter is denoted by magenta line with rectangles.
refers to subject #1, while the best correlative parameters for this subject were parameter #1 (please refer to ).
one can see graphs taken with subject #1 which show glucose concentration in blood (denoted by blue line with triangles) and the optically measured parameter of the ratio between positive and negative peak (parameter #9) of pulse profile (denoted by magenta line with squares). shows an exact inverse ratio between the estimated and the reference glucose level. Note that parameter #9 is actually a ratio between parameter #1 and #5, therefore it is not sensitive to the vibrations of the optical system. Some of the results showed very high correlation with the reference measurement for the full cycle of glucose changes in blood. In one can see that the estimated parameter is tracking after the reference glucose level (in opposite direction). It comprises the positive and the negative slopes, therefore presenting a full cycle of increase and decrease of glucose level in the blood.
Data of subject #1: Glucose level in blood and the ratio between positive and negative peak (parameter #9). Glucose level is denoted by blue line with triangles and the optically measured parameter is denoted by magenta line with rectangles.
The correlation coefficient of −0.916 was obtained between the two curves. RMSE estimator for this parameter was calculated between the inverse function of the normalized estimated parameter (one minus the normalized values) and the reference. RMSE estimator is equal to 0.17 in this case. However, this estimator was working well only for one subject out of four.
refers to subject #2, while the best correlative parameter for this subject was found to be positive pulse amplitude (parameter #1).
refers to subject #3, while the best correlative parameter for this subject was found to be parameter #1 as well. refers to subject #4, with the best correlative parameter #1.
Data of subject #2: Glucose level in blood and amplitude of positive peak (parameter #1). Glucose level is denoted by blue line with triangles and the optically measured parameter is denoted by magenta line with rectangles.
Data of subject #3: Glucose level in blood and amplitude of positive peak. Glucose level is denoted by blue line with triangles and the optically measured parameter is denoted by magenta line with rectangles.
summaries all correlation coefficients, while
summaries all RMSE estimator coefficients from the graphs presented in –.
Summary of Correlation Coefficients from All the Tests Taken with the Four Subjects
Summary of RMSE Estimator Coefficients from All the Tests Taken with the Four Subjects