JP4696300B2 - Baseline stabilization method using wavelength difference for biological optical measurement using near infrared light - Google Patents

Description

  The present invention uses a near-infrared light that stabilizes a baseline during measurement and a stable baseline during measurement in a biological optical measurement method such as a brain function measurement method using near-infrared light. The present invention relates to a biological light measurement device.

The brain function measurement method using near infrared spectrophotometry (NIRS) is a non-invasive estimation of each concentration in brain tissue based on the difference in near-infrared absorption spectra of oxygen-bound and oxygen-desorbed hemoglobin. It is a technique to do. Like fMRI, it can monitor oxygen metabolism associated with brain activity, has sub-second sub-second time resolution, and is an optical measurement, so it is more resistant to electromagnetic noise than MRI, MEG, EEG, etc. Another advantage is that the mounting equipment on the head is small and light. Therefore, it is expected to be applied to various objects in a wide range of fields including medicine and brain science, and the spread of measurement systems using stationary light has begun.

  However, in NIRS measurement using stationary light, various fluctuations that are not synchronized with brain function activity are observed (see Non-Patent Document 1), which is a problem in realizing highly reproducible measurement. It is known that some of these fluctuations are synchronized with physiological activities other than brain activities such as heartbeat and respiration (see Non-Patent Document 2), and bending of the neck, articulation of the jaw, and fiber cable It is also easily generated by moving. Currently, temporal averaging, multiple multiplication, and drift subtraction processing by linear approximation are used to eliminate such baseline time fluctuations. No proposal for countermeasures has been made.

In recent years, in estimation of each hemoglobin concentration using NIRS, a two-degree-of-freedom model that does not substantially take into account temporal changes in absorption, scattering, and light attenuation by other substances has been widely used (see Non-Patent Documents 3 and 4). ). However, it is a well-known fact that the absorbance of scattered light and other absorbing substances changes in the visible to near-infrared region with neural activity (see Non-Patent Document 5). In addition, since the aspect of Mie scattering of erythrocytes varies depending on the hematocrit value (see Non-Patent Document 6), the coupling between neural activity and capillary blood flow may cause a change in light scattering intensity. Furthermore, the contribution of temporal changes in light attenuation in the scalp and cerebrospinal fluid is also controversial.

Chance CE. et al., Proc. Natl. Acad. Sci. USA, 90, 3770-3774, (1993) Obrig H. et al., Neouroimage 12 (6): 623-639 (2000) Obrig H. et al., J. Cereb.Blood Flow Metab., 23, 1-18, (2002) Boas DA. Et al., NeuroImage, 23, S275-S288, (2004) Vellringer A. et al., Trend Neurosci., 20, 435-442, (1997) Steinke JM. Et al., Appl. Optics, 27, 4027-4033, (1988)

  As described above, measurement with NIRS using stationary light is problematic in that various fluctuations that are not synchronized with brain functional activity are observed. It is considered useful to consider the mechanism of occurrence of An object of the present invention is to eliminate baseline fluctuations in a brain function measurement method using NIRS and obtain an accurate measurement value.

The conventional NIRS measurement problems have the following theoretical background.
Spectroscopic model of measurement Of the light irradiated on the scalp surface, the light path of the scalp surface that propagates through the living tissue and reaches the brain tissue and then a certain distance away has a banana-shaped outline as shown in Fig. 1. It is known by calculation (Okada E. et al, Appl. Opt., 36, 21-31, (1997)). The light attenuation in this optical path can be approximately expressed by the modified Lambert-Beer law using the number of various absorbing molecules X _i (t), the absorption coefficient μ _{i, λ} and the scattering contribution S _λ (t). In actual NIRS, there are media such as transmission fiber, scalp, bone tissue, and cerebrospinal fluid in addition to brain tissue between the light source and the receiver, and light attenuation corresponding to the characteristics occurs under various conditions. If the attenuation coefficient in this process is now generally expressed as U _λ (t), the intensity I _λ (t) of light irradiated from the light source with the intensity of I _{λ, 0} and returning to the receiver is as follows: I can express. In the present specification, a hat symbol representing an estimated value may be represented as “X hat”, which means that a hat symbol is added on the letter X.

Absorbance in brain tissue is defined as follows.

Here, an absorbance change amount a _λ (t) at time t with reference to an absorbance value at arbitrarily defined time t = 0 is introduced.

By defining the amount of change x _i (t) of each absorbing molecule, the amount of scattering s _λ (t) in brain tissue, and the amount of light attenuation u _λ (t) in other pathways as follows: ,

From the number (1), the following relationship holds.

In measuring this a _λ (t), an observation error also occurs due to factors other than the above. These are left in the form of error terms n _λ (t), and the observation quantities are expressed as follows.

The change x _i (t) of the molecular species i can be expressed as the product of the average length l _λ (t) of the light propagation path in the brain tissue and the change c _i (t) of the average concentration: .

In general, it is considered that l _λ (t) changes with time as the absorption and scattering of the system changes. However, if the amount of change in absorption and scattering is small, l _λ (t) is almost stable over time. Can be considered. Therefore, in the following, the discussion proceeds as a quantity to obtain x _i (t). As can be seen from Equation (8), the observed change in absorbance a _λ (t) is the noise due to the light absorption of each absorbing molecule, light scattering on the light path in brain tissue, light attenuation in the measurement path, and others. Is the sum of contributions. It is an object of the present invention to estimate x _i (t) from any of these by some method.

Estimating the number of molecules of each substance by multi-wavelength measurement When the system is composed of N kinds of substance groups, x _i (t) is estimated by measuring the absorbance at N or more measurement wavelengths. This results in solving a linear simultaneous equation consisting of N or more equations having the form (8). The vertical vectorization of a _λ (t), s _λ (t), u _λ (t) and n (t) obtained at each wavelength is a (t), s (t), u (t ), N (t), and the vertical vector of the number-of-molecules variation x _i (t) of each molecular species i is x (t), the linear simultaneous equations are given as follows.

If N measurement wavelengths are selected, M is an N × N-dimensional matrix having the molecular absorption coefficient μ _{i, λ} of the molecular species i at the wavelength _λ as a matrix element. If M is regular, x (t) can be obtained using the inverse matrix M ⁻¹ . That is,

By the way, normally, the estimated value x hat (t) of the amount of change in the number of molecules must be calculated without having accurate information on s (t), u (t), and n (t). That is,

At this time, x hat (t) has the following.

  In addition to the target x (t), the number (13) includes the contribution of scattering s (t), light attenuation u (t) derived from the measurement path and its state, and other noise n (t). I understand that. As a technique for suppressing the influence of these additional terms, processing such as time averaging, multiple integration, and drift subtraction by linear approximation is usually performed. These can only reduce their contribution to components that vary independently of cranial nerve activity. It should also be noted that the time resolution and measurement time are sacrificed and some assumptions are introduced in the time variation of drift.

The observed spectroscopic change and the degree of freedom (13) of the estimation model can be solved by preparing an appropriate M for the estimated value vector x hat (t) of any number of dimensions. Conventionally, as a model to describe spectroscopic changes obtained from observation of brain activity, we focused only on changes in oxygen-binding hemoglobin (x ₁ ) and oxygen-desorbed hemoglobin (x ₂ ). Two-degree-of-freedom models have been widely used that assume that changes in tissue scattering s (t) and other light attenuation u (t) are small enough (Obrig H. et al., J. Cereb. Blood Flow Metab., 23, 1-18, (2002); Boas DA, et al., NeuroImage, 23, S275-S288, (2004)). From this standpoint, if the number of dimensions of a (t) in number (12), that is, the number of selected wavelengths, is greater than the degree of freedom 2 of the estimation model, in order to make the most desirable estimation in the sense of the least square error A pseudo inverse matrix M ⁺ is used instead of M ⁻¹ (BoasDA. Et al., NeuroImage, 23, S275-S288, (2004); Okamoto M. et al, NeuroImage, 21, 1275-1288, (2004) ).

Although this method can certainly give you the only estimated value x hat p (t), it is an error to think that the two-degree-of-freedom model is valid. Assuming that a _λ (t) is measured at a plurality of wavelengths of three or more, it is assumed that x ₁ (t) using any two measured values a _k (t) and a _l (t). ), X ₂ (t) can be estimated. If the two-degree-of-freedom model is valid, the estimates using any combination of a _k (t), a _l (t) should be in good agreement with each other. FIG. 4A shows an example of a discrepancy in time change between a plurality of estimated values obtained by such processing. The measurement was performed using a commercially available apparatus (Shimadzu Corporation, NIRStation) in which three observation wavelengths 780 nm, 805 nm, and 830 nm were set. Using the absorbance change a (t) obtained here, a _{780 nm} (t) and a _{805 nm} (t), a _{805 nm} (t) and a _{830 nm} (t), a _{830 nm} (t) and a _{780 nm} (t) X ₁ (t) and x ₂ (t) were estimated using the three combinations. FIG. 4A is an estimation example of the amount of change in each hemoglobin at the top of the head when an adult male performs a tapping task. Similar results to the discrepancies between the estimated values shown here were found in other channels of this measurement, and many other measurement examples gave similar results. Such a result is considered to show doubt about the validity of the two-degree-of-freedom model.

However, one of the causes of the discrepancy between the estimated values may be an error in the molecular absorption coefficient μ _{i, λ} that is an element of the matrix M in the equation (13). Actually reported molecular absorption coefficient of hemoglobin (Takatani S. et al., IEEE Trans. Biomed. Eng., BME-26 (12), 656-664, (1979); Wray S. et al., Biochim. Biophys. Acta, 933, 184-192, (1988); Matcher
SJ. Et al., Anal. Biochem. 227 (1): 54-68 (1995)) shows a discrepancy of up to 10%. There is also a possibility that the actual light source wavelength includes an error of about 10 nm from the assumed wavelength. In that case, according to the reported value, an error of the molecular absorption coefficient of about 10% at maximum is generated. These errors in M result in errors in the estimation of x (t). Therefore, in order to eliminate this error factor, a matrix M ′ in which the molecular absorption coefficient is optimized so as to minimize the variance between the estimated values was obtained. Assuming that the wavelength error is actually about ± 10 nm and each absorption coefficient may have an error accordingly, the result of optimizing M ′ and estimating x ₁ (t) and x ₂ (t) Shown in FIG. 4B.
Although the time variation of each estimated value is different from that in FIG. 4A, the three estimated values still do not sufficiently match each other. Therefore, it can be seen that the discrepancy between the estimated values cannot be sufficiently explained from the viewpoint of the error of the matrix M.

On the other hand, for spectroscopic changes associated with cranial nerve activity, neural excitation (Malonek D. et al.,
Science, 272 (26), 551-554 (1996); Rector DM. Et al, J. Neurophysiol. 78, 1707-1713 (1997)) and subsequent neuronal morphological changes (Malonek D. et al. , Science, 272 (26), 551-554 (1996); MacVicar BA. Et al., J. Neourosci., 11, 1458-1469, (1991))
Already pointed out the possibility of changes in light scattering intensity of brain tissue due to blood flow and changes in red blood cell Mie scattering due to changes in blood flow (Steinke JM. Et al., Appl. Optics, 27, 4027-4033, (1988)) Has been. These suggest the existence of spectroscopic degrees of freedom that are not included in the two-degree-of-freedom model.

  From the above, the present inventor made an estimation based on a three-degree-of-freedom model in which the scattering s (t) of brain tissue and light attenuation u (t) due to other factors are collectively introduced as the third component in addition to each hemoglobin. .

  The present inventor spectroscopically considered the process of estimating each hemoglobin concentration by NIRS using steady light, and showed that scattering and other time-varying factors can lead to artifacts in estimating each hemoglobin change. . In addition, from the actual experimental results, it has been clarified that the estimation by the conventional two-degree-of-freedom model that does not take them into consideration has a significant difference in the results depending on the selected wavelength. Based on these considerations, we proposed an extension of the estimation model to three degrees of freedom, introduced a wavelength difference method to eliminate non-wavelength dependent fluctuation factors and a Wiener filter for stabilization, and formulated a new estimation method. Turned into. Furthermore, these were applied to actual measurement examples, and were shown to be more effective than the conventional estimation method based on the two-degree-of-freedom model in stabilizing the baseline.

  The present invention relates to a light irradiation unit that irradiates a plurality of irradiation positions of a subject with near-infrared light having a plurality of wavelengths, and a light detection unit that receives and detects light transmitted through the subject or scattered or reflected in the subject. And a biological light measurement device including a data processing unit that processes data using an optical signal detected by the light detection unit, wherein the data processing unit receives the signal, processes the signal by a wavelength difference method, and performs non-wavelength processing. Baseline fluctuations that occur in the measured values by removing the signal components that are dependent on wavelength and then processing the signal from which signal components that are not dependent on wavelength are removed using a Wiener filter to remove wavelength-dependent error components and noise. This is a biological light measurement method characterized by removing the light.

  As a biological light measuring device using near infrared light used in the method of the present invention, there is a near infrared spectral brain function measuring device capable of using multi-wavelength near infrared light, for example, Shimadzu Corporation NIRStation.

Further, according to the present invention, the signal processing by the wavelength difference method creates a coefficient matrix having matrix elements of absorption coefficients of a plurality of molecular species at a plurality of wavelengths based on the detection signal detected by the light detection unit, and the signal difference This is a process for obtaining an estimated value from which a non-wavelength dependent variation component is removed by taking a value, and a process using a Wiener filter is performed with respect to the coefficient matrix when estimating an observed value using the estimated value as an input value. This biological light measurement method is characterized in that the regularization parameter is introduced and regularized using a Wiener filter to obtain an observation value as an output value.

  Furthermore, the present invention provides the biological light measurement method, wherein the regularization parameter in the Wiener filter is determined in correlation with the error component and the magnitude of noise included in the input value to be subjected to the Wiener filter processing, and the wavelength difference processing is wavelength modulation spectroscopy. The biological light measurement method described above, which is used for the biological light measurement method and the brain function measurement performed based on the photometric measurement value, and the measurement target molecular species are oxygen-binding hemoglobin and oxygen desorption hemoglobin.

  Furthermore, the present invention is a biological light measurement device that performs data processing by the above method. That is, a light irradiation unit that irradiates a plurality of irradiation positions of a subject with near-infrared light of a plurality of wavelengths, a light detection unit that receives and detects light transmitted through the subject, or scattered or reflected in the subject, and the A biological light measurement apparatus including a data processing unit that performs data processing using an optical signal detected by a light detection unit, wherein the data processing unit receives a signal, processes the signal by a wavelength difference method, and has a non-wavelength dependent signal component By removing the non-wavelength-dependent signal component and processing the signal using the Wiener filter to remove the wavelength-dependent error component and noise, it is possible to eliminate baseline fluctuations that occur in the measured value. It is the biological light measuring device characterized.

  Further, according to the present invention, the signal processing by the wavelength difference method creates a coefficient matrix having matrix elements of absorption coefficients of a plurality of molecular species at a plurality of wavelengths based on the detection signal detected by the light detection unit, and the signal difference This is a process for obtaining an estimated value from which a non-wavelength dependent variation component is removed by taking a value, and a process using a Wiener filter is performed with respect to the coefficient matrix when estimating an observed value using the estimated value as an input value. The above-mentioned biological light measurement apparatus is a process for introducing the regularization parameter of, normalizing using a Wiener filter, and obtaining an observation value as an output value.

  Furthermore, the present invention provides the above-described biological light measurement device in which the regularization parameter in the Wiener filter is determined in correlation with the error component and the magnitude of noise included in the input value to be subjected to the Wiener filter processing, and the wavelength difference processing is performed by wavelength modulation spectroscopy. The biological light measurement device described above that is performed based on photometric measurement values, and the biological light measurement device that is used for brain function measurement and whose measurement target molecular species are oxygen-binding hemoglobin and oxygen desorption hemoglobin.

  As shown in FIG. 3, in the brain function measurement using near infrared light, according to the estimation method based on the three-degree-of-freedom model of the present invention, the conventional two-degree-of-freedom model is used to stabilize the baseline. It is more effective than the estimation method based on

An outline of the biological light measurement device used in the method of the present invention is shown in FIG. The biological optical measurement device of the present invention detects a light source 10 that generates light of a predetermined wavelength, a light irradiation unit that irradiates light to an examination site of a subject, and light that is transmitted, reflected, or scattered through the examination site of the subject. A probe 12 including a light detection unit, a light measurement unit 14 for measuring a light signal detected by the light detection unit, a data processing unit 16 for data processing of light measured by the light measurement unit, and data processed by the data processing unit are displayed. The display unit 18 is included. The light source unit and the probe and the probe and the optical measurement unit are connected by a light guide such as an optical fiber. An LED (light emitting diode), LD (laser diode), or the like can be used as the light source of the light source unit. The light detection unit includes a light receiving element such as a photodiode or a phototransistor, and further includes a photomultiplier tube. Also good. The optical signal detected by the light detection unit is converted into an electric signal and sent to the data processing unit via the optical measurement unit. The data processing unit processes the received data, sends the processed data to the display unit, and the data is displayed on the display unit. As the data processing unit, a personal computer or the like can be used, a memory for recording a signal from the optical measurement unit, a central processing unit (CPU) for processing a signal from the optical measurement unit, and an arithmetic process in the central processing unit It includes a storage device such as a hard disk that stores necessary conditions and parameters, and stores calculation results. The data display unit includes a monitor and a printer that display data.

A flowchart of processing in the data processing unit is shown in FIG.
The details of the signal processing method in the data processing unit in the method of the present invention will be described below, and the effectiveness in stabilizing the baseline will be compared with the estimation method based on the conventional two-degree-of-freedom model.

Wavelength difference method The number (13) indicates the possibility that some temporal variation different from them may be added to the estimation of the amount of change x (t) of the substance group involved in brain activity. Specifically, changes in the above-mentioned factors represented by s (t) in brain tissue, spectroscopic changes in other biological tissues, contact state between optical probe and subject, bending loss of transmission fiber U (t) due to the above. There are various reports on the scattering coefficient of brain tissue and living tissue (Bevilacqua F. et al., Appl. Opt. 38 (22): 4939-4950 (1999); Yaroslavsky AN. Et al., Phys. Med. Biol., 47, 2059-2073, (2002)), most of which can be considered to have a certain finite value in the near infrared wavelength region. In addition, it has been reported that the temporal change of the scattering intensity due to the neural activity does not depend on the wavelength (Frostig RD. Et al., Proc. Natl. Acad. Sci. USA., 87, 6081-6086, (1990) ). The change in absorption coefficient of cytochrome c oxidase, which is often pointed out, is considered to be no more than 10% of the estimated value in the error evaluation with the conventional model (Uludag K. et al., J. Biomed. Opt. 7, 51 -59, (2002)).

Based on these, s _λ (t) and u _λ (t) can be treated as a third variable having little wavelength dependence, though it varies with time. However, it is assumed here that the transmission loss due to the contact state between the probe and the subject and the bending load of the fiber has no wavelength dependence. The outline of the wavelength difference method performed based on this premise is described below.

となる。ここで、異なる波長k、lでの吸光度変化量a_k(t)、a_l(t)の波長に関する差分量をδa_k,l(t)とすると、以下のように表せる。

From the assumption that s _λ (t) and u _λ (t) have no wavelength dependence, the number (8) is

It becomes. Here, when the amount of difference in the absorbances a _k (t) and a _l (t) at different wavelengths _{k and l} is Δa _{k, l} (t), it can be expressed as follows.

と表し、波長k、lでの分子種iの吸収係数の差μ_i,l-μ_i,kを行列要素とする係数行列をm
とすると、以下のように定式化される。

Since n _λ (t), which is an observation error of the change in absorbance, takes an independent value for each observation at each wavelength, it cannot be removed by wavelength difference processing and remains in the form of δn _{k, l} (t). Following the equation (10), the vertical vectors of the difference values δa _{k, l} (t) and δn (t) between N different wavelengths are

And a coefficient matrix having a matrix element of the difference μ _{i, l} -μ _{i, k} of the absorption coefficient of molecular species i at wavelengths k and l
Then, it is formulated as follows.

If m is regular, x (t) can be estimated from the above equation,

When this result is compared with the number (13), fluctuation terms such as s (t) and u (t) are removed by estimation using the wavelength difference amount δa (t), and a value closer to true can be estimated. I understand that.

Wiener filter
NIRS measurement is based on spectroscopic assumptions to estimate the number of molecules (for example, if the wavelength is close, the optical path is almost the same, and scattering and other fluctuation factors can be considered to be almost the same). The wavelength is required to be as close as possible. Under this condition, the regularity of the matrix m in the number (18) is poor, and the presence of a small δn greatly affects the estimation of x (t). In order to cope with this, the estimated value x hat w (t) was obtained using the Wiener filter K as a procedure for regularizing m.

Here, R _x and R _δn are autocorrelation matrices of x and δn (t). For simplicity, assuming that the autocorrelation matrices are σ ² I and ζ ² I, respectively, as the nature of x and the observation error n (t),

となる。このとき、Ｋは以下の形で得られる。

It becomes. At this time, K is obtained in the following form.

Assuming that oxygen-binding hemoglobin (x ₁ ) and oxygen-desorbed hemoglobin (x ₂ ) are the main molecules involved in the observation of brain activity, and their absorption coefficients are μ _{o, λ} , μ _{d, λ} , m Is given below.

Results Based on the data of the change in absorbance δa (t) used in the estimation of FIG. 4, each hemoglobin is newly obtained by the combination of the wavelength difference method and the inverse matrix (Equation 19) and the combination of the wavelength difference method and the Wiener filter (Equation 20). The amount of change was estimated, and the results were compared with those estimated by the conventional method (Equation 14). The absorption coefficient of hemoglobin is the existing reported value (Matcher SJ. Et al, Anal. Biochem. 227 (1): 54-68.
(1995)). The regularization parameter ζ ² / σ ² was set to 10 ⁻³ . FIG. 5 shows an estimation result of each hemoglobin change amount based on the measurement data of FIG. It can be seen that the conventional method is based on the two-degree-of-freedom model and thus has a long-period baseline fluctuation, whereas the two estimations based on the three-degree-of-freedom model significantly reduce this kind of fluctuation. Also, between the two estimations using the three-degree-of-freedom model, white noise-like fine fluctuations are reduced more when the Wiener filter K is used than when the inverse matrix m ⁻¹ is used. I understand that.

In order to investigate the effect of reducing fluctuations in each frequency included in the baseline, FFT analysis was performed on the estimated value of each hemoglobin change in the parietal region of an adult male while resting in the sitting position, and a power spectrum was obtained. FIG. 6 shows a comparison between the estimation using the wavelength difference method and the Wiener filter K and the estimation using the conventional method. It can be seen that in both methods, noise having frequency dependence is included in the low frequency region of about 1 Hz or less, unlike white noise. In addition to this, in the estimation of the amount of oxygen-linked hemoglobin change by the conventional method, a remarkable peak derived from the heartbeat was observed around 1.2 Hz.
On the other hand, this peak was not recognized in the estimation using the wavelength difference method and the Wiener filter K. From this, it can be seen that in the estimation based on the three-degree-of-freedom model, baseline fluctuations derived from physiological activities that were difficult to reduce without using conventional time-average processing and multiple integration can be removed.

Traditionally, NIRS measurements are known to have baseline fluctuations of various periods (Chance CE. Et al., Proc. Natl. Acad. Sci. USA, 90, 3770-3774, (1993) ), The relationship with physiological activity has been discussed (Obrig H. et al., Neouroimage 12 (6): 623-639 (2000)). According to the method described above, it has been found that the fluctuation of the baseline occurring in the time domain of several tens of seconds or more can be stabilized as seen in FIG. Moreover, it was found from the comparison of the power spectra of the estimated values at the long-term rest (FIG. 6) that the fluctuation component corresponding to the heartbeat can be removed by this method even in a shorter time region. In the conventional method, baseline fluctuations inherent in the estimation results have been reduced by using time averaging, multiple integration, drift subtraction by linear approximation, etc., but these are at the expense of time resolution and measurement time, or are appropriate for linear approximation. There was a problem in that the sex was unquestioned. This method is considered to be highly useful in that it can realize the stabilization of the baseline by using the measurement values obtained by the conventional multi-wavelength measurement apparatus as they are without using these processes.

  Multi-component substance concentration estimation using spectrophotometry can be viewed mathematically as an example of an inverse problem. The advantage of observing in the near-infrared region is, of course, its high permeability to living tissue, but on the other hand, the instability in solving the inverse problem is inevitably generated due to the spectroscopic properties of the measurement target described below. . That is, even in multi-wavelength measurement, the optical path must be regarded as almost the same, so the measurement is limited to close wavelengths within several tens of nm, but in this wavelength region, the substance shows only a slow spectral structure in the range of several tens of nm. It is. Under this condition, the rank of the coefficient matrix tends to be substantially reduced from the number of components of this system. For this reason, in estimating the amount of change of each component, the output value exhibits instability that depends on the accuracy of the coefficient matrix and the accuracy of the input value. In order to avoid this instability, the accuracy of input values was improved and the coefficient matrix was regularized as follows.

  First, the wavelength difference method was used to eliminate non-wavelength dependent fluctuation components and to stabilize the input value. Conventionally, physiological activities that have been regarded as a cause of baseline fluctuation are accompanied by changes in blood flow volume and blood flow velocity in the blood circulation. A change in the light scattering intensity on the optical path can be sufficiently assumed only by considering the variation in the number of blood cells in the tissue. In the inventor's model, a non-wavelength-dependent component was newly introduced, so that it was possible to remove temporal fluctuations caused by scattering that actually had almost no wavelength dependence, and contributed to improving the accuracy of the input value.

In addition, we introduced regularization parameter ζ ² / σ ² for coefficient matrix and tried to regularize by Wiener filter. As a result, it was possible to stably estimate the amount of change in each hemoglobin as the output value. The magnitude of the regularization parameter ζ ² / σ ² is determined in correlation with the magnitude of noise included in the input value. At that time, the degree of freedom of the output information is reduced as a price for securing the regularity of the matrix. If the accuracy of the input value is improved, the regularization parameter closer to zero, that is, closer to the true inverse matrix can be used. At this time, output information having a degree of freedom closer to the original degree of freedom of the input information can be obtained. In that sense, improving the measurement accuracy of input values is essential in brain function measurement using NIRS. By combining the wavelength difference processing method proposed by the present inventor with wavelength modulation spectrophotometry and the like, it is possible to improve the measurement accuracy of the input value.

  Bending of the neck and contact with the optical cable causes fluctuations in the absorbance value of NIRS measurement. As a result of this, it is well known that artifacts occur in the estimated values of each hemoglobin change amount in the conventional method. It has been confirmed that at least a part of this kind of artifact cannot be removed even in the estimation by this method. The reason may be that the light attenuation due to the change in the contact state between the scalp and the optical probe caused by these disturbances may be wavelength-dependent. In the above technique, the present inventor made the scattering term S (t) in the brain tissue and the light attenuation U (t) due to other factors both non-wavelength dependent in the number (15), but the wavelength in U (t) It is also possible to introduce and handle dependencies.

It is a figure which shows the optical model of each hemoglobin variation | change_quantity estimation in cerebral blood flow by near-infrared spectroscopy measurement. It is a figure which shows the outline | summary of the biological light measuring device of this invention. It is a figure which shows the flowchart of the process in the signal processing part of the biological light measuring device of this invention. It is a figure which shows the estimated value of each hemoglobin variation | change_quantity of the parietal part at the time of the tapping task by a 2 degree-of-freedom model. FIG. 4A shows an estimated value calculated based on the reported value of the molecular absorption coefficient, and FIG. 4B corrects the molecular absorption coefficient so that the standard deviation between the estimated values is minimized in consideration of the error of the light source wavelength. The estimated value calculated by 4A and 4B are estimated from changes in absorbance at black: 780 nm, 805 nm, dark ash: 805 nm, 830 nm, and light ash: 830 nm, 780 nm. The upper row shows x ₁ (oxygen-binding hemoglobin), the middle row shows x ₂ (oxygen-desorbed hemoglobin), and the lower row shows tasks. It is a figure which shows the comparison of the estimated value of the hemoglobin variation | change_quantity by each estimation method. The upper row is estimated by the wavelength difference method and the inverse matrix m ^-1 , the middle row is estimated by the wavelength difference method and the Wiener filter K, the lower row is estimated by the pseudo inverse matrix (conventional method) M ⁺ , and each result is normalized by the standard deviation. did. The black line represents oxygen-binding hemoglobin, and the ash line represents oxygen-desorbed hemoglobin. It is a figure which shows the comparison of the power spectrum of the estimated value of each hemoglobin variation | change_quantity of the parietal part at the time of the adult male rest by each estimation method. FIG. 6A shows oxygen-binding hemoglobin, and FIG. 6B shows oxygen-desorbed hemoglobin. In both FIGS. 6A and B, black dots indicate estimation by the wavelength difference method and the Wiener filter K (three-degree-of-freedom model), and gray points indicate estimation by a pseudo inverse matrix (two-degree-of-freedom model) M ⁺ .

Explanation of symbols

10 light source part 12 probe (light irradiation part and light detection part)
14 Optical measurement part 16 Data processing part 18 Display part

Claims (10)

  A light irradiating unit that irradiates a plurality of irradiation positions of near-infrared light of a plurality of wavelengths, a light detecting unit that receives and detects light transmitted through the subject, or scattered or reflected within the subject, and the light detection In a biological optical measurement method for processing an optical signal detected using a biological optical measurement device provided with a data processing unit that performs data processing using the optical signal detected by the unit, the signal received by the data processing unit is subjected to a wavelength difference method Measured values by removing non-wavelength-dependent signal components and processing the signal from which non-wavelength-dependent signal components have been removed using a Wiener filter to remove wavelength-dependent error components and noise. A biological light measurement method characterized by removing baseline fluctuations that occur in   The signal processing by the wavelength difference method is not performed by creating a coefficient matrix having matrix elements of absorption coefficients of a plurality of molecular species at a plurality of wavelengths based on the detection signal detected by the light detection unit, and taking the difference value of the signal. A process for obtaining an estimated value from which a wavelength-dependent variation component has been removed. A process using a Wiener filter introduces a predetermined regularization parameter for the coefficient matrix when estimating an observed value using the estimated value as an input value. The living body light measurement method according to claim 1, wherein the biological light measurement method is a process of regularizing using a Wiener filter to obtain an observation value as an output value.   The biological light measurement method according to claim 2, wherein the regularization parameter in the Wiener filter is determined in correlation with an error component and a noise level included in an input value to be subjected to the Wiener filter process.   The biological light measurement method according to claim 1, wherein the wavelength difference processing is performed based on a wavelength modulation spectrophotometric measurement value.   The biological light measurement method according to any one of claims 1 to 4, which is used for brain function measurement, and the molecular species to be measured are oxygen-binding hemoglobin and oxygen-desorbed hemoglobin.   A light irradiating unit that irradiates a plurality of irradiation positions of near-infrared light of a plurality of wavelengths, a light detecting unit that receives and detects light transmitted through the subject, or scattered or reflected within the subject, and the light detection A biological optical measurement device including a data processing unit that processes data using an optical signal detected by the unit, wherein the data processing unit receives the signal and processes it by a wavelength difference method to remove non-wavelength dependent signal components Furthermore, it is characterized in that baseline fluctuations that occur in measured values are eliminated by processing the signal from which non-wavelength dependent signal components have been removed using a Wiener filter to remove wavelength dependent error components and noise. A living body light measuring device.   The signal processing by the wavelength difference method is not performed by creating a coefficient matrix having matrix elements of absorption coefficients of a plurality of molecular species at a plurality of wavelengths based on the detection signal detected by the light detection unit, and taking the difference value of the signal. A process for obtaining an estimated value from which a wavelength-dependent variation component has been removed. A process using a Wiener filter introduces a predetermined regularization parameter for the coefficient matrix when estimating an observed value using the estimated value as an input value. The biological light measurement device according to claim 6, wherein the biological light measurement device is a process of regularizing using a Wiener filter to obtain an observation value as an output value.   The biological light measurement apparatus according to claim 7, wherein the regularization parameter in the Wiener filter is determined in correlation with an error component and a noise level included in an input value to be subjected to the Wiener filter process.   The biological light measurement device according to any one of claims 6 to 8, wherein the wavelength difference processing is performed based on a wavelength modulation spectrophotometric measurement value.   The biological optical measurement device according to any one of claims 6 to 9, which is used for brain function measurement, and the measurement target molecular species are oxygen-binding hemoglobin and oxygen-desorbed hemoglobin.

Images