This invention relates to apparatus and methods for monitoring the level of a predetermined substance, such as oxygen, in blood. The invention may employ invasive or non-invasive measurement techniques, and is suitable for determining blood oxygen saturation in patients in any context, for example central venous SO ₂ monitoring, pulmonary artery SO ₂ monitoring, extracorporeal SO ₂ monitoring, amputation level assessment and free-flap SO ₂ monitoring, for example.

The standard way to measure blood oxygen saturation in a patient is to direct light onto a target area of the patient's body containing the blood, and to measure the intensity of light reflected by or transmitted through the blood in the target area, either at discreet wavelengths or over a substantially continuous spectral range, and then to calculate the SO2 as a function of the measured light intensity values. Such measurement apparatus is known from WO 94/03102 for example. However many factors reduce the accuracy of such known SO2 measurement apparatus. One significant factor is the degradation of the accuracy of the apparatus due to patient movement, that is due to motion that leads to a change in the path length of the light through the biological tissue and hence to variation in the intensity of the detected light. This problem is particularly severe in critical health care applications where continuous monitoring is essential. Another factor that reduces the accuracy of such measurement apparatus is skin pigmentation, in that, for measurements made at the patient's skin, the quantity of detected light will vary depending on skin colour which ranges from fair to dark as the concentration of melanin increases.

Furthermore WO 00/09004 describes SO2 measurement apparatus that operates by passing light through biological tissue and monitoring the transmitted or reflected light continuously by means of a photodetector. However, the subsequent analysis of the detected light is based on the use of only a limited number of wavelengths, including some isobestic wavelengths at which oxygenated blood and deoxygenated blood absorb the same amount of light. The fact that only a limited number of wavelengths are used in such analysis in such apparatus limits the accuracy of the SO2 determination.

Furthermore the apparatus is sensitive to ambient light interference, for example due to the use of fluorescent lighting.

Furthermore WO 00/01294 describes apparatus for measuring one or more analytes in blood in a patient's body which comprises a light transmitter comprising a plurality of transmitting fibres positioned to transmit light to the body, and a light detector comprising a plurality of light detector fibres positioned to detect light transmitted through or reflected from the body. Light of only certain discreet wavelengths, including some of the isobestic wavelengths, is transmitted to and reflected from the blood, and accordingly such apparatus also suffers from limited accuracy and is sensitive to ambient interference.

It is an object of the invention to provide apparatus and methods for monitoring the level of a predetermined substance in blood that is of inherently greater accuracy in use.

According to one aspect of the present invention there is provided apparatus for monitoring the level of a predetermined substance in blood, comprising a light source for directing light onto a target area; a detector for detecting light from the source that has been reflected by or transmitted through blood in the target area; and an analyser for analysing the spectral content of the detected light, the analyser comprising first memory means for storing values representative of a first reference spectrum corresponding to a relatively low level of said substance in the blood, second memory means for storing values representative of a second reference spectrum corresponding to a relatively high level of said substance in the blood, third memory means for storing values representative of a third reference spectrum corresponding to changes in the spectral content of the detected light due to one or more factors other than the level of said substance in the blood, calculating means for calculating a set of coefficients relating the spectral content of the detected light to the first, second and third reference spectrum values relatively weighted by the coefficients, and determining means for determining a parameter representative of the level of said substance in the blood on the basis of the relative values of the coefficients calculated for weighting of the first and second reference spectrum values.

According to another aspect of the present invention there is provided a method of monitoring the level of a predetermined substance in blood, comprising directing light from a light source onto a target area; detecting light from the source that has been reflected by or transmitted through blood in the target area; obtaining values representative of a first reference spectrum corresponding to a relatively low level of said substance in the blood; obtaining values representative of a second reference spectrum corresponding to a relatively high level of said substance in the blood; obtaining values representative of a third reference spectrum corresponding to changes in the spectral content of the detected light due to one or more factors other than the level of said substance in the blood; calculating a set of coefficients relating the spectral content of the detected light to the first, second and third spectral values relatively weighted by the coefficients; and determining a parameter representative of the level of said substance in the blood on the basis of the relative values of the coefficients calculated for weighting of the first and second reference spectrum values.

According to a further aspect of the present invention there is provided a computer readable storage medium for monitoring the level of a predetermined substance in blood by analysing the spectral content of light that has been reflected by or transmitted through the blood, the storage medium comprising a program for carrying out the following analysis steps: obtaining values representative of a first reference spectrum corresponding to a relatively low level of said substance in the blood; obtaining values representative of a second reference spectrum corresponding to a relatively high level of said substance in the blood; obtaining values representative of a third reference spectrum corresponding to changes in the spectral content of the detected light due to one or more factors other than the level of said substance in the blood; calculating a set of coefficients relating the spectral content of the detected light to the first, second and third spectral values relatively weighted by the coefficients; and determining a parameter representative of the level of said substance in the blood on the basis of the relative values of the coefficients calculated for weighting of the first and second reference spectrum values.

Preferred embodiments of the claimed invention provide significant advantages in use over known techniques. Since the noise in the detection data is fairly evenly distributed, there is no need to remove the noise prior to analysing the data. It is also straightforward to determine the magnitude of the haemoglobin in the blood in order to make a subjective assessment as to the meaningfulness of the measured values in the case of SO2 monitoring. Prior art techniques actually amplify the noise content where only a small haemoglobin signal is present and this may result in meaningless SO ₂ values being obtained. It is also particularly advantageous if the detector is adapted to simultaneously detect light of different wavelengths that have been reflected by or transmitted through blood in the target area, to enable simultaneous analysis of the spectral content of said different wavelengths by the analyser. This enables the measurement to be made more quickly than in prior art arrangements in which detection of different wavelengths is made sequentially, and most importantly avoids any apparent change in the spectrum due to relative movement between the detector and the target area while the measurement is being effected. Where detection of all the different spectral components is being made simultaneously, any such movement will affect all the spectral components equally and does not adversely effect the measurement.

Furthermore the technique of the invention places no reliance on isobestic points, and would be valid even if the detected spectrum were decomposed into fundamental absorption peaks. By contrast prior art techniques are particularly sensitive to isobestic positioning, and the precise positioning of these points is known to change with temperature and pH, for example. Another advantage of the technique of the invention is that the input data for analysis is in the form of a set of independent coordinates (absorption versus wavelength), and it is possible to remove some of these coordinates without affecting the analysis. This would enable certain problem wavelengths (corresponding to fluorescent peaks, for example) to be eliminated, or values detected by faulty pixels in the detector pixel array to be eliminated completely. Furthermore the measurement results can be given with greater precision than with prior art techniques in which the precision is limited by the finite number of reference profiles.

In one embodiment of the invention the third memory means stores values representative of substantially linear variation of at least one of said factors with wavelength. These factors may comprise several different species of melanin that exhibit approximately linear absorption characteristics in the wavelength range of interest. However, the applied technique could easily be adapted so that the third memory means stores values representative of non-linear variation of at least one of the factors with wavelength.

Preferably, in addition to calculation of coefficients relating the spectral content of the detected light to the first, second and third reference spectrum values, the calculating means also serves to calculate an offset value that is substantially invariable with wavelength.

It is also a particular advantage of the invention that is enables the simultaneous monitoring of the levels of a number of different substances in blood.

In order that the invention may be more fully understood, a preferred embodiment of SO2 measurement apparatus in accordance with the invention will now be described, by way of example, with reference to the accompanying drawings, in which:

Figure 1 is a diagram of a detector assembly used in the apparatus;

Figure 2 is a graph of the extinction coefficients of Hb and HbO ₂ against wavelength;

Figure 3 is a graph of the extinction coefficients of pheomelanin and eumelanin against wavelength; and

Figure 4 is a graph showing the effect of ambient fluorescent light interference on the light level detected in use of the apparatus as a function of wavelength.

Figure 1 diagrammatically shows a detector assembly for monitoring the oxygenation of blood, as used in the SO ₂ measurement apparatus constituting a preferred embodiment

of apparatus in accordance with the present invention. In such apparatus a probe is has a LED at one end for directing white light onto a target area of a patient's body or a tube carrying the patient's blood, and an optical fibre for receiving the light reflected by or transmitted through the blood in the target area and for transmitting this to a monitoring unit incorporating a spectrometer assembly. The probe is coupled to the monitoring unit by a lead incorporating the optical fibre and the necessary electrical wiring for supplying power to the LED. In practice two or more probes will be coupled to the monitoring unit, including, in one implementation, a probe for monitoring the venous absorption spectrum and a probe for monitoring the arterial absorption spectrum. In alternative implementations of the invention a substantially greater number of probes may be provided for measuring the absorption at a number of points on the patient's skin.

The transmitted or reflected light from the target area is inputted to the spectrometer assembly 1 by means of the optical fibre 2 and is reflected by a mirror 3 towards a splitter 4 that serves to split the light into its constituent spectral components. The spectral components of the light are then reflected by a curved mirror 5 towards a linear CCD photodetector array 6 comprising a series of pixels spaced apart along a detection axis so that spectral components of different wavelengths are detected by different pixels and electrical output signals are obtained from the photodetector 6 indicative of the relative intensities of the detected spectral components. The light transmitted or reflected by the venous blood and the arterial blood may be detected separately by a common detector assembly or by means of two separate detector assemblies. Other types of spectrometer can be used in alternative implementations of the invention.

In a typical measurement cycle of a preferred method in accordance with the invention, a "light sample" measurement is first obtained utilising the transmitted or reflected light from the target area as described above, followed by a "dark sample" measurement obtained under the same conditions as the "light sample" but with the light from the LED turned off. The output signals from the array for the dark sample are then subtracted, pixel by pixel, from the output signals from the array for the light sample. This compensates for first and second aberrations in the detected data. The first

aberration is a feature of CCD electronic detectors in that they give an output even when no light is present, and this output is a function of integration time, temperature and physical properties of the individual CCD detector. The second aberration is ambient light entering the system. The light-dark sample obtained in this manner is known as a remittance spectrum, and an absorbance spectrum is obtained from the remittance spectrum by applying the following formula A = -log (Rb / Rw) on a pixel by pixel basis, where Rb is the remitted blood sample as described above, and Rw is a remitted white reference sample.

Conceptually Rw is the sample that would have been obtained by measurement of the remitted spectrum of a pure white target, although in practice such a sample is never actually measured directly, due to the complex calibration procedure used in this technique. The inclusion of the white reference sample here compensates for the nonuniform nature of various parts of the apparatus, namely the detector, the light source etc. The log transform, using the mathematical identity log (AX) = log (A) + log (X) or log (AX) = k + log(X) where k = log (A), turns scalar changes in Rw and Rb into linear shifts in the absorbance spectrum so that, if, instead true remittance being measured, a constant fraction of remittance is measured, the log curve corresponds to the true absorbance plus a (different) constant. Thus, provided that the SO ₂ calculation can compensate for linear shifts in the absorbance curve, it is never necessary to know the illumination levels in absolute terms, and the same result is obtained regardless of how much of the remitted light is actually collected.

The generation of reference spectra corresponding to high and low levels of oxygenation of the blood in this method is substantially the same as in the method described in WO 00/09004 in that high oxygen saturation of a sample of blood is artificially induced, after which, when stability has been reached, the current absorbance curve is taken as the high level reference spectrum, along with a corresponding SO ₂ value obtained from a blood gas analyser. Blood gas analysers are both invasive and destructive, but give much higher precision. The measurement step is then repeated with a sample of blood of low oxygen saturation to obtain the low level reference spectrum.

Figure 2 is a graph of the light absorption by blood over a range of wavelengths expressed in terms of the extinction coefficient at each wavelength for fully deoxygenated blood Hb and fully oxygenated blood HbO ₂ . In practice the absorption spectra, corresponding to the low level reference spectrum and the high level reference spectrum referred to above, measured for an actual patient will always be for blood that has some intermediate level of oxygenation between 0% and 100%, and the detected values obtained for venous and arterial blood will correspond to two different levels of oxygenation between 0% and 100%. The function of the apparatus is to determine what levels of oxygenation correspond to the detected spectral values.

Referring to the two absorption spectra 10 and 11 of Figure 2 corresponding to the extinction coefficients of Hb and HbO ₂ , it will be appreciated that these exhibit various peak values and additionally a number of wavelengths, that is the so-called isobestic wavelengths, at which the light absorption of the blood is independent of the degree of oxygenation. In analysing the detected spectral values obtained with a particular sample of blood, it is necessary to determine the relative contributions to the absorption spectrum of deoxygenated haemoglobin Hb and oxygenated haemoglobin HbO ₂ . It is known to match the measured absorption, reflectance or transmission spectrum against two or more reference spectra representing different predetermined levels of blood oxygenation such that the match yields a measure of actual blood oxygen saturation.

It may be shown that, for a medium consisting of several absorbers A ₁ , A ₂ , A ₃ etc., the resulting absorption spectrum takes the form:

A(λ) = l _a ( C ₁ C ₁ (X) + c ₂ e ₂ (λ) + c ₃ e ₃ (λ) + )

where l _a is the average photon path length, cl, c2, c3 ... are the molar concentrations of the absorbing species, and e^λ), e ₂ (λ), e ₃ (λ)... are the extinction coefficients of the absorbers expressed as functions of wavelength.

In the measurements made by the apparatus of the invention we are concerned with three principle absorbing species, namely melanin (actually several different species in various proportions), HbO ₂ and Hb.

The extinction coefficients of these absorbers shown in Figure 2 and 3 (Figure 3 showing separately the extinction coefficients of pheomelanin and eumelanin) give rise to the familiar double hump absorption spectra between 500 nm and 600 nm.

It is known that in tissue the melanin species vary but importantly all exhibit approximately linear absorption characteristics within the wavelength range of interest. It is therefore possible to model the total melanin contribution as a linear function of wavelength:

A _m (λ) = m _m λ + c _m

It is therefore possible to model the detected absorption spectrum in terms of four coefficients m _m) c _m , CHbO2 and CHb as follows:

A(λ) = c _m + m _m λ + Cπb02-eHb02(λ) + Cπb-eHb(λ)

where eii _b0 2(λ) and eii _b (λ) are the extinction functions of HbO ₂ and Hb shown in Figure 2, and m _m) c _m , CH _bO 2 and CH _b are the set of coefficients characterising the sample.

Absorbance as a measurement is independent of integration time. However it is convenient to alter the integration time to optimise the dynamic range of the received signal whilst still using a white light reference derived at a fixed integration time. This has the effect of introducing an offset in the absorbance curve approximating to the logarithm of the integration time ratios.

The other significant variable to be aware of relates to the relative concentrations of the three principle absorbers. For lower haemoglobin concentrations the absorption peaks are flatter.

The analysis arrived at in the analyser of the apparatus of the present invention uses a Levenberg-Marquardt algorithm to simplify the analysis process by directly decomposing the observed spectra into the four required coefficients. Such an algorithm is discussed in Marquardt, D. M., "An Algorithm for Least Squares Estimation of Nonlinear Parameters", J. 5Oc. Ind. Appl. Math. 11, 431-441 (1963). The method is also described in "Numerical Recipes in C++, The Art of Scientific Computing", Second Edition, Cambridge University Press 1988-2002 by William H. Press, Saul A. Teukolsky, William T. Vetterling, Brian P. Flannery, Chapter 15.5, pages 686 to 694. The method uses an elegant mathematical technique to perform multivariate non-linear least squares fitting. The algorithm requires a definition of all the individual absorption functions into which the sample curve is to be split and provides the relative amplitudes of these components which best match the detected absorption spectrum of the sample. Other algorithms can also be used for such analysis, such as the Newton Raphson algorithm, as described in "Numerical Recipes in C++, The Art of Scientific Computing", Chapter 9.4, pages 366 to 373.

Since the oxygen saturation SO ₂ is by definition the ratio OfHbO ₂ to total HbO ₂ + Hb it follows that it can then be directly calculated from the coefficients without needing to apply the normalisation required in prior art techniques.

Since the noise in the data sets characterising the detected absorption spectrum is normally distributed there is no requirement to remove it prior to fitting. Noisy and filtered versions of the same data converge to the same solution.

By extracting the magnitude of the haemoglobin contribution to the detected absorption spectrum it is possible to make a subjective decision as to its usefulness. This is an improvement over prior art techniques which, by manipulating (normalising) the spectrum, actually amplify the noise content in the case where only a small haemoglobin signal is present. This could potentially lead to meaningless SO ₂ values in tissue samples where some semblance of fit still exists.

The algorithm treats the input data characterising the detected absorption spectrum as a set of independent coordinates (absorption versus wavelength). Furthermore it is possible to remove certain coordinates without affecting the convergence. This could be useful in eliminating problem wavelengths (fluorescent etc) or building a bad pixel list if it becomes commercially prudent to do so.

The technique places no reliance on isobestics points, etc. and will be equally valid should it be required to decompose the spectra into fundamental absorption peaks. The key difference between the preferred method of the invention and the prior art method of WO 00/09004 relates to the manner in which the SO ₂ estimation is obtained from a given run time absorbance curve. The method of the invention relies on designing a mathematical model for absorbance, and using this model to calculate concentrations for Hb, HbO ₂ , melanin, dc shift values and any other analyte that has a characteristic spectrum. On the other hand the prior method of WO 00/09004 relies on the numerical values at the isobestic points to allow dc offset and melanin to be compensated for, and on comparing the measured curve to all 101 possible mathematically generated SO ₂ curves from 0 to 100, the right curve being that which gives the least sum of pixel by pixel differences between the reference curve and the measured curve. In such prior methods, an error in the measurement of those few pixels around the isobestic points could throw out the whole calculation. By contrast, in use of the method of the invention, those points have no more weighting than all the other pixels.

Various modifications of the above described analysis are also possible within the scope of the invention to take account of different applications. In particular the technique can be adapted for the purpose of detecting different analytes in blood other than oxygen. For example one possibility would be for the technique to be modified to detect the level of carbon monoxide in blood, that is the level of haemoglobin bound to carbon monoxide as HbCO. The technique could also be adapted to compensate for the presence of interferents other than melanin. A particular example would be to add a term to the algorithm enabling the interferents due to fluorescent lighting, which can have the effect of disrupting the absorbance spectrum as shown in Figure 4 for example, to be compensated so as to either improve the signal quality or identify possible error

conditions (or a combination of the two). Advantageously the technique can be used to simultaneously measure the level of three or more analytes in the blood.

Furthermore, since the linear melanin model already discussed with reference to Figure 2 is only an approximation, the technique could be modified to include a non-linear melanin function.

The part of the detected absorption spectrum that is usually used for the analysis is in the blue-green to yellow-red areas, but the unique signatures of the HbO ₂ and Hb spectra extend right down to the blue part of the visible spectrum, and this area could potentially be used in the analysis provided that the detector is sensitive enough to resolve the reduction in signal due to the increased extinction coefficient in this area.

In more complex models it is possible to add restraints to the parameters to prevent convergence to unrealistic solutions.