US20130024129A1

US20130024129A1 - Detecting chemical components from spectroscopic observations

Info

Publication number: US20130024129A1
Application number: US13/625,707
Authority: US
Inventors: Steven E. Pav
Original assignee: Nellcor Puritan Bennett LLC
Current assignee: Nellcor Puritan Bennett LLC
Priority date: 2008-03-31
Filing date: 2012-09-24
Publication date: 2013-01-24
Also published as: US8292809B2; US20090247844A1

Abstract

Embodiments disclosed herein may include methods and systems capable of estimating the underlying concentrations of chromophores in a sample. The photon scattering and absorption model may be based on Laplace and stable distributions, which may reveal that measurements in diffuse reflectance may follow a Beer-Lambert and Kohlrausch-Williams-Watts (KWW) product. This Beer-Lambert portion of the product may dominate in high absorption sample areas, while the KWW portion of the product may dominate in low absorption sample areas.

Description

RELATED APPLICATIONS

This application is a divisional of U.S. application Ser. No. 12/412,956, filed Mar. 27, 2009, which claims priority to U.S. Provisional Application No. 61/072,580, filed Mar. 31, 2008, the disclosures of which are hereby incorporated by reference in their entirety.

BACKGROUND

The present disclosure relates generally to the field of spectroscopy and, more particularly, to a system and method of optimizing the processing spectroscopic data.
This section is intended to introduce the reader to various aspects of art that may be related to various aspects that are described and/or claimed below. This discussion is believed to be helpful in providing the reader with background information to facilitate a better understanding of these various aspects. Accordingly, it should be understood that these statements are to be read in this light, and not as admissions of prior art.
Spectroscopy may be employed to ascertain the existence and/or concentration of component chemicals in a sample. To perform a spectroscopic analysis on a sample, a source may first send electromagnetic radiation through the sample. The spectrum of electromagnetic radiation which passes through the sample may indicate the absorbance of the sample. Based on the amount and spectrum of the sample absorbance, the presence and/or concentration of distinct chemicals may be detected by employing methods of spectrographic data processing.
Typically, the analysis includes modeling the underlying concentrations of chromophores in a sample from spectroscopic observations. The most common method for estimating these chromophores concentrations includes applying a photon scattering and absorption model based solely on the Beer-Lambert Law and utilizing multiple linear regression techniques to approximate the chromophores concentrations. However, the current methods may result in errors on the order of several percent. As such, a method and system for closer approximation of underlying concentrations of chromophores in a sample from spectroscopic observations is needed.

SUMMARY

Certain aspects commensurate in scope with the originally claimed subject matter are set forth below. It should be understood that these aspects are presented merely to provide the reader with a brief summary of certain embodiments and that these aspects are not intended to limit the scope of the claims. Indeed, the claims may encompass a variety of aspects that may not be set forth below.
In accordance with an embodiment, a method of processing spectrographic data may include transmitting an optical signal from an emitter to a sample, receiving the optical signal having passed through the sample at a detector, and analyzing the data associated with the received sample by numerically calculating an approximation of underlying concentrations of chromophores by applying a photon scattering and absorption model based on a mixed Beer-Lambert/Kohlrausch-Williams-Watts Model (KWW) for photon diffusion. In a another embodiment, a method for using Kernel Partial Least Squares (KPLS) Regression to formulate a model to be used in conjunction with analyzing spectrographic data includes collecting a number of data samples of optical signals passed through a sample from an emitter to a detector, measuring an affine function of the concentrations of the components for each given sample, and performing a KPLS regression to find a model for estimating future spectroscopic data.

BRIEF DESCRIPTION OF THE DRAWINGS

Certain embodiments may be understood upon reading the following detailed description and upon reference to the drawings in which:

FIG. 1 illustrates a perspective view of a pulse oximeter in accordance with an embodiment;

FIG. 1A illustrates a perspective view of a sensor in accordance with the pulse oximeter illustrated in FIG. 1;

FIG. 2 illustrates a simplified block diagram of a pulse oximeter in FIG. 1, according to an embodiment;

FIG. 3 illustrates a graph of diffuse reflectance of a sample area measured by the pulse oximeter in FIG. 1, according to an embodiment.

DETAILED DESCRIPTION

Various embodiments will be described below. In an effort to provide a concise description of these embodiments, not all features of an actual implementation are described in the specification. It should be appreciated that in the development of any such actual implementation, as in any engineering or design project, numerous implementation-specific decisions must be made to achieve the developers' specific goals, such as compliance with system-related and business-related constraints, which may vary from one implementation to another. Moreover, it should be appreciated that such a development effort might be complex and time consuming, but would nevertheless be a routine undertaking of design, fabrication, and manufacture for those of ordinary skill having the benefit of this disclosure.
The present disclosure is related to a photon scattering and absorption model which may be applied as an alternative to a strict application of the Beer-Lambert Law for estimation of the underlying concentrations of chromophores in a sample. The photon scattering and absorption model may be based on Laplace and stable distributions which reveal that measurements in diffuse reflectance may follow a Beer-Lambert and Kohlrausch-Williams-Watts (KWW) product. This Beer-Lambert portion of the product may dominate in high absorption sample areas, while the KWW portion of the product may dominate in low absorption sample areas.
Turning to FIG. 1, a perspective view of a medical device is illustrated in accordance with an embodiment. The medical device may be a pulse oximeter 100. The pulse oximeter 100 may include a monitor 102. The monitor 102 may be configured to display calculated parameters on a display 104. As illustrated in FIG. 1, the display 104 may be integrated into the monitor 102. However, the monitor 102 may be configured to provide data via a port to a display (not shown) that is not integrated with the monitor 102. The display 104 may be configured to display computed physiological data including, for example, an oxygen saturation percentage, a pulse rate, and/or a plethysmographic waveform 106. As is known in the art, the oxygen saturation percentage may be a functional arterial hemoglobin oxygen saturation measurement in units of percentage SpO₂, while the pulse rate may indicate a patient's pulse rate in beats per minute. The monitor 102 may also display information related to alarms, monitor settings, and/or signal quality via indicator lights 108.
To facilitate user input, the monitor 102 may include a plurality of control inputs 110. The control inputs 110 may include fixed function keys, programmable function keys, and soft keys. Specifically, the control inputs 110 may correspond to soft key icons in the display 104. Pressing control inputs 110 associated with, or adjacent to, an icon in the display may select a corresponding option. The monitor 102 may also include a casing 118. The casing 118 may aid in the protection of the internal elements of the monitor 102 from damage.
The monitor 102 may further include a sensor port 112. The sensor port 112 may allow for connection to an external sensor. FIG. 1A illustrates a sensor 114 that may be used with the monitor 102. The sensor 114 may be communicatively coupled to the monitor 102 via a cable 116 which connects to the sensor port 112. The sensor 114 may be of a disposable or a non-disposable type. Furthermore, the sensor 114 may obtain readings from a patient, which can be used by the monitor to calculate certain physiological characteristics such as the blood-oxygen saturation of hemoglobin in arterial blood, the volume of individual blood pulsations supplying the tissue, and/or the rate of blood pulsations corresponding to each heartbeat of a patient. The sensor 114 and the monitor 102 may combine to form the pulse oximeter 100.
Turning to FIG. 2, a simplified block diagram of a medical device is illustrated in accordance with an embodiment. The medical device may be the pulse oximeter 100. The pulse oximeter 100 may include a sensor 114 having one or more emitters 202 configured to transmit electromagnetic radiation, i.e., light, into the tissue of a patient 204. For example, the emitter 202 may include a plurality of LEDs operating at discrete wavelengths, such as in the red and infrared portions of the electromagnetic radiation spectrum. Alternatively, the emitter 202 may be a broad spectrum emitter, or it may include wavelengths for measuring water fractions.
The sensor 114 may also include one or more detectors 206. The detector 206 may be a photoelectric detector which may detect the scattered and/or reflected light from the patient 204. Based on the detected light, the detector 206 may generate an electrical signal, e.g., current, at a level corresponding to the detected light. The sensor 114 may direct the electrical signal to the monitor 102 for processing and calculation of physiological parameters.
In this embodiment, the monitor 102 may be a pulse oximeter, such as those available from Nellcor Puritan Bennett L.L.C. The monitor 102 may include a light drive unit 218. Light drive unit 218 may be used to control timing of the emitter 202. An encoder 220 and decoder 222 may be used to calibrate the monitor 102 to the actual wavelengths being used by the emitter 202. The encoder 220 may be a resistor, for example, whose value corresponds to the actual wavelengths and to coefficients used in algorithms for computing the physiological parameters. Alternatively, the encoder 220 may be a memory device, such as an EPROM, that stores wavelength information and/or the corresponding coefficients. For example, the encoder 220 may be a memory device such as those found in OxiMax® sensors available from Nellcor Puritan Bennett L.L.C. The encoder 220 may be communicatively coupled to the monitor 102 in order to communicate wavelength information to the decoder 222. The decoder 222 may receive and decode the wavelength information from the encoder 220. Once decoded, the information may be transmitted to the processor 214 for utilization in calculation of the physiological parameters of the patient 108.
Further, the monitor 102 may include an amplifier 208 and a filter 124 for amplifying and filtering the electrical signals from the sensor 114 before digitizing the electrical signals in the analog-to-digital converter 212. Once digitized, the signals may be used to calculate the physiological parameters of the patient 204. The monitor 102 may also include one or more processors 214 configured to calculate physiological parameters based on the digitized signals from the analog-to-digital converter 212 and further using algorithms programmed into the monitor 102. The processor 214 may be connected to other component parts of the monitor 102, such as one or more read only memories (ROM) 216, one or more random access memories (RAM) 218, the display 104, and the control inputs 110. The ROM 216 and the RAM 218 may be used in conjunction, or independently, to store the algorithms used by the processors in computing physiological parameters. The ROM 216 and the RAM 218 may also be used in conjunction, or independently, to store the values detected by the detector 206 for use in the calculation of the aforementioned algorithms.
In an embodiment, the algorithm stored in the ROM 216 for use by the processor 214 to compute physiological parameters may be a Beer-Lambert and Kohlrausch-Williams-Watts (KWW) product for measuring characteristics of a sample, such as chromophore concentrations in a patient 204. The probability that an emitted photon passes through a sample and arrives at a detector 206 is
$I (μ_{a}, μ_{s}, g) = \int_{0}^{\infty} e^{- x μ_{a}} f (x) \partial x .$
In this expression, μ_smay represent the scattering coefficient of the medium and g may represent the anisotropy coefficient of the medium. Furthermore, μ_amay be the absorption coefficient. This expression may be derived by assuming that ƒ is the density function for photon path lengths for a fixed configuration of an emitter 202, a detector 206, and a sample site, for example, on a patient 204. Assuming that the medium is non-absorbing at the given wavelength supplied by the emitter 202, i.e μ_a=0, then the function for determining the probability of a photon passing through a the zero absorption sample across a distance l, where l is a distance between a to b, (where a to b may be the maximum distance through the medium between the emitter 202 and the detector 206), may be found by ∫_a ^bƒ(x)dx. Therefore, in the absence of absorption,
I(0,μ_s ,g)=∫₀ ^∞ƒ(x)dx
where I(μ_a, μ_s, g) represents the detected intensity at the detector 206 for the given absorption, scattering, and anisotrophy coefficients μ_a, μ_s, and g. However, real world situations may occur where the absorption coefficient does not equal zero.
In the case where absorption does not equal zero, according to the Beer-Lambert Law, the probability that a single photon traveling a distance l through a medium with an absorption coefficient of μ_awill be absorbed is equal to e^−tμ ^a, which follows from the memoryless property and definition of μ_a. Combined, this yields
$I (μ_{a}, μ_{s}, g) = \int_{0}^{\infty} e^{- x μ_{a}} f (x) \partial x .$
Moreover, I(μ_a, μ_s, g), may then be the Laplace transform of ƒ i.e. I(μ_a, μ_s, g)=L{f}(μ_a). Thus, the probability that an emitted photon passes through the sample of, for example, a patient 204 is
$\int_{0}^{\infty} e^{- x μ_{a}} f (x) \partial x = ℒ {f} (μ_{a}) .$
Moreover, the path length distribution function, ƒ(x), may be shown to follow a sum-stable distribution. The Laplace transform of a stable distribution with a parameter α is e^−s ^α. Therefore, since ƒ(x) follows a stable distribution, then I (μ_a, μ_s, g) should contain a factor of the form e^−μ ^a ^β. Modeling ƒ(x) for the KWW distribution results in
I(μ_a,μ_s ,g)=C ₂(μ_s ,g)e ^−C ² ^(μ ^s ^,g)μ ^a ^β.
In this equation, C₁(μ_s, g) may be strictly due to scattering and the geometry of the emitter 202, the detector 206, and a sample site, for example, on a patient 204. However, since ƒ(x)=0 for all x smaller than the Euclidian distance from the source to the detector, ƒ(x) should be a shift of a stable distribution. Addition of an extra factor of e^−C ^3μato the Laplace transform compensates for the shift, where C₃may represent the offset distance. The offset distance may be equal to the Euclidean distance between the emitter 202 and the detector 206. Inclusion of the shift factor results in
I(μ_a,μ_s ,g)=C ₁(μ_s ,g)e ^−(C ² ^(μ ^s ^,g)μ ^a ^β ^+C ³ ^μ ^a ⁾.
In this embodiment, the model can be extended to include the case of collimated, i.e. non-diffused, light where some of the light detected has not been scattered, while other portions of the light has been scattered. For this embodiment, the path length distribution function, ƒ(x), can be described as
ƒ(x)=g(x)+C ₄ e ^−C ³ ^μ ^sδ(x−C ₃).
Here, g(x) may be a stable distribution, C₄may represent a coefficient inclusive of the intensity of the emitter 202 and the coupling efficiency of the test geometry, and δ(x) may be the Dirac delta. The coupling efficiency of the test geometry may take include such factors as the aperature size of the detector 206 as well as the beam diameter. By the linearity of the Laplace transform, this yields
I(μ_a,μ_s ,g)=C ₁(μ_s ,g)e ^−(C ² ^(μ ^s ^,g)μ ^a ^β ^+C ³ ^μ ^a ⁾ +C ₄ e ^−C ^g ^(μ ^s ^+μ ^a ⁾.
This equation represents the general attenuated KWW model for the detected intensity at the detector 206 for the given absorption, scattering, and anisotrophy coefficients μ_a, μ_s, and g. This general attenuated KWW model may be stored in the ROM 216 for use by the processor 214 in calculating physiological parameters based on the digitized signals from the analog-to-digital converter 212.
In an embodiment (in the case of diffuse reflectance), the second summand equals zero, for the case when the detector 206 may not be located in the beam path of the emitter 202. The log of the general attenuated KWW model may be taken, resulting in
log I(μ_a,μ_s ,g)=−log C ₁(μ_s ,g)+C ₂(μ_s ,g)μ_a ^β +C ₃μ_a.
As log C₁(μ_s, g) can be estimated, then log C₁(μ_s, g)−log I(μ_a, μ_s, g) versus μ_amay be plotted graphically. FIG. 3 illustrates a graph 300 of log C₁(μ_s, g)−log I(μ_a, μ_s, g) versus μ_a. As seen from the graph 300, diffuse reflectance 302 may closely follow the predicted KWW model 304 of diffuse reflectance in sample areas with low absorption rates. Conversely, diffuse reflectance 302 may closely follow the predicted Beer-Lambert model 306 of diffuse reflectance in sample areas with high absorption rates. The crossover point 308 where the two terms trade dominance occurs at
$μ_{a} = {(C_{2} (μ_{s}, g) / C_{3})}^{\frac{1}{1 - β}},$
while far from the crossover point on each end of the diffuse reflectance 302 may be well approximated by the summands C₂(μ_s, g)μ_a ^β for the predicted KWW model 304, and C₃μ_afor the Beer-Lambert model 306.
The tendencies of the KWW model 304 and the Beer-Lambert model 306 may be used in the estimation of the concentrations of chemical components of known absorptions. Thus, when a sample consists of l chemical components of varying concentrations c may be subject to light of different wavelengths, and the intensity at the detector 206 has been recorded, then the bulk absorption coefficient may be proportional to U_ac, where U_arepresents the matrix of absorption coefficients of the different components l. If μ_sis taken to vary slowly with respect to the wavelength, then the offset and scaling factors will vary slowly with respect to the wavelength, and may be approximated with, for example, B-splines or quadratic polynomials. Thus, the general attenuated KWW model becomes
m=F ₁ c ₁+(F ₂ c ₂)⊙(U _a c)^β +C ₃(U _a c),
where m represents the vector of the negative log intensity values, F₁and F₂may represent matrices whose columns span the spaces containing the approximations of offset and scaling values, and “⊙” represents the Hadamard, i.e. element by element, product. Furthermore, (U_ac)^β may be a Hadamard exponential.
For given estimates of c and p, the optimal values for c₁and c₂may be easily computed. Thus, determining the values used as estimations for c and β remains. Let ĉ, {circumflex over (β)}, ĉ₁, ĉ₂, Ĉ₃represent estimates of the unknown quantities. The residual may then be defined as
ε=m−F ₁ ĉ ₁−(F ₂ ĉ ₂)⊙(U _a ĉ)^{{circumflex over (β)}} −Ĉ ₃(U _a ĉ),
while the square error of the approximation may be
Ø(ĉ ₁ ,ĉ ₂ ,Ĉ ₃ ,ĉ,{circumflex over (β)})=ε^Tε.
Therefore, to find the ĉ₁ĉ₂and Ĉ₃, which minimize ⊙ for fixed ĉ, {circumflex over (β)}, we may estimate the vector of the negative log intensity values as
$\begin{matrix} m \approx F_{1} {\hat{c}}_{1} + diag ({(U_{a} \hat{c})}^{\hat{β}}) F_{2} {\hat{c}}_{2} + {\hat{C}}_{3} U_{a} \hat{c} \\ = [F_{1} diag ({(U_{a} \hat{c})}^{\hat{β}}) F_{2} U_{a} \hat{c}] [\begin{matrix} {\hat{c}}_{1} \\ {\hat{c}}_{2} \\ {\hat{C}}_{3} \end{matrix}] \\ = A [\begin{matrix} {\hat{c}}_{1} \\ {\hat{c}}_{2} \\ {\hat{C}}_{3} \end{matrix}], \end{matrix}$
where diag (ν) represents the square diagonal matrix with diagonal ν and A=(ĉ,{circumflex over (β)})=
A=(ĉ,{circumflex over (β)})=[F ₁diag((U _a ĉ)^{{circumflex over (β)})} F ₂ U _a ĉ].
The least squares optimal ĉ₁ĉ₂and Ĉ₃can then be described by the normal equation form
$[\begin{matrix} {\hat{c}}_{1} \\ {\hat{c}}_{2} \\ {\hat{C}}_{3} \end{matrix}] = {(A^{T} A)}^{- 1} A^{T} m .$
In an embodiment, the least squares solution may be calculated by the processor 214 using a software program which may be stored on ROM 216.
Ø may be considered a function of ĉ and {circumflex over (β)}. Optimization of Ø may be accomplished by an iterative numerical scheme used to compute the gradient of the objective with respect to the vector of free variables. In an embodiment, the iterative numerical scheme used may be the Broyden-Fletcher-Goldfarb-Shanno (BFGS) method. In another embodiment, the iterative numerical scheme used may be the conjugate gradient method. The gradient will depend on ĉ₁, ĉ₂, and Ĉ₃, and in an embodiment, the partial derivatives of ĉ₁, ĉ₂, and Ĉ₃may be incorporated into the computation. In another embodiment, computational time may be reduced by approximating the gradient by assuming fixed values for ĉ₁, ĉ₂, and Ĉ₃. Under this assumption, the gradients of Ø, which can be used to minimize Ø with respect to ĉ and {circumflex over (β)}, can be found from
$\frac{\partial \emptyset ({\hat{c}}_{1}, {\hat{c}}_{2}, {\hat{C}}_{3}, \hat{c}, \hat{β})}{\partial \hat{β}} = - 2 {(ε ⊙ F_{2} {\hat{c}}_{2} ⊙ {(U_{a} \hat{c})}^{\hat{β}})}^{T} \log (U_{a} \hat{c})$ $and$ $\nabla_{\hat{c}} \emptyset ({\hat{c}}_{1}, {\hat{c}}_{2}, {\hat{C}}_{3}, \hat{c}, \hat{β}) = - 2 U_{a}^{T} (\hat{β} ε ⊙ F_{2} {\hat{c}}_{2} ⊙ {(U_{a} \hat{c})}^{\hat{β} - 1} + {\hat{C}}_{3} ε) .$
When the processor 214 determines that the μ_avalues fall into the dominant region of either the predicted KWW model 304 or the predicted Beer-Lambert model 306, the final summand of
m=F ₁ C ₁+(F ₂ C ₂)⊙(U _a c)^{{circumflex over (β)}} +C ₃(U _a C)
may be eliminated. For example, when the Beer-Lambert model 306 dominates, then {circumflex over (β)} may tend towards “1”. Assuming that the space spanned by the columns of F₂represent a constant, which occurs if F₂spans a B-spline or a polynomial space (in μ_a). In this form, ĉ₁and ĉ₂may be found by
$[\begin{matrix} {\hat{c}}_{1} \\ {\hat{c}}_{2} \end{matrix}] = {(B^{T} B)}^{- 1} B^{T} m, where$ $B = B (\hat{c}, \hat{β}) = [F_{1} diag ({(U_{a} \hat{c})}^{\hat{β}}) F_{2}] .$
This results in the gradients of Ø being solved by
$\frac{\partial \emptyset ({\hat{c}}_{1}, {\hat{c}}_{2}, \hat{c}, \hat{β})}{\partial \hat{β}} = - 2 {(ε ⊙ F_{2} {\hat{c}}_{2} ⊙ {(U_{a} \hat{c})}^{\hat{β}})}^{T} \log (U_{a} \hat{c})$ $and$ $\nabla_{\hat{c}} \emptyset ({\hat{c}}_{1}, {\hat{c}}_{2}, \hat{c}, \hat{β}) = 2 \hat{β} U_{a}^{T} (ε ⊙ F_{2} {\hat{c}}_{2} ⊙ {(U_{a} \hat{c})}^{\hat{β} - 1}) .$
In another embodiment, a method for using KPLS Regression to formulate a model to be used in conjunction with analyzing spectrographic data may be employed. This method may include the preprocessing of the data with a nonlinear transform to a given space before performing the linear regression into that same space. This may be achieved by use of a kernel function, κ, which may be used to compute the dot product in the given space of two vectors in the data space, without having to perform a transform on that space. This may be accomplished by building a nonlinear model, which may begin with y, some affine function of the concentrations of the components of a given sample. The KPLS may proceed by collecting a number of data samples of optical signals passed through a sample from an emitter 202 to a detector 206 and measured, for example, spectrographically. The processor 214 may then measure an affine function y of the concentrations of the components for each given sample and store it in a vector y. The processor 214 may performing a KPLS regression to find a model of the form
$y_{i} \approx y_{o} + \sum_{j} a_{j} k (X_{j, :}, X_{i, :}) .$
This model may then be used to estimate y.
To determine κ, we can set a value for β, or we may determine it from the procedure described above. In either case, an x measurement should take the form of
x=h+m(μ_a ^T c)^β.
Since a lower bound on h can be determined for a given sensor 114, its value may be set as small relative to the second summand. From
$μ_{a}^{T} c = {[(x - h) / m]}^{\frac{1}{β}},$
using a Taylor expansion, we may determine
$μ_{a}^{T} c \approx {\frac{1}{m}}_{}^{\frac{1}{β}} [x^{\frac{1}{β}} - \frac{1}{β} x^{\frac{1}{β} - 1} h] .$
This suggests the use of
$κ (x, y) = {(x^{T} y)}^{\frac{1}{β}} or κ (x, y) = {(x^{T} y)}^{\frac{1}{β}} + {(x^{T} y)}^{\frac{1}{β} - 1}$
as kernels.
A specific embodiment may include β=½, where μ_s, and g are fixed. In this case, the function
(C ₂(μ_s ,g)μ_a ^β +C ₃μ_a)
may be an injunctive function, i.e. one-to-one, with respect to μ_a. Thus, the function has an inverse. As such, an optical observation may be transformed to a quantity proportional to μ_a, thus linearizing the signal. Therefore, once C₁(μ_sg) is estimated, the quantity
M(μ_a;μ_s ,g)=_dflog C ₁(μ_s ,g)−log I(μ_a,μ_s ,g)=C ₂(μ_s g)μ_a ^β +C ₃μ_a
may be computed from the optical observations of I(μ_a, μ_s, g). Moreover, since β=½, the observations are explicitly invertible to
$M (μ_{a}; μ_{s}, g) = C_{2} (μ_{s}, g) μ_{a}^{\frac{1}{2}} + C_{3} μ_{a}$ $0 = {C_{3} (\sqrt{μ_{a}})}^{2} + C_{2} (μ_{s}, g) \sqrt{μ_{a}} - M (μ_{a}; μ_{s}, g) .$
The quadratic equation may be used to yield:
$\sqrt{μ_{a}} = \frac{- C_{2} (μ_{s}, g) \pm \sqrt{{C_{2} (μ_{s}, g)}^{2} + 4 C_{3} M (μ_{a}; μ_{s}, g)}}{2 C_{3}},$
the negative root of which may be ignored to generate
$\sqrt{μ_{a}} = - \frac{C_{2} (μ_{s}, g)}{2 C_{3}} + \sqrt{{(\frac{C_{2} (μ_{s}, g)}{2 C_{3}})}^{2} + \frac{M (μ_{a}; μ_{s}, g)}{C_{3}}} .$
Furthermore, by letting κ=C₂(μ_s, g)/2C₃, the equation becomes:
$μ_{a} = {(- κ + \sqrt{κ^{2} + \frac{M (μ_{a}; μ_{s}, g)}{C_{3}}})}^{2} = 2 κ^{2} + \frac{M (μ_{a}; μ_{s}, g)}{C_{3}} - 2 κ \sqrt{κ^{2} + \frac{M (μ_{a}; μ_{s}, g)}{C_{3}}} .$
Accordingly, because C₁(μ_s, g) and C₂(μ_s, g) may depend on geometry and scattering, while C₃may depend on the test geometry, only the estimation of μ_s, and g is required to be made by the pulse oximeter 100. This may be accomplished through assumptions as to the tissue sample of the patient 204 which may be stored in the ROM 216 and/or the RAM 218 for use in the calculation of μ_a.
Another embodiment may be applied when observations are made over time with changes in the absorption of the medium and negligible changes in the scattering properties of the medium. For
$- \frac{\partial \log I (μ_{a}, μ_{s}, g, t)}{\partial t} = C_{2} (μ_{s}, g) {βμ}_{a}^{β - 1} \frac{\partial μ_{a}}{\partial t} + C_{3} \frac{\partial μ_{a}}{\partial t},$
it may be shown that
$- \frac{\partial \log I (μ_{a}, μ_{s}, g, t)}{\partial t}$
may be large when μ_ais small, which contrasts with the expected values from the Beer-Lambert Law that
$- \frac{\partial \log I (μ_{a}, μ_{s}, g, t)}{\partial t} = C_{3} \frac{\partial μ_{a}}{\partial t}$
holds steady for all values of μ_a. When observations are made at n wavelengths over a sample containing l chemical components of varying concentrations c(t), with U_aas the (n×l) matrix of absorption coefficients of the different components at the different wavelengths, then the vector of μ_aat the n wavelengths is U_ac(t). If m is the n-vector of observed
$- \frac{\partial \log I (μ_{a}, μ_{s}, g, t)}{\partial t}$
values at a fixed time at the given wavelength,
then
$m = [C_{2} (μ_{s}, g) {β (U_{a} c (t))}^{β - 1} + C_{3} \overset{⇀}{I}] ⊙ U_{a} \frac{\partial c (t)}{\partial t} .$
As such, when U, c(t) is estimated, then the left Hadamard multiplicand may be estimated, resulting in
$g \approx C_{2} (μ_{s}, g) {β (U_{a} c (t))}^{β - 1} + C_{3} \overset{⇀}{I}$ $and$ $m \approx (diag (g) U_{a}) \frac{\partial c (t)}{\partial t},$
for which
$\frac{\partial c (t)}{\partial t}$
be estimated, for example, using the least squares method.
Specific embodiments have been shown by way of example in the drawings and have been described in detail herein. However, it should be understood that the claims are not intended to be limited to the particular forms disclosed. Rather the claims are to cover all modifications, equivalents, and alternatives falling within their spirit and scope.

Claims

1. A method for determining the concentration of a component in a tissue sample of a patient, comprising:

collecting a plurality of physiologic signals from a tissue sample of a patient;

measuring, for each physiologic signal, a concentration of a component of the tissue sample; and

performing a Kernel Partial Least Squares regression on the concentration of the component.

2. The method of claim 1, comprising determining a model for the tissue sample based at least in part upon the Kernel Partial Least Squares regression.

3. The method of claim 2, comprising applying the model to a second tissue sample.

4. The method of claim 2, wherein the model comprises a nonlinear model.

5. The method of claim 1, comprising calculating a kernel function for use with the Kernel Partial Least Squares regression.

6. The method of claim 1, wherein collecting the plurality of physiologic signals comprises:

emitting light into the tissue of the patient;

detecting scattered or reflected light from the tissue; and

generating the plurality of physiologic signals corresponding to the detected light.

7. The method of claim 1, wherein the concentration comprises a concentration of chromophores in the tissue sample.

8. A system for determining the concentration of a component in a patient's tissue, comprising:

a sensor configured to obtain a plurality of physiologic signals from a patient;

a monitor configured to receive from the sensor the plurality of physiologic signals; and

a processor configured to:

measure, for each physiologic signal, a concentration of a component of the tissue; and

perform a Kernel Partial Least Squares regression on the concentration of the component.

9. The system of claim 8, wherein the sensor is configured to emit light into a tissue of the patient and to detect scattered or reflected light from the tissue.

10. The system of claim 9, wherein the plurality of physiologic signals corresponds to the detected light.

11. The system of claim 8, wherein the monitor comprises a pulse oximeter.

12. The system of claim 8, wherein the processor is configured to determine a model for the tissue based at least in part upon the Kernel Partial Least Squares regression.

13. The system of claim 12, wherein the processor is configured to apply the model to a second tissue sample.

14. The system of claim 12, wherein the model comprises a nonlinear model.

15. The system of claim 8, wherein the processor is configured to calculate a kernel function for use with the Kernel Partial Least Squares regression.

16. The system of claim 8, further comprising a memory device coupled to the processor and storing an assumption regarding the tissue, for use in the Kernel Partial Least Squares regression.

17. The system of claim 8, wherein the concentration comprises a concentration of chromophores in the tissue.

18. A medical device for determining the concentration of a component in a tissue sample, comprising:

a processor configured to:

collect a plurality of physiologic signals from a tissue sample of a patient;

measure, for each physiologic signal, a concentration of a component of the tissue sample; and

19. The medical device of claim 18, wherein the processor is configured to determine a model for the tissue sample based at least in part upon the Kernel Partial Least Squares regression.

20. The medical device of claim 19, wherein the processor is configured to apply the model to a second tissue sample.

21. The medical device of claim 18, wherein the medical device comprises a pulse oximeter.

22. A method for determining a chromophore concentration in a tissue area of a patient, comprising:

emitting electromagnetic radiation into a tissue area of a patient;

detecting radiation scattered and/or reflected by the tissue area;

generating a physiologic signal corresponding to the detected radiation;

collecting a plurality of data samples from the signal;

measuring, for each data sample, an affine function of a chromophore concentration; and

performing a Kernel Partial Least Squares regression to obtain a model for the tissue area.