WO2007144977A1

WO2007144977A1 - Apparatus for measuring biological light

Info

Publication number: WO2007144977A1
Application number: PCT/JP2006/325814
Authority: WO
Inventors: Naoki Tanaka; Masashi Kiguchi; Atsushi Maki; Shingo Kawasaki; Noriyoshi Ichikawa; Fumio Kawaguchi
Original assignee: Hitachi Medical Corporation
Priority date: 2006-06-15
Filing date: 2006-12-25
Publication date: 2007-12-21
Also published as: JPWO2007144977A1

Abstract

In automatically judging diseases of subjects from the characteristic wave patterns of changes in the hemoglobin level that are determined by measuring biological light, the separation of a group of healthy subjects from a group of unhealthy subjects and the separation of individual disease groups are improved. In an apparatus (10) for measuring biological light, an analysis section, wherein the wave pattern is analyzed, is provided with a judgment section (40) by which diseases are judged based on classification models. For conducting hierarchical classification, multiple classification models are provided in the judgment section. As the classification models, parameters obtained by synthesizing characteristic amounts of multiple kinds are employed. These classification models are optimized by the optimization method and automatically adjusted in association with changes in the data.

Description

Specification

Biological light measurement device

Technical field

[0001] The present invention relates to a biological optical measurement device that supports disease diagnosis non-invasively.

Background art

[0002] A biological light measurement device is a device that irradiates a living body with near-infrared light and measures light that has passed through the living body or reflected within the living body, and has blood circulation, hemodynamics, and hemoglobin content inside the living body. Since changes can be easily measured with low restraint and no harm to the subject, clinical application is expected.

[0003] As clinical applications of biological light measurement, diagnosis of epilepsy, cerebral ischemia, and research of language function have been reported. Non-patent documents 1 and 2 below report that abnormalities occur in the pattern of hemoglobin changes in the frontal lobe by biophotometry in patients with mental illness such as depression and schizophrenia (schizophrenia). . Specifically, it has been reported that the integration values during the trial of the hemoglobin time waveform have different characteristics of large, small, and medium when comparing healthy subjects, depressed patients, and schizophrenic patients. In schizophrenia, it has been reported that a re-elevation change of hemoglobin is observed after completion of the task.

[0004] When estimating a subject's disease with such waveform characteristics, it is difficult to ensure objectivity if the judgment is left to the examiner. Therefore, as a method of automation that emphasizes objectivity, the present applicant determines a reference waveform for each disease group and uses the correlation between the waveform of each subject and the reference waveform of each disease group as a reference. A technique for determining a disease is proposed (for example, Patent Document 1). In addition, a method for calculating and displaying a disease determination score by calculating the Mahalanobis distance between the measured feature value of the hemoglobin change waveform force and the disease-specific dictionary data is proposed (Patent Document 2). In the present invention, as a criterion for determination, the dictionary data for each disease is classified and categorized based on the disease name added by the user.

Non-Patent Document 1: “Dynamics of frontal lobe regional cerebral blood volume in neuropsychiatric disorders `` Examination using Graph '' Masato Fukuda, JSPS Grant-in-Aid for Scientific Research 2001-2002

Non-Patent Document 2: "The Mind Seen by Light" Masato Fukuda et al., Mind and Society No. 34 卷 1 separate volume, Japan Mental Health

Patent Document 1: Japanese Patent Laid-Open No. 2003-275191

Patent Document 2: International Publication WO2005Z025421

Disclosure of the invention

Problems to be solved by the invention

[0005] However, in these technologies, there is an aspect that the separation of the healthy group and the non-healthy group is not sufficient, and in the separation of the disease group, a simple categorization based on the waveform feature is made. ! / There was a cunning habit.

Therefore, the present invention provides a biological optical measurement device that improves separation of healthy groups and non-healthy groups, enables simple category generation in disease classification, and can perform effective disease classification. The purpose is to provide.

Means for solving the problem

[0006] In order to solve the above problems, the present invention provides a determination unit that determines a disease by a classification model in an analysis unit that performs waveform analysis of a biological light measurement device, and performs hierarchical classification in the determination unit. Several classification models were provided. In the present invention, a determination unit for determining a disease using a classification model is provided, and a variable obtained by synthesizing a plurality of types of feature amounts is used as a classification model used by the determination unit.

That is, the biological light measurement device of the present invention irradiates a subject with light having a wavelength belonging to the visible to infrared region, detects light that has passed through the subject, and measures a hemoglobin change waveform. In addition, a feature amount calculation unit that extracts and analyzes a plurality of types of feature amounts from the measured waveform, and a classification model that is set in advance using the plurality of types of feature amounts extracted by the feature amount calculation unit. 1. It comprises a determination unit that performs disease determination and a display unit that displays a determination result of the determination unit, and the determination unit includes a plurality of classification models having a hierarchical structure as the classification model. To do.

[0008] Further, in the biological light measurement device of the present invention, the determination unit includes a plurality of types as the classification model. It is characterized by using a variable that combines feature quantities.

In the biological optical measurement device of the present invention, the classification model is, for example, a variable obtained by combining a plurality of types of feature amounts, for example, a linear combination of a plurality of types of feature amounts, and a plurality of subject groups for which disease determination is confirmed. Each is determined to maximize the probability of being classified into the class corresponding to the disease.

[0009] In addition, the biological optical measurement device of the present invention can be used to construct a classification model, for example, a slope value immediately after the start of a task of a hemoglobin waveform, an integral value during task execution, and a re-rise degree after the task ends. The center of gravity value of the entire waveform is used as the feature amount. The feature amount calculation unit extracts a feature amount from a hemoglobin waveform obtained in each of the multichannels of the biological light measurement unit.

Furthermore, the present invention provides a disease diagnosis support apparatus in which components other than the biological light measurement unit are separated from the force of the biological light measurement unit. This disease diagnosis support device is the same as the biological light measurement device of the present invention except that the biological light measurement unit is independent.

The invention's effect

[0010] According to the present invention, the effect of separating a healthy person group from a non-healthy person group (disease patient group) can be enhanced by performing disease classification hierarchically. In addition, even if the classification of disease patient groups is adopted, it is possible to classify diseases effectively by creating a simple category by incorporating hierarchical properties.

BEST MODE FOR CARRYING OUT THE INVENTION

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an outline of the biological light measurement device 100 of the present invention. The biological light measurement device 100 extracts features from the biological light measurement unit 10 and the hemoglobin change waveform measured by the biological light measurement unit 10 and performs various analyzes based on the input of the operator. The feature amount calculation unit 20, the input unit 30 for inputting information necessary for calculation in the feature amount calculation unit 20, and the plurality of types of feature amounts extracted by the feature amount calculation unit 20 A patient is classified into five types, and a decision unit 40 that supports disease diagnosis and analysis results of biological light measurement data obtained from a large number of subjects are stored as disease dictionary data together with definitive diagnosis results and The storage unit 50 stores the waveform data measured by the optical measurement unit 10 and the classification model used by the determination unit 40, and the determination results of the determination unit 40 are registered as a dictionary. And a display unit 60 for displaying information together with the information. Hereinafter, each element will be described.

[0012] The biological light measurement unit 10 irradiates a human head with light, receives light reflected and scattered in the vicinity of the head surface at a plurality of detection positions, and changes signals in the blood (here, hemoglobin). This is a multi-channel device that measures the amount of hemoglobin change for each channel when a subject is exposed to a fluency task with language recall. As shown in FIG. 2, the specific configuration mainly includes a light source unit 102, a light receiver 108, a lock-in amplifier 109, an A / D converter 110, a measurement control computer 111, and the like. The letters at the end of these codes mean that there are multiple elements. The light source unit 102 and the subject 106 and the light receiver 108 and the subject 106 are connected by an irradiation optical fiber 105 and a light receiving optical fiber 107, respectively.

[0013] The light source unit 102 includes two light source units 102a and 102b having a first wavelength (for example, 690 nm or 780 nm) and light source units 102c and 102d having a second wavelength (for example, 830 nm). Connected, modulated differently depending on the measurement position, and irradiated to the head of the subject 106 via the couplers 104a and 104b and the irradiation optical fibers 105a and 105b. The light reflected / scattered near the head is collected by receiving optical fibers 107a to 107fC arranged close to the irradiation optical fiber, and photoelectrically converted by the optical receivers 108a to 108fC, respectively. The distal ends of the irradiation optical fiber 105 and the light receiving optical fiber 107 are arranged so that the light receiving positions are arranged at the same distance (for example, 30 mm) from the light irradiation position force. As the light receiver 108, a photoelectric conversion element represented by a photomultiplier tube or a photodiode is used.

Electric signals representing the intensity of light passing through the living body photoelectrically converted by the light receiver 108 are respectively input to the lock-in amplifier 109. The intensity modulation frequency from the oscillator 101 is input to the lock-in amplifier 109 as a reference frequency. The light intensity signals from the light sources 102a and 102b and 102c and 102d are separated by the reference frequency, respectively. Is output.

[0015] The separated transmitted light intensity signal of each wavelength, which is the output of the lock-in amplifier 109, is analog-digital converted by the analog-digital converter 110, and then sent to the measurement control computer 111. The measurement control computer 111 calculates the detection signal power at each detection point, the relative change in oxygen concentration to deoxygenated hemoglobin concentration, deoxygenated hemoglobin concentration, and total hemoglobin concentration. Store. Oxygenated hemoglobin concentration One or a plurality of hemoglobin change signals, which are relative changes in the deoxygenated hemoglobin concentration and the total hemoglobin concentration, are used depending on the disease to be determined. In addition to the above calculation, the measurement control computer 112 controls the operation of the biological light measurement unit 20. Note that FIG. 2 shows a configuration of a biological light measurement unit that separates a plurality of lights by a modulation method. However, the present invention is not limited to this. It is also possible to use a time-sharing method for discriminating between the two.

In the example shown in FIG. 2, the elements other than the biological light measurement unit 10, that is, the feature amount calculation unit 20, the determination unit 40, and the storage unit 50 are all in the calculator 112, and the input unit 30 and the display unit 60 are It is provided in computer 112. However, the computer 111 and the computer 112 may be a single computer. Further, it may be placed at a completely distant place and connected via a communication means such as the Internet. That is, the computer 112 can also be a disease diagnosis support device that is independent of the biological light measurement unit.

[0017] The feature amount calculation unit 20 analyzes the feature amount based on the waveform data of the measured changes in local oxygenated hemoglobin, deoxygenated hemoglobin, and total hemoglobin. The function of the feature quantity calculation unit 20 will be described later. Waveform data and features are passed to the storage unit 50

[0018] The storage unit 50 temporarily stores the measurement information of the subject and enables subsequent processing. On the other hand, for example, when there is a definitive diagnosis, the measurement information is stored as dictionary data. The dictionary data is used not only when optimizing the classification model (automatic parameter adjustment) used by the determination unit 40 as will be described later, but also when diagnosing with this device.

The determination unit 40 uses the feature quantities and classification models stored in the storage unit 50 to determine the subject's disease, specifically, whether a person is healthy or unhealthy for mental illness, Judgment is made regarding which disease the determined person belongs to. Details will be described later.

[0019] The input unit 30 displays a GUI for the user to input data and commands for control and calculation performed by the computer 112, and enables communication between the user and the computer 112 using the GUI.

The display unit 60 stores the feature amount of the subject extracted by the feature amount calculation unit 20 in the storage unit 50. Database information, the result of classification by the determination unit ₄₀ , and the like are displayed.

Next, the function of the feature amount calculation unit 20 will be described.

The hemoglobin change signal to be processed by the feature calculation unit 20 is, for example, as shown in FIG. 3, a change in signal intensity in a predetermined time consisting of a waiting time before starting a task and a pause time during and after the task. This is obtained as a waveform 300 indicating (mMmm). The two vertical lines shown in the figure represent the start time 301 and the end time 302 of the task, respectively. Usually, tasks are repeated multiple times with a set of load and pause. The hemoglobin waveforms obtained from multiple measurements are averaged and subjected to preprocessing such as smoothing and baseline processing as necessary. Of course, there are cases where the amount of hemoglobin change is determined in a single trial without repeating multiple times. FIG. 3 shows one hemoglobin change waveform, which is obtained for each channel of the biological light measurement unit 10.

[0021] It has been reported that such a hemoglobin waveform has different characteristics for each psychiatric disorder (Non-Patent Documents 1 and 2). Figure 4 shows the waveform feature patterns for each of the load diseases for subjects with a language fluency task involving language recall. As shown in the figure, in healthy subjects, the hemoglobin change is large and monotonously decreases after the task ends.In schizophrenia, the hemoglobin change is moderate and the waveform re-rises after the task ends. A small feature is that bipolar disorder has large hemoglobin changes and peaks in the second half of the task.

[0022] The feature amount calculation unit 20 digitizes these features from the measured hemoglobin waveform 300. Specifically, the slope value 311, the integral value 313, the re-elevation degree 315, and the center of gravity 317 are calculated as feature quantities. Slope 311 is the slope of the hemoglobin waveform calculated from the task start time to 5 seconds. The slope 311 represents the speed of response to the task. The integral value 313 is obtained by integrating the hemoglobin waveform in the section during the task trial. The integrated value 313 is considered to reflect the magnitude of the reaction. The re-elevation degree 315 is obtained by connecting the hemoglobin value at the end of the task and the hemoglobin value at the end of the measurement with a straight line, and calculating the area above the straight line. The rise of 315 seems to reflect a mental tendency not to follow the task order. The centroid value 317 represents the relative time at which the centroid of the waveform is located. The measurement start is zero and the measurement end is one. The center of gravity value 317 is considered to represent the speed of sustained response. It is done. The biological light measuring unit 10 can obtain a hemoglobin waveform and a feature quantity of a plurality of channels in one measurement. Depending on the type of feature quantity, the average waveform value of the measurement region or the maximum value of all channels is used. In the present embodiment, the maximum value of all channels is used for the degree of re-elevation, and the average waveform value in the measurement region is used otherwise.

[0023] The feature quantity calculation unit 20 further performs a calculation according to the classification model of the determination unit 40 using these feature quantities. The calculation result is used by the determination unit 40 to determine which disease the subject belongs to by applying the classification model.

Hereinafter, the classification model of the determination unit 40 will be described in detail.

In the biological optical measurement device of the present embodiment, the measured waveforms are classified into five types by hierarchical classification. Responses to subjects of healthy subjects and patients with diseases are very complex, and it is difficult to classify them in one step even by combining the four types of features described above. It becomes possible to classify into groups. Figure 5 shows the hierarchical classification adopted in this embodiment. In this embodiment, there are three layers, the first layer 501 classifies the Typel group and the non-Typel group, and the second layer 502 classifies the non-Typel group into three groups Type2 / Type3, Type4, and Type5. In the third hierarchy 503, Type2 / Type3 group is classified into Type2 group and Type3 group. The Typel group consists mainly of healthy individuals, the Type2 and Type5 groups mainly consist of schizophrenic patients, the Type3 group consists of bipolar disorder patients, and the Type4 group consists mainly of depressed patients.

Next, a classification model used in each hierarchy will be described. FIGS. 6 to 8 are diagrams showing classification models of the first hierarchy, the second hierarchy, and the third hierarchy and the threshold value thr for determination, respectively.

[0026] The classification model of the first layer that classifies into the Typel group and the non-Typel group consists of four types of feature values, slope value (D), integral value (I), re-rise (R), and center of gravity (C). It is a variable X_l that is normalized and combined by combining them linearly.

X_1 = C 1 * 1 + C2 * D + C3 * R + C4 * C (1)

Underline in Equation 1 indicates normalization. When the value of the variable X_l is greater than or equal to thr_l, the waveform belongs to the Typel group.

[0027] The second hierarchy for classifying the group classified as a non-Typel group in the first hierarchy is further divided into two hierarchies of classification using slope values and classification using integral values. First, when the slope value is smaller than thr_a, Typ It is determined as e5. Next, when the slope force is greater than ¾hr_a and the integral value is greater than or equal to thr_b, it belongs to the Type2 / Type3 group. .

[0028] The classification model of the third layer is a classification model for classifying Type2 / Type3 group into Type2 group and Type3 group. It is a variable X_23 that is synthesized by combining the above four types of feature values, slope value (D), integral value (1), re-rise (R), and center of gravity (C), and combining them linearly.

X_23 = D 1 * 1 + D2 * D + D3 * R + D4 * C (2)

Underline in Equation 2 indicates normalization. When the value of variable X_23 is greater than or equal to thr_23, the waveform is determined to belong to Type 2 group, and otherwise, it belongs to Type 3 group.

[0029] These classification models in the first to third layers are combinations of features that are considered to represent the features of each disease in a straightforward manner, and the threshold thr and linear combination coefficients C1 to C4, D1 ˜D4 is obtained by optimizing the model using dictionary data. Hereinafter, optimization of each classification model will be described.

[0030] First, the classification model of the first layer is optimized by considering a unit vector c facing a certain direction in a four-dimensional space with a normalized feature amount, and with the vector χθ corresponding to one data. Take the product and let it be X. The minimum and maximum values of X are determined for dictionary data, and an appropriate value thr is taken and classification is performed. In this case, when the probability that a healthy person is classified as Typel is pi and the probability that a non-healthy person is not classified as Type 1 is p_l, a vector that maximizes the evaluation function f (u) = u * pl + (lu) * p_l Find c and the value thr. Let the optimized c elements (cl, c2, c3, c4) and the value thr be Cl, C2, C3, C4, and thr_l in equation (1).

Next, the optimization of the classification model in the second hierarchy uses, for example, an automatic clustering technique. There are 3 types for the threshold combination j (TYPE (j, n) n = l, 2,3; corresponding to the above Type2 / Type3, Type4, Type5), and for each type, healthy, schizophrenia, Presence probability pNC (j, n), pSO.n), pD (j, n), pBP (j, n) is determined for each disease group of depression and bipolar disorder. Between these probabilities

pNCO.n) + pSO.n) + pD (j, n) + ρΒΡθ, η) = 1

The relationship holds. The threshold value is selected so that the existence probability of each disease group in each type is as biased as possible. That is, the entropy sum E corresponding to the combination j of threshold values (thr_a, thr_b) (j) is

[Number 1]

E (J) = ∑P J, ")

E (J, «) =-∑ poc (j, n) ldg ₂ p (j, n)

= NC, S, D, BP Here, pn is the ratio of data included in type n to threshold combination j. Threshold combination force that minimizes the entropy sum E (j). Gives the best classification (ie clustering). Minimizing entropy corresponds to selecting thresholds so that the probability of existence of each disease group in each type is as biased as possible.

[0032] Finally, optimization of the classification model of the third hierarchy will be described. Consider a unit vector d pointing in a certain direction in a four-dimensional space with normalized features, and let X be the inner product of the vector χθ corresponding to one piece of data. The minimum and maximum values of X are determined for the dictionary data, and an appropriate value thr is taken for classification. At this time when the p_2 the probability of classifying a bipolar disorder patients belonging the probability of classifying the schizophrenic patients belonging to the Type2 / Type3 group to Type2 to p2, Type2 / Type3 group to Typ e3, the evaluation function g _(W) = Find the vectors d and thr that maximize _W * p2 + (lw) * p_2. Let the optimized elements of d (dl, d2, d3, d4) and thr be D1, D2, D3, D4, thr-23 in equation (2).

[0033] The determination unit 40 performs disease determination using the classification model described above. Figure 9 shows the disease determination algorithm. As shown in the figure, when the feature amount (in this case, the slope value, the integral value, the re-rise degree, and the center of gravity) calculated by the feature calculation unit 20 in the hemoglobin waveform force is input, the determination unit 40 receives the first layer 901. Determine whether the value X_l in equation (1) is larger than thr_, and classify it as either Typel or Non-Typel. If it is classified as non-Typel, it is further classified as Type2 / Type3, Type4, or Type5 in the second layer 902, 903. If it is classified as Type2 / Type3, it is further classified as either Type 4 or Type 5 in the third layer 904. For example, a subject who has a hemoglobin change with an integration value of 102, a slope of 0.003, a re-elevation degree of 5, and a center of gravity of 0.15 is determined to belong to a healthy person (Typel).

[0034] According to this hierarchical classification, Typel contains the most healthy individuals, Type2 contains schizophrenia, Type3 contains bipolar disorder, Type4 contains depression, and Type5 contains schizophrenia. Will be. However, since these only indicate a high probability, the judgment result is expressed as a high probability. As a display method, for example, it is possible to display findings such as “high possibility of schizophrenia” and “possibility of depression”, but display them on the scatter chart of the classification model. You can also.

FIG. 10 shows an example of a scatter diagram. Figure 10 is a scatter plot of the classification model in the second layer, and the Type area divided by the thr value of the classification model is shown on the graph with the horizontal axis tilted and the vertical axis the integrated value. Each data in the dictionary data is indicated by a symbol. By plotting the value of the subject on such a scatter diagram, the type to which the subject belongs can be confirmed at a glance, and the distance from other types can also be confirmed. When the first layer is classified as belonging to the non-Type 1 group, subjects who have hemoglobin changes with the same integral value 102 and slope 0.003 as in the previous case are distributed in the upper right area on the scatter diagram. It can be recognized that it belongs to Type2 / Type3 group.

FIG. 11 shows a procedure from the feature amount extraction described above to the disease determination. An input screen as shown in FIG. 12 is displayed on the input unit 30 to make a disease determination. Accordingly, information necessary for disease determination, examination date, subject number, etc. are entered (step 1101). In this example, the name of the force identifying the subject by number may be used. If there is a confirmed diagnosis result, check the mouth of item 1. When input, the measurement waveform of the corresponding subject is read from the biological light measurement unit 10, and a plurality of types of feature quantities are calculated (steps 1102 and 1103). Next, hierarchical classification is performed based on the calculated feature amount (step 1104). The judgment result is displayed on a scatter diagram as shown in FIG. 10 (step 1105). When there is a definitive diagnosis of the subject, the test result is automatically stored in the database (storage unit 50).

[0037] As described above, according to the biological optical measurement device of the present embodiment, a mental disorder having a complex aspect of waveform characteristics for each disease by performing hierarchical classification using variables that combine feature quantities. Can be automatically discriminated from a healthy person and between disease groups, and the accuracy of discrimination can be increased. In particular, it is possible to grasp the overall trend by comparing and comparing the feature values of the subject (or variables that combine them) and the feature values of the dictionary data (or variables that combine them) on a scatter diagram. However, it becomes possible to determine the disease category.

[0038] By the way, the classification model used by the determination unit 40 is constructed from dictionary data by an optimization method. However, the accuracy of classification improves as the number of dictionary data increases. Therefore, it is preferable that the living body optical measurement device of the present invention has a function of updating (automatically adjusting) the classification model!

The automatic adjustment function of the classification model will be described below.

As already mentioned, if there is a definitive diagnosis result of the subject, the test result is automatically stored in the database. The dictionary data changes by such a dictionary data storage function. The automatic adjustment function is a mechanism that optimizes the classification model parameters again as the dictionary data changes. The optimization method is the same as the optimization method when the classification model is constructed. For the first and third layer classification models, the evaluation function is set for the set solid c and value thr. Set the maximum vector c and value thr, and use the optimized c element and value thr as the coefficients and thresholds in equation (1) or (2). For the classification model of the second layer, the automatic clustering technology is used and the threshold value is selected so that the existence probability of each disease group is as biased as possible for each type.

In the optimization of the classification model, a scatter diagram can be used. In this case, the thr value calculated by optimization is displayed on a graph plotting each data in the dictionary data. Fig. 13 shows an example of the display screen showing the classification model. In the example shown in Fig. 13, along with a scatter plot of each classification model, a bar graph shows what diseases are present in each disease category classified by the calculated threshold (disease abundance ratio)! . Such a display makes it possible to immediately confirm the classification result (classification accuracy).

Example

[0041] For 107 cases with confirmed diagnosis results, disease determination was performed using the hierarchical classification of the biological optical measurement device of the present invention! The accuracy of disease determination was evaluated.

[0042] <First level>

As a classification model for the first layer, a variable X_l synthesized by linearly combining four types of normalized features, slope value (D), integral value (I), re-rise (R), and centroid (C) Use

X_1 = C 1 * 1 + C2 * D + C3 * R + C4 * C (1)

The coefficients in Equation (1) were Cl = 0.31, C2 = 0.15, C3 = -0.66, C4 = -0.66, and thr_l = 0.482. As a result, 35 cases out of 107 cases were determined as typel. 64% of healthy people were judged as Typel On the other hand, the unhealthy person was determined to be typel with 6% strength.

As a comparative example, C4 = 0 was fixed, and determination was performed with Cl = 0.56, C2 = 0.42, C3 = −0.71, and thr_l = 0.191. In this case, the correct answer rate was the highest, but 46% of healthy subjects and 31% of non-healthy subjects were judged as typel. Furthermore, when C3 = 0 was also fixed, 21% of healthy subjects and 48% of unhealthy subjects were determined to be typel. From this, it can be seen that the center of gravity (C) and the re-elevation degree (R) are important features for the separation of Typel and non-Typel groups.

[0044] Second level>

For the 72 cases classified into the non-Typel group, a judgment was made using slope and integral as the second-layer classification. Good results were obtained when the slope threshold thr_a = -0.0012 and the integral threshold thr_b = 12. The scatter diagram shown in Fig. 10 is the result of this example. In Fig. 10, it can be seen that the four groups are distributed so as to overlap each other rather than being clearly separated. Still, it is distributed in Type 5, which is a small area of schizophrenia, but with a small slope. The remaining patients with schizophrenia are distributed with the majority of patients with bipolar disorder in the type 2 and type 3 regions where the slope and integral are relatively large. Most patients with depression have a relatively large slope and a small integrated value, and are distributed in the region Type4. Since the number of healthy individuals is large, a considerable number is distributed among non-Typel, especially Type2 / Type3.

[0045] <Third layer>

In Example 2, it was determined that Type 2 / Type 3 were 37 cases, and the four types of feature values, slope value (D), integral value (I), and re-elevation degree (R) were further normalized. Judgment was performed using variable X_23, which was synthesized by linearly combining the center of gravity (C).

X_23 = D 1 * 1 + D2 * D + D3 * R + D4 * C (2)

When Dl = 0.15, D2 = 0.15, D3 = 0.98, D4 = 0.0, and thr_l = 0.129, 17 out of 37 cases were judged as type2. When each type was identified with the main component disease label, the correct answer rate for each disease group was finally over 64%. When D3 = 0 is fixed, Dl = 0.55, D2 = 0.76, D3 = 0.0, D4 = 0.34, thr_l = 0.349, the average correct answer rate is maximum, but each disease group classification correct answer rate is only 39% I stayed at. From this, it can be seen that the degree of re-elevation is an important feature for the classification of Type 2 and Type 3 groups. The bar graph shown in FIG. 13 is a diagram showing the determination result of the third hierarchy according to the present example. In this figure, Typel is healthy, Type2 is schizophrenia, Type3 is Bipolar disorder, Type 4 has depression, Type 5 has many schizophrenia. Type2 and Type3 showed a tendency to separate schizophrenia and bipolar disorder, respectively.

As described above, the embodiments of the present invention have been described. The main feature of the present invention is that the analysis is performed in advance using a plurality of types of feature amounts in the analysis of the waveform measured by the biological light measurement device. This is that a determination function for determining a disease according to a similar model is provided, and various modifications and applications can be made within the scope described in the claims, not limited to the above embodiment. For example, the present invention also includes that the hierarchical classification model automates only the classification of the first hierarchy using a composite variable of a plurality of feature amounts, and allows the user to make subsequent determinations. In addition to the display using a scatter diagram, the display method of the determination result can be appropriately combined with display using characters, display using a dull, and the like.

The present invention relates to a disease determination function using an optical measurement result, and can be used for a disease diagnosis support apparatus such as a mental illness.

Brief Description of Drawings

FIG. 1 is a diagram showing a schematic configuration of a biological light measurement device according to the present invention.

FIG. 2 is a diagram showing a configuration of a biological light measurement unit.

FIG. 3 is a diagram for explaining extraction of feature values of hemoglobin change waveform force.

FIG. 4 is a diagram showing waveform characteristics for each disease.

FIG. 5 is a diagram schematically showing a hierarchical classification in a determination unit.

FIG. 6 is a diagram showing a classification model in the first layer.

FIG. ⁷ is a diagram showing a classification model of the second hierarchy.

FIG. 8 is a diagram showing a classification model of the third hierarchy.

FIG. 9 is a diagram showing a disease determination algorithm.

FIG. 10 is a diagram showing a display example of determination results.

FIG. 11 is a diagram showing a disease state determination flow.

FIG. 12 is a diagram showing an example of an input screen of the input unit.

FIG. 13 is a diagram showing the result of optimizing the classification model.

Explanation of symbols

[0049] 10 ··· Biometric light measurement unit, 20 ··· Feature amount calculation unit, 30 ··· Input unit, 40 ··· Determination unit, 50 ··· Memory, 60 ··· Display, 102… Light source, 108 ··· Receiver, 106 ··· Inspection object (subject), 111 ·

• Measurement control computer, 112 ··· computer (disease diagnosis support device).

Claims

The scope of the claims

[1] Visible force A biological light measurement unit that irradiates a subject with light having a wavelength belonging to the infrared region, detects the light passing through the subject, and measures the hemoglobin change waveform, and the measured waveform force A feature amount calculation unit that extracts and analyzes the feature amount, a determination unit that performs disease determination according to a preset classification model using a plurality of types of feature amounts extracted by the feature amount calculation unit, And a display unit that displays a determination result of the determination unit, wherein the determination unit includes a plurality of classification models having a hierarchical structure as the classification model.

[2] In the biological light measurement device according to claim 1,!

A biological light measurement apparatus, wherein a variable obtained by combining the plurality of types of feature quantities is used for at least one of the plurality of classification models.

[3] In the biological optical measurement device according to claim 1,!

The biological light measurement apparatus according to claim 1, wherein the uppermost classification model among the plurality of classification models having a hierarchical structure is a classification model for classifying a subject into either a normal group or a non-normal group.

[4] In the biological light measurement device according to claim 1,!

Among the plurality of classification models having the hierarchical structure, the uppermost classification model is a classification model for classifying the subject into either a healthy group or a non-healthy person group, and the lower classification model is a non-healthy person. A biological light measurement apparatus, which is a classification model for finally classifying subjects classified into groups into one of a plurality of disease groups.

[5] Visible force A biological light measurement unit that irradiates a subject with light having a wavelength belonging to the infrared region, detects light passing through the subject, and measures a hemoglobin change waveform, and a plurality of measured waveform forces A feature amount calculation unit that extracts and analyzes the feature amount, a determination unit that performs disease determination according to a preset classification model using a plurality of types of feature amounts extracted by the feature amount calculation unit, and And a display unit that displays a determination result of the determination unit, wherein the determination unit uses a variable obtained by combining a plurality of types of feature amounts as the classification model.

[6] In the biological light measurement device according to claim 2, The biometric optical measurement apparatus, wherein the synthesized variable is a linear combination of the plurality of types of feature quantities.

[7] In the biological light measurement device according to claim 1,!

The biometric measurement is characterized in that the classification model is determined so as to maximize the probability that each of the plurality of subject groups for which disease determination is confirmed is classified into a class corresponding to the disease. apparatus.

[8] The biological light measurement device according to claim 1,

The feature value used by the determination unit is one or more feature values selected from the slope value of the hemoglobin time waveform immediately after the start of the task, the integral value during the task execution, the re-rise degree after the task ends, and the centroid value of the entire waveform. A biological light measuring device comprising:

[9] The biological light measurement device according to claim 1,

The biological light measurement unit has a multi-channel structure that measures a plurality of waveforms from a plurality of measurement positions of the subject, and the feature amount calculation unit extracts a feature amount from waveforms obtained from each of the multi-channels. A biological light measuring device characterized in that the maximum value is an analysis target.

[10] The biological light measurement device according to claim 1,

The determination unit is a method for optimizing the classification model based on the waveform measured with the disease confirmed subject and Z or a plurality of feature amounts extracted from the waveform. A biological light measuring device comprising:

[11] The biological light measurement device according to claim 5,

The biometric measurement is characterized in that the classification model is determined so as to maximize the probability that each of a plurality of subject groups for which disease determination is confirmed is classified into a class corresponding to the disease. apparatus.

[12] The biological light measurement device according to claim 5,

[13] The biological light measurement device according to claim 5, The biological light measurement unit has a multi-channel structure that measures a plurality of waveforms from a plurality of measurement positions of the subject, and the feature amount calculation unit extracts a feature amount from waveforms obtained from each of the multi-channels. A biological light measuring device characterized in that the maximum value is an analysis target.

The biological light measurement device according to claim 5,

The determination unit is a method for optimizing the classification model based on the measured waveform and Z or a plurality of types of feature values extracted from the waveform for a subject whose disease has been confirmed. A biological light measuring device comprising: