Classifications

• • A—HUMAN NECESSITIES
  • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
  • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
  • A61B5/00—Measuring for diagnostic purposes; Identification of persons
  • A61B5/145—Measuring characteristics of blood in vivo, e.g. gas concentration or pH-value ; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid or cerebral tissue
  • A61B5/1455—Measuring characteristics of blood in vivo, e.g. gas concentration or pH-value ; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid or cerebral tissue using optical sensors, e.g. spectral photometrical oximeters
• • A—HUMAN NECESSITIES
  • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
  • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
  • A61B5/00—Measuring for diagnostic purposes; Identification of persons
  • A61B5/145—Measuring characteristics of blood in vivo, e.g. gas concentration or pH-value ; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid or cerebral tissue
  • A61B5/14532—Measuring characteristics of blood in vivo, e.g. gas concentration or pH-value ; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid or cerebral tissue for measuring glucose, e.g. by tissue impedance measurement

Definitions

• This invention relates to glucose monitoring, and more particularly, to glucose level monitoring using laser-induced emission spectroscopy.
• infrared absorption techniques are susceptible to accuracy problems, in part because glucose has more than 20 infrared absorption peaks, many of which overlap with the absorption peaks of other constituents in the body.
• Fluorescence spectroscopy using ultraviolet (UV) excitation light has been introduced for monitoring glucose levels. This technique requires, among other things, the monitoring of a spectral peak within the induced fluorescence spectrum. Accurately locating the peak may be difficult for a low-level fluorescence signal in the presence of noise. Increasing the intensity of the excitation light may not be a desirable option because of concerns of UV exposure to the body. Also, known fluorescence spectroscopic techniques generally fail to take full advantage of information contained in the fluorescence spectrum at wavelengths other than the peak wavelength and fail to account for certain nonlinear relationships between the glucose level and the resulting emission spectra.
• UV ultraviolet
• the present invention is embodied in an apparatus, and related method, that determines the concentration of glucose in a sample that includes water, by directly monitoring induced glucose ultraviolet and visible (UV-visible) emission light from the sample.
• the apparatus compensates for nonlinearities between these parameters and the glucose.
• the apparatus includes a light source, a sensor, and a processor.
• the light source emits ultraviolet excitation light of at least one predetermined energy level.
• the excitation light is directed at a sample to produce return light from the sample.
• the return light includes induced emissions of light produced as a result of interactions between the excitation light and any glucose with water present in the sample.
• the sensor monitors the return light and generates at least three electrical signals indicative of the intensity of return light associated with glucose concentration distinguishing characteristics of the emission light.
• the processor processes the electrical signals, using a predictive model, to determine the concentration of glucose in the sample.
• the predictive model is defined using six latent variables. The latent variables are used to derive prediction coefficients that are associated with the glucose concentration distinguishing characteristics.
• the intensity of the excitation light remains relatively constant while the sensor generates the electrical signals.
• the at least three electrical signals indicate the intensity of return light within a respective number of predetermined wavelength bands within the wavelength band of the emission light.
• the sensor may includes a first detector adapted to detect the return light within a first wavelength band and generate a first electrical signal, a second detector adapted to detect the return light within a second wavelength band and generate a second electrical signal, and a third detector adapted to detect the return light within a third wavelength band and generate a third electrical signal.
• the senor monitors the intensity of return light within eight different wavelength bands and generates eight electrical signals, each indicative of the intensity of return light within a respective wavelength band. More particularly, using an excitation light having a wavelength of about 308 nanometers, the eight wavelength bands may be centered at about 342, 344, 347, 352, 360, 370, 385 and 400 nanometers, respectively. Alternatively, the sensor may generate a plurality of electrical signals that indicate the intensity of return light substantially continuously across an extended wavelength spectrum associated with the emission light.
• the energy of the excitation light is varied over several predetermined energy levels
• the sensor generates, at each intensity level, a first electrical signal based on the intensity of return light within a wavelength of the emission light associated with ra an scattering, and a second electrical signal based on the intensity of return light within a wavelength band of the emission light associated with a peak of a broad glucose emission band.
• the apparatus may include one or more waveguides for transmitting the excitation light from the light source to the sample and for transmitting the return light from the sample to the sensor.
• a regression model is provided that accounts for a nonlinear relationship between the concentration of glucose in a sample and an electrical signal based on certain glucose concentration distinguishing characteristics of a light emission spectrum that includes UV-visible emission light produced by glucose related interactions with the excitation light. Further, a sample is caused to produce a light emission spectrum that includes emission light produced by any glucose related interaction or direct fluorescence, and a plurality of electrical signals are generated that represent the glucose concentration distinguishing characteristics. Finally, the plurality of electrical signals are processed, using the regression model, to determine the glucose concentration and an electrical signal generated based on the glucose concentration determined using the regression model.
• FIG. 1 is a block diagram of a glucose monitoring system embodying the invention.
• FIG. 2 is a graph of the intensity of glucose emission versus wavelength for different concentrations of glucose in water illuminated with laser excitation light having a wavelength of 308 nanometers.
• FIG. 3 is a graph of the intensity of glucose emission at two wavelengths verses glucose concentration in water, illuminated with laser excitation light having a wavelength of 308 nanometers and an excitation energy of 1 millijoule per pulse.
• FIG. 4 is a graph of the regression coefficient verses the latent variable number, derived from a partial least square (PLS) analysis using the intensities at eight wavelength indicated in the graph of FIG.2.
• PLS partial least square
• FIG. 5 is a graph of the prediction residual sum of squares (PRESS) versus number of latent variables, using one spectra at a time to test the PLS model derived from intensities at the eight wavelengths indicated in the graph of FIG.2.
• FIG. 6 is a graph of the PRESS versus number of latent variables using two spectra at a time to test the PLS model derived from intensities at the eight wavelengths indicated in the graph of FIG.2.
• FIG. 7 is a graph of the predicted concentration verses the actual concentration of glucose for the PLS model using six latent variables and for a multiple linear regression (MLR) model derived from the graph of FIG. 2.
• MLR multiple linear regression
• FIG. 8 is a graph of the predicted concentration verses the actual concentration of glucose for the PLS model using seven latent variables and for a multiple linear regression (MLR) model derived from the graph of
• FIG. 9 is a graph of the PRESS versus number of latent variables using one spectra at a time to test a PLS model derived from the whole spectra of the graph of FIG.2.
• FIG. 10 is a graph of the actual concentration verses the predicted concentration for the PLS model using six latent variables derived from the whole spectrum of the graph of FIG. 2.
• FIG. 11 is a graph of the intensity of glucose emission verses wavelength, at different excitation energy levels, for glucose in water at a concentration of 500 milligrams per deciliter.
• FIG. 12 is a graph of emission intensity versus wavelength for distilled water excited at an excitation energy of 5 millijoules per pulse.
• FIG. 13 is a graph of the emission intensity verses wavelength for ultra-anhydrous glucose excited at an excitation energy of 5 millijoules per pulse.
• FIG. 14 is a graph of the emission intensity verses wavelength for anhydrous glucose excited at 5 millijoules per pulse.
• FIG. 15 is a graph of the emission intensity versus wavelength for anhydrous glucose excited with excitation light having an energy at different levels between 0.25 and 10 millijoules per pulse.
• FIG. 16 is a graph of the intensity of glucose emission verses wavelength for different concentrations of glucose in water, illuminated with laser excitation light having a wavelength of 308 nanometers and an excitation energy of 7 millijoules per pulse.
• the present invention is embodied in a glucose monitoring system 10, and related method, for determining the concentration of glucose in a sample 12 by monitoring the glucose ultraviolet and visible (UV-visible) light emission spectra at several wavelengths or excitation intensities while compensating for the nonlinear relationship between the glucose concentration of these parameters.
• the system provides good accuracy in spite of the nonlinearities or low signal-to-noise levels.
• an excitation source 14 directs ultraviolet excitation light to the sample 12 through an optical fiber 16, to induce any glucose within the sample to absorb and reemit or to scatter the excitation light.
• An optical fiber or fiber bundle 18 collects return light emitted by the sample. The return light includes any glucose emissions induced by the excitation light.
• a sensor 20 monitors the return light within different wavelength bands of interest and generates a series of electrical signals based on the intensity of return light received in the wavelength band or bands of interest.
• the sensor includes a spectrograph 22 which resolves the return light by separating the return light by wavelength.
• An analyzer 24 or processor having a prediction model that associates the intensity of return light of interest with the concentration of glucose in the sample, processes the electrical signals generated by the sensor, compares the results with the model, and generates an electrical signal representing the concentration of glucose in the sample.
• a useful excitation source 14 is an excimer laser producing light having a wavelength of about 308 nanometers, a full width half maximum (FWHM) pulse width of about 120 nanometers, and a repetition rate of about 5 hertz. It is believed that glucose more efficiently absorbs excitation light having a wavelength between 260 to 280 nanometers and it would be desirable to have an excitation wavelength in that range. However, as presently understood, excitation sources operating at these wavelengths generally are of limited availability.
• the excitation light can be provided by any type of generally narrow-band ultraviolet light source and generally can have a wavelength from about 250 to 350 nanometers.
• the excitation light is guided to the sample 12 through a fused silica fiber 16 having a 600 micron core diameter.
• the excitation light's energy, emitted from the fiber is set to predetermine levels from about 0.5 to 10 millijoules per pulse (0.54 to 36 millijoules per square millimeter per pulse) .
• the induced emissions from the sample, or return light preferably is collected using a bundle of six fused silica fibers 18, each fiber having a 300 micron core.
• the collection fibers guide the return light to the sensor 20.
• the sensor 20 may include individual light-sensitive diodes, with appropriate bandpass filters, or as discussed above, may include a spectrograph 22 that resolves a broad spectrum of the return light.
• a spectrograph was used to collect the data discussed below.
• a long pass filter 26 (Schott WG335) is placed within the spectrograph to filter from the return light, any excitation light that may have been collected by the collection fibers 18.
• An ultraviolet enhanced grating 150 grooves per millimeter
• a silicon diode array detector 28 having 1024 elements collects the dispersed return light and generates an electrical signal that serially represents the intensity of return light collected in each element.
• An EG&G optical multichannel analyzer (OMA III) receiving the electrical signal can display a graph representing the intensity of return light within the desired wavelength band or bands of interest.
• the spectrum shown in FIG. 2 is the emission spectra of different glucose concentrations after excitation with an ultraviolet excimer laser light having a wavelength of 308 nanometers. Each spectrum is shown to have a double peak shape. One spectral peak is associated with a narrow wavelength band centered at about 346 nanometers, apparently produced as a result of vibrational raman scattering, and a broad emission band 28 centered at approximately 440 nanometers, believed to be produced largely by direct glucose and water fluorescence.
• the wavelength of the peak associated with the narrow raman scattering band is approximately 30 to 50 nanometers longer than the wavelength of the excitation light and shifts generally in proportion to shifts in the wavelength of the excitation light.
• the shape and emission wavelengths of the broad glucose emission band generally is not a direct function of the excitation wavelength.
• the emission intensity associated with eight representative wavelengths does not vary linearly with glucose concentration over the clinically relevant range of 80 to 300 milligrams per deciliter.
• the eight representative wavelength are indicated by the vertical lines in the graph of FIG. 2.
• the relationship between measured intensity and glucose concentration is highly nonlinear and presents a different profile at different wavelengths. More particularly, as the glucose concentration in water increases, the intensity at a wavelength of 370 nanometers generally increases as the glucose concentration increases until the concentration reaches about 500 milligrams per deciliter. At this point, the intensity then begins to taper off or decrease with increasing concentration. Similarly, the intensity at at a wavelength of 347 nanometers, generally the wavelength of the raman scattering peak generally increases and then decreases with increasing glucose concentration. Note however, that the rate of change for the intensity versus glucose concentration is different for each of the curves.
• Polynomial curve fitting for providing a predictive model is achieved using statistical techniques based on a least squares regression method.
• a common regression technique known as partial least squares (PLS) regression has been found to provide a robust model in that the model parameters do not change significantly when new samples are taken.
• PLS partial least squares
• the PLS regression technique begins by "autoscaling" each variable such that all the variables are equally influential in the prediction.
• the PLS regression technique uses principle component analysis, also known as singular value decomposition or eigenvector analysis, to represent the dependent and independent matrices.
• principle component analysis a NIPALS algorithm is used to define a data matrix of independent variables.
• PLS regression techniques introduce a weighting factor into the regression model.
• the PLS algorithm gives a sequence of models, the best model being the one that minimizes the cross-validation.
• a data matrix of independent variables (the glucose concentration is the dependent variable) , consisting of the emission intensity at the different wavelengths, is provided to a data processing routine that performs the PLS regression.
• the data processing routine is included in the "PLS_Toolbox Version 1.3" available from Barry M. Wise, 1415 Wright Avenue, Richland, WA 99352 (E-mail: bm_wise@pnl.gov) .
• the routines in the "Toolbox” are presently intended for use with the MATLABTM software package available from The Mathworks, Inc., 24 Prime Park Way, Natick, MA 01760.
• the matrix associated with the spectral intensities at each wavelength and the matrix associated with the concentration values have their means removed before processing.
• the routine calculates a regression vector shown in FIG. 4 and in Table II below.
• the scalar components of the regression vector are the prediction coefficients for each wavelength.
• the intensity at each of the eight wavelengths is measured. These eight measured values are scaled and multiplied by the regression vector, i.e., the eight wavelength coefficients in Table II. The result is a scaled concentration prediction.
• the scaled predicted concentration must be rescaled to provide a concentration value in the original units.
• the cross- validation calculation is used to determine the optimum number of latent variables to use in the model.
• the cross-validation is performed by using one spectra chosen at random to test the model.
• the cross-validation is repeated ten times, randomly choosing a different spectra to test the model.
• the results of the cross-validation are shown in the press plot of FIG. 5 as a plot of the prediction residual sum of squares (PRESS) versus the number of latent variables used in the model.
• the PLS analysis yielded a model of six latent variables.
• FIGS. 5 and 6 shows that the minimum PRESS exists between five to seven latent variables.
• FIG. 7 shows the results of a prediction test using samples of known glucose concentration in the PLS prediction model using six latent variables, derived from Table I, to define the model.
• the PLS model provides a fairly accurate prediction of the glucose concentration.
• the test was repeated for a multiple linear regression (MLR) model based on the same input data.
• the PLS model generally performs better than the MLR model at lower concentration levels while the MLR model performs better at at higher concentration levels.
• FIG. 8 shows the results of another prediction test again using samples of known glucose concentration in testing PLS and MLR models.
• the PLS model uses seven latent variables to define the model.
• both models provide substantially the same results.
• using additional latent variables in the model does not necessarily improve the model's prediction accuracy.
• the predictive model can be improved by using a greater number of wavelengths for generating the model.
• the emissions spectra from the 1,024 elements of the detector array provides a like number of intensity values. Approximately 200 of these points are in the wavelength range of glucose UV-visible emission light (approximately 335 to 450 nanometers) and the data is truncated to this range. To reduce the effects of noise, the spectra is measured three to five times for each glucose concentration. An average of each of these spectra is used to generate the model.
• the data for the truncated smoothed spectra was converted into a smaller file by averaging three points at a time to arrive at one point. For example, 180 points become 60 points.
• 60 wavelengths for each concentration level, preconditioned as discussed above, are analyzed in this example to arrive at a predictive model using the PLS regression technique, instead of the eight different wavelengths from Table 1 used in the previous example.
• the PRESS plot for the model using the whole spectra indicates a minimum PRESS at six latent variables.
• a test of the model using samples of known concentration is shown in FIG. 10.
• the PLS predictive model, using the preconditioned spectra provides a very accurate prediction of the glucose concentration. Given the generally noisy nature of the spectral measurements, and the non-linear relationship between the glucose concentration and the emission intensity at any given wavelength of interest, the results indicated in FIG. 10 are indeed surprising.
• FIG. 11 shows emission spectra, at a single glucose concentration, resulting from excitation light delivered at different intensity levels. As shown in Table IV below, the emission intensity at a wavelength associated with the raman peak, normalized with respect to the broader florescence peak, is nonlinear with respect to the excitation energy at given concentration level.
• Excitat ion Energy (mj/pul, se)
• Table IV can be used to provide a predictive model, using the PLS regression technique, as discussed above, with respect to Table I.
• the glucose concentration of an unknown sample can be determined using a predictive model provided by PLS analysis.
• FIG. 12 shows the emission spectrum of distilled water illuminated by excitation light having an energy of 5 millijoules per pulse (18 millijoules per millimeter per square millimeter) .
• the graph shows that the florescence spectra for distilled water exhibits a broad florescence band with a peak at approximately 370 nanometers and a narrow raman scattering band at approximately 346 nanometers.
• the raman scattering band results from scattered incident light having its wavelength shifted by the energy (rotational and translational) of the water molecules.
• the emission spectrum of ultra anhydrous glucose is shown in FIG. 13.
• the resulting spectrum has a single broad fluorescence band that peaks at approximately 450 nanometers.
• the emission spectrum of anhydrous glucose which has absorbed a small but spectrally significant amounts of water, exhibits two narrow raman scattering bands that peak at 341 nanometers and 346 nanometers, respectively, and one broad emission band that peaks at about 420 nanometers.
• the raman scattering peak at 346 nanometers corresponds to the raman peak of water shown in FIG. 12.
• the raman scattering peak at 341 nanometers apparently results from interaction between the water and glucose molecules.
• the spectrum of the anhydrous glucose is shifted to shorter wavelengths when compared with the spectrum of the ultra anhydrous glucose shown in FIG. 13.
• the emissions spectra of anhydrous glucose, as the excitation energy is varied, are shown in FIG. 15.
• the intensity of spectra generally increase as the excitation energy increases.
• the intensity ratio between the peaks of the raman bands and the broad emission band does not remain constant as the excitation energy increases.
• the ratio between the raman scattering band and the broad emission band similarly does not remain constant as the concentration increases. Accordingly, the interaction between the water and glucose molecules, and the energy density of the excitation light all appear to effect the resulting emission spectra. Accordingly, simple linear models are effective as an approximation only along very narrow, discrete segments of possible glucose concentrations of interest.
• the glucose concentration can be accurately predicted in spite of signal noise and nonlinear relationships between the glucose concentration and certain spectroscopic parameters of interest.
• the prediction is performed using a model developed from a PLS regression analysis.

Landscapes

• Health & Medical Sciences (AREA)
• Life Sciences & Earth Sciences (AREA)
• Physics & Mathematics (AREA)
• Medical Informatics (AREA)
• Surgery (AREA)
• Biophysics (AREA)
• Pathology (AREA)
• Engineering & Computer Science (AREA)
• Biomedical Technology (AREA)
• Heart & Thoracic Surgery (AREA)
• Veterinary Medicine (AREA)
• Molecular Biology (AREA)
• Optics & Photonics (AREA)
• Animal Behavior & Ethology (AREA)
• General Health & Medical Sciences (AREA)
• Public Health (AREA)
• Emergency Medicine (AREA)
• Spectroscopy & Molecular Physics (AREA)
• Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
• Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
• Investigating Or Analysing Materials By Optical Means (AREA)
• Investigating Or Analysing Biological Materials (AREA)

Priority Applications (7)

Applications Claiming Priority (2)

Publications (1)

Family

ID=24262374

Family Applications (1)

Country Status (10)

Cited By (3)

Families Citing this family (374)

Citations (7)

Family Cites Families (13)

• 1995
  • 1995-12-01 US US08/566,313 patent/US6232609B1/en not_active Expired - Lifetime
• 1996
  • 1996-11-19 AT AT96941397T patent/ATE249166T1/de not_active IP Right Cessation
  • 1996-11-19 AU AU10552/97A patent/AU715281B2/en not_active Ceased
  • 1996-11-19 EP EP96941397A patent/EP0863718B1/en not_active Expired - Lifetime
  • 1996-11-19 PT PT96941397T patent/PT863718E/pt unknown
  • 1996-11-19 ES ES96941397T patent/ES2206610T3/es not_active Expired - Lifetime
  • 1996-11-19 JP JP9521294A patent/JP2000501830A/ja not_active Ceased
  • 1996-11-19 DE DE69629937T patent/DE69629937T2/de not_active Expired - Fee Related
  • 1996-11-19 WO PCT/US1996/018532 patent/WO1997020495A1/en not_active Ceased
  • 1996-11-19 DK DK96941397T patent/DK0863718T3/da active
• 2005
  • 2005-08-01 JP JP2005223243A patent/JP2005351907A/ja active Pending

Patent Citations (7)

Also Published As

Similar Documents

Legal Events