Optical interference signal generation method, optical interference signal generation device, learned machine model, machine model learning method, machine model learning device, eye tissue evaluation method, eye tissue evaluation device, and program

Abstract

An optical interference signal generation method (S3) comprises: a property value setting step (S31) for setting a property value of a pseudo-scatterer; a signal generation step (S32) for generating a pseudo-optical interference signal from a pseudo-scatterer which has a set density; a noise generation step (S33) for using a noise model, which represents noise of the pseudo-optical interference signal, to generate pseudo-noise; and a superimposition step (S34) for superimposing the pseudo-optical interference signal and the pseudo-noise.

Description

Optical interference signal generating method, optical interference signal generating device, trained machine model, machine model learning method, machine model learning device, ocular tissue evaluation method, ocular tissue evaluation device, and program

[0001] The present invention relates to an optical interference signal generating method, an optical interference signal generating device, a trained machine model, a machine model training method, a machine model training device, an ocular tissue evaluation method, an ocular tissue evaluation device, and a program.

[0002] Evaluation or diagnosis of a subject using OCT (Optical Coherent Tomography) is performed based on changes in the characteristics (abnormalities) of the subject itself. Many of these changes manifest as changes in the density of scatterers (e.g., cell nuclei in biological tissue) within the subject. Generally, evaluation of scatterer density using OCT is performed by measuring OCT signals. OCT signals are signals that measure the intensity of interference light between incident light entering the subject and the light (backscattered light) reflected by scatterers within the subject. However, because the intensity of the OCT signal is strongly affected by factors other than scatterer density, such as the morphology of the subject itself, the depth of the measurement site, the probe light intensity of the OCT device, the signal-to-noise ratio, and the image resolution of the OCT, it has been difficult to actually estimate scatterer density from OCT signals.

[0003] On the other hand, a technique is known in which the influence of the various disturbances described above is eliminated by analyzing the local pattern of the OCT signal using a deep convolved neural network (DCNN, deep learning) to estimate the scattering density of the target (Patent Document 1).

[0004] OCT is an extremely widely used imaging diagnostic technique in the field of ophthalmology. In particular, OCT is widely used to diagnose glaucoma risk by quantifying the thickness of each retinal layer. This technique is classically performed by evaluating the thickness of the optic nerve fibers near the optic disc. Recently, it has become clear that glaucoma can be detected with greater sensitivity by evaluating the thickness of the three layers associated with ganglion cells in the macula of the retina: the optic nerve fiber layer (NFL), the ganglion cell layer (GCL), and the inner plexiform layer (IPL). Many ophthalmic OCT diagnostic devices (products) are equipped with diagnostic algorithms based on this method.

[0005] Recently, as disclosed in Non-Patent Document 1, it is expected that higher diagnostic accuracy can be achieved by directly visualizing and counting ganglion cells. However, such visualization requires a complex and expensive adaptive optics ophthalmoscope, and has not yet been put to clinical use. Therefore, it has become important to evaluate abnormalities in ocular tissues using widely used OCT diagnostic techniques.

[0006] International Publication No. WO2021 / 206076A1

[0007] Zhuolin Liu, et al., “Imaging and quantifying ganglion cells and other transparent neurons in the living human retina,” Proc. Nat. Accad. Sci. 114, 12803 (2017)

[0008] According to the studies of the inventors of the present invention, it has been found that when the signal-to-noise ratio is low, that is, when the signal generated by OCT contains a lot of noise, the estimated scatterer density value is biased, making it difficult to accurately estimate (measure) the scatterer density. For example, in the field of clinical ophthalmology, which is the largest market for OCT technology, the retina and cornea, which are the measurement targets, are highly transparent, and therefore the OCT signal intensity is weak and the signal-to-noise ratio is often low. Therefore, it is considered difficult to accurately estimate the scatterer density in this field.

[0009] For example, a machine model that has learned the intensity pattern of an optical interference signal through deep learning is thought to be able to estimate characteristic values ​​such as the density of a target scatterer from the intensity pattern of the optical interference signal, including noise. However, in order to train a machine model, optical interference signal data is required to serve as training data (learning data). However, until now, there has been no training data for optical interference signals that reflect the characteristics of noise.

[0010] Therefore, it is desirable to realize a technology that generates an optical interference signal that accurately reflects the characteristics of noise to be used as training data for a mechanical model, and that uses the generated optical interference signal to train the mechanical model.

[0011] Furthermore, conventional techniques have not always been able to accurately evaluate or diagnose abnormalities in ocular tissues, such as glaucoma and age-related macular degeneration, using OCT. Specifically, glaucoma is currently evaluated or diagnosed by evaluating the thickness of each retinal layer using OCT images. However, this diagnosis has limited accuracy, and furthermore, it has drawbacks in that an appropriate diagnosis cannot be obtained when diabetic retinopathy or traction caused by a premacular membrane is also present.

[0012] An OCT image is an image of the intensity of interference light between incident light incident on ocular tissue and reflected light (backscattered light) from the incident light reflected by scatterers in the ocular tissue. However, the measured interference light contains noise components, and the noise varies depending on the measurement target, light source, detector, measurement conditions, etc. One of the reasons why the accuracy of evaluation or diagnosis of abnormalities in ocular tissue using OCT images is not necessarily high is thought to be that the noise fluctuations cannot be accurately reflected in the evaluation or diagnosis.

[0013] Therefore, it is desirable to use OCT images to evaluate the state of ocular tissues more accurately than conventional techniques.

[0014] An optical interference signal generating method according to one aspect of the present invention includes a characteristic value setting step of setting characteristic values ​​of a pseudo-scattering medium, a signal generating step of generating a pseudo-optical interference signal from the pseudo-scattering medium having the set characteristic values, a noise generating step of generating pseudo-noise using a noise model representing noise in the pseudo-optical interference signal, and a superimposing step of superimposing the pseudo-optical interference signal and the pseudo-noise.

[0015] Furthermore, an optical interference signal generating device according to one aspect of the present invention includes a characteristic value setting unit that sets characteristic values ​​of a pseudo-scatterer, a signal generating unit that generates a pseudo-optical interference signal from the pseudo-scatterer having the set characteristic values, a noise generating unit that generates pseudo-noise using a noise model that represents noise in the pseudo-optical interference signal, and a superimposing unit that superimposes the pseudo-optical interference signal and the pseudo-noise.

[0016] Furthermore, a method for training a machine model according to one aspect of the present invention includes a characteristic value setting step of setting a characteristic value of a pseudo-scatterer; a superimposed signal generation step of generating a superimposed optical interference signal by superimposing a pseudo-optical interference signal from the pseudo-scatterer having the set characteristic value and a pseudo-noise generated using a noise model representing the noise of the pseudo-optical interference signal; and a learning step of using the set characteristic value and the generated superimposed optical interference signal to train a machine model that takes an optical interference signal obtained by actually measuring an object as an input and outputs at least one of the characteristic values ​​of the object.

[0017] Furthermore, a machine model learning device according to one aspect of the present invention includes a characteristic value setting unit that sets characteristic values ​​of a pseudo-scatterer; a superimposed signal generation unit that generates a superimposed optical interference signal by superimposing a pseudo-optical interference signal from the pseudo-scatterer having the set characteristic value and a pseudo-noise generated using a noise model that represents noise in the pseudo-optical interference signal; and a learning unit that uses the set characteristic value and the generated superimposed optical interference signal to learn a machine model that receives an optical interference signal obtained by actually measuring an object as input and outputs at least one of the characteristic values ​​of the object.

[0018] Furthermore, a trained machine model according to one aspect of the present invention is a trained machine model that takes as input an optical interference signal obtained by actually measuring an object to be measured, and outputs at least one of the characteristic values ​​of the object to be measured, the trained model using a set characteristic value, an optical interference signal superimposed with pseudo-noise generated using a pseudo-optical interference signal from an object having the characteristic value and a noise model representing the noise contained in the pseudo-optical interference signal.

[0019] An ocular tissue evaluation method according to one aspect of the present invention includes an acquisition step of acquiring an interference signal between incident light and scattered light of the incident light when light in a wavelength range from infrared light to visible light is incident on a target ocular tissue; a generation step of generating a local signal intensity pattern by numerically processing the local signal intensity of the interference signal; and a comparison step of comparing the signal intensity pattern of the target with a normal signal intensity pattern obtained by numerically processing the local signal intensity of the interference signal obtained by incidenting the light on normal ocular tissue.

[0020] An ocular tissue evaluation device according to one aspect of the present invention includes an acquisition unit that acquires an interference signal between incident light and scattered light of the incident light when light in a wavelength range from infrared light to visible light is incident on a target ocular tissue; a generation unit that performs numerical processing on the local signal intensity of the interference signal to generate a local signal intensity pattern; and a comparison unit that compares the signal intensity pattern of the target with a normal signal intensity pattern obtained by numerically processing the local signal intensity of the interference signal obtained by incidenting the light on normal ocular tissue.

[0021] The optical interference signal generating device, learning device, and ocular tissue evaluation device (hereinafter referred to as "devices") according to aspects of the present invention may be realized by a computer. In this case, the control program for the ocular tissue evaluation device, which causes the computer to operate as each part (software element) of the ocular tissue evaluation device, thereby realizing the ocular tissue evaluation device on the computer, and the computer-readable recording medium on which the control program is recorded, also fall within the scope of the present invention.

[0022] According to one aspect of the present invention, it is possible to generate an optical interference signal that accurately reflects noise characteristics and is used as training data for a machine model. Furthermore, it is possible to realize a technique for training a machine model using the generated optical interference signal.

[0023] According to one aspect of the present invention, abnormalities in ocular tissue can be evaluated using OCT images with greater accuracy than conventional techniques.

[0024] 1 is an explanatory diagram for explaining an overview of measurement signal processing according to a first embodiment of the present invention. FIG. 1 is a block diagram showing an example of the configuration of a measurement signal processing device 100 according to the first embodiment. FIG. 1 is a configuration diagram showing an example of an OCT system 1 according to the first embodiment. FIG. 2 is an explanatory diagram showing an example of a mathematical model according to the first embodiment. FIG. 2 is a schematic diagram showing the number, type, number of kernels, kernel size, stride, activation function, and axis for each layer in each column. FIG. 3 is a flowchart showing the flow of an optical interference signal generation method S3 for generating an optical interference signal according to a second embodiment of the present invention. FIG. 4 is a block diagram showing the configuration of an optical interference signal generation device 200 according to the second embodiment. FIG. 5 is a flowchart showing the flow of a machine model training method S4 according to the second embodiment. FIG. 6 is a block diagram showing the configuration of a machine model training device 300 according to the second embodiment. FIG. 7 is a graph showing the results of simulating scattered light from a pseudo object including pseudo scatterers. FIG. 8 is a graph showing the evaluation results of scatterer density when using the conventional technology and the noise model of the present embodiment when microparticles are assumed as pseudo scatterers. FIG. 9 is a graph showing the results of evaluating a tumor spheroid using the noise model of the present embodiment. FIG. 10 is a block diagram showing the configuration of an ocular tissue evaluation device 400 according to a third embodiment. FIG. 11 is a block diagram showing the configuration of an ocular tissue evaluation device 400A, which is another example of the third embodiment. 1 is a flowchart showing the flow of an ocular tissue evaluation method S5 according to embodiment 3. FIG. 1 is a graph showing the correlation between measured OCT signal intensity of the NFL and age, and the correlation between estimated cell density of the NFL and age. FIG. 2 is a graph showing the correlation between measured OCT signal intensity or estimated cell density in the GCL and age. FIG. 3 is a graph showing the correlation between measured OCT signal intensity or estimated cell density in the IPL and age. FIG. 4 is a diagram showing en face sectors of the retina. FIG. 5 is a graph showing the correlation coefficient r between OCT signal intensity or estimated cell density and age in each sector of the NFL. FIG. 6 is a graph showing the correlation coefficient r between OCT signal intensity or estimated cell density and age in each sector of the IPL.

[0025] [Embodiment 1] Hereinafter, a first embodiment of the present invention will be described in detail. The first embodiment is an embodiment that forms the basis for the second and subsequent embodiments. Hereinafter, the embodiments of the present invention will be described with reference to the drawings. First, an overview of this embodiment will be described. FIG. 1 is an explanatory diagram for explaining an overview of measurement signal processing according to this embodiment. A measurement signal processing device 100 according to this embodiment is capable of executing a learning step (S01) and an estimation step (S02). The learning step S01 is a process of determining model parameters for estimating a predetermined target value among characteristic values ​​that indicate the characteristics of the measurement signal from an input measurement signal according to a predetermined mathematical model. The predetermined target value may be one type or multiple types. In the learning step S01, the measurement signal processing device 100 performs supervised learning. Supervised learning means using a set consisting of known input values ​​and output values ​​as learning data (also referred to as supervised data, training data, teacher data, etc.) and recursively calculating model parameters so that an estimated value calculated for the input value according to a mathematical model approximates the output value.

[0026] In the learning step S01, the measurement signal processing device 100 uses a measurement signal (e.g., an OCT signal) and a target value as input and output values, respectively. At least one predetermined characteristic value of the measurement signal (in the example shown in FIG. 1 , the number of scatterers per unit volume (SPUV)) is used as the target value. In the learning step S01, the measurement signal processing device 100 generates multiple sets of input and output values ​​(hereinafter referred to as learning sets) and calculates model parameters so that the difference between the estimated value and the output value (hereinafter referred to as model error) is minimized across the multiple learning sets. For example, a convolutional neural network (CNN) can be used as the mathematical model. Here, minimization does not necessarily mean absolute minimization, but rather means performing calculations to estimate or search for model parameters with the goal of minimizing them as much as possible. Therefore, the model error does not necessarily decrease monotonically during the minimization process and may temporarily increase.

[0027] The learning step S01 according to this embodiment includes, for example, the following steps S102 to S108. (Step S102) The measurement signal processing device 100 sets multiple sets of multiple types of characteristic values ​​that indicate the characteristics of the measurement signal as characteristic value sets. In the measurement signal processing device 100, one characteristic value set includes multiple types of characteristic values, and at least one predetermined characteristic value among the multiple types of characteristic values ​​is set in advance as a target value. In other words, one characteristic value set may include multiple types of target values. Furthermore, one target value is determined to be common to multiple characteristic value sets. The number of types of characteristic values ​​included in one characteristic value set is set to be greater than the number of types of target values. As illustrated in FIG. 1 , when SPUV is adopted as one type of characteristic value as the target value, other types of characteristic values ​​such as intensity, SNR (Signal-to-Noise Ratio), spatial resolution, and ENS (Effective Number of Scatterers), or a combination of some or all of these, may be included in the characteristic value set.

[0028] (Step S104) The measurement signal processing device 100 determines, for each set of characteristic values, the distribution of scatterers in the sample located in a predetermined observation region as the tissue structure of the sample, using SPUV as an example of the target value set in step S102. (Step S106) The measurement signal processing device 100 generates (simulates) a measurement signal in the observation region according to a predetermined signal model, using the tissue structure determined for each set of characteristic values ​​and a type of characteristic value other than SPUV, which is the target value. The measurement signal processing device 100 may generate an OCT signal as an example of the measurement signal, or may generate OCT image data for visualizing the state of the sample based on the OCT signal (OCT imaging).

[0029] (Step S108) The measurement signal processing device 100 sets a set of measurement signals and target values ​​(SPUV in the example shown in FIG. 1 ) generated for each characteristic value set as a learning set including, as input values ​​and output values, respectively. The measurement signal processing device 100 calculates (learns) model parameters across multiple learning sets so as to minimize the model error between the estimated values ​​calculated for the measurement signals according to a mathematical model and the output values.

[0030] On the other hand, in the estimation step S02, the measurement signal processing device 100 executes, for example, the processes of the following steps S112 and S114. (Step S112) A measurement signal indicating the state of the sample in a predetermined observation region where the sample is placed is acquired by OCT measurement or the like. (Step S114) The measurement signal processing device 100 calculates an estimate of the target value (in this example, SPUV) using the model parameters obtained in the learning step S01 according to a predetermined mathematical model using the acquired measurement signal as an input value (target value estimation).

[0031] In this way, multiple sets of characteristic values ​​are determined for one target value, and measurement signals generated based on each of the characteristic value sets consisting of multiple types of characteristic values ​​are used as input values. Model parameters are determined so that estimated values ​​for the input values ​​across the multiple characteristic value sets approximate the target value. Using the determined model parameters, values ​​approximating the target value are calculated as estimated values ​​for measurement signals acquired separately as input values. Thus, the measurement signal processing device 100 can determine characteristic values ​​that are robust or invariant with respect to characteristic values ​​(e.g., intensity, resolution, etc.) other than the target value that are factors that cause fluctuations in the measurement signal.

[0032] (Measurement signal processing device) Next, a configuration example of the measurement signal processing device 100 according to this embodiment will be described. Fig. 2 is a block diagram showing a configuration example of the measurement signal processing device 100 according to this embodiment. The measurement signal processing device 100 includes a control unit 110, a storage unit 130, an input / output unit 140, a display unit 150, and an operation unit 160.

[0033] Some or all of the functions of the control unit 110 are realized as a computer including a processor such as a CPU (Central Processing Unit). The processor reads a program stored in advance in the storage unit 130 and performs processing instructed by instructions written in the read program to fulfill its functions. In this application, performing processing instructed by instructions written in a program may be referred to as program execution, program execution, etc. Some or all of the control unit 110 is not limited to general-purpose hardware such as a processor, but may also be configured to include dedicated hardware such as an LSI (Large Scale Integration) or an ASIC (Application Specific Integrated Circuit). The control unit 110 includes a measurement system control unit 112, a measurement signal acquisition unit 114, a training data generation unit 116, a model training unit 118, a characteristic estimation unit 120, and an output processing unit 122.

[0034] The measurement system control unit 112 controls the range of the observation area of ​​the object to be measured used as a sample installed in the measurement system. The measurement system control unit 112 sets the position or size of the observation area based on, for example, an operation signal input from the operation unit 160, and controls the position of the observation point within the set observation area. The measurement system control unit 112 drives, for example, a drive mechanism that changes the positions of galvanometer mirrors 60a and 60b in the OCT system 1 (FIG. 3) described below, to scan the observation point of the sample Sm. The observation point of the sample Sm is scanned in a direction intersecting the depth direction of the sample Sm (for example, a direction parallel to the front surface of the sample Sm).

[0035] The measurement signal acquisition unit 114 acquires measurement signals indicating the state of the object to be measured from the measurement system. For example, the measurement signal acquisition unit 114 sequentially acquires measurement signals from the spectrometer 70 in the OCT system 1. Based on the acquired measurement signals, the measurement signal acquisition unit 114 acquires signal values ​​indicating the intensity distribution of reflected light in the depth direction of the sample Sm for each observation point arranged at a predetermined interval. The measurement signal acquisition unit 114 repeats the process of acquiring the intensity distribution of reflected light for each observation point changed by scanning. The acquired intensity distribution of reflected light is based on the distribution of the refractive index of the sample Sm in the depth direction. The measurement signal acquisition unit 114 can acquire, as a measurement signal, an OCT signal indicating the state of the sample Sm within the observation area. The OCT signal has a signal value indicating, for each observation point, the complex amplitude of the intensity of interference light resulting from interference between a reference light incident on the sample Sm and light reflected from the sample Sm. The measurement signal acquisition unit 114 stores the acquired measurement signals in the storage unit 130.

[0036] The training data generation unit 116 generates a training set for each set of characteristic values. The training data generation unit 116 stores a series of data groups including the training sets generated for each set of characteristic values ​​in the storage unit 130 as training data. One training set includes a measurement signal as an input value and at least one predetermined characteristic value as a target value. First, the training data generation unit 116 generates multiple characteristic value sets. Each characteristic value set includes multiple types of characteristic values, and at least one of the multiple types of characteristic values ​​is used as a target value. However, the training data generation unit 116 determines each characteristic value set so that one target value is common to two or more characteristic value sets. The combinations of the multiple types of characteristic values ​​constituting each characteristic value set are different between the individual characteristic value sets. In the present application, a characteristic value set that shares some types of characteristic values ​​with a certain characteristic value set but differs from the remaining types of characteristic values ​​is distinguished as a separate characteristic value set. For example, in two sets of characteristic values ​​1 and 2, each including SPUV and SNR as characteristic values ​​and with SPUV as a target value, the training data generator 116 generates characteristic value set 1, in which the SPUV is 0.01 particles / μm³ and the SNR is 10 dB, and characteristic value set 2, in which the SPUV is 0.01 particles / μm³ and the SNR is 20 dB, as different characteristic value sets, although they share the same SPUV of 0.01 particles / μm³. Even when each target value is a set of characteristic values ​​including multiple types of characteristic values, the training data generator 116 distinguishes target values ​​that differ in at least one type of characteristic value as different target values. Generally, the training data generator 116 sets multiple target values. In other words, the total number of generated characteristic value sets is the sum of the number of characteristic value sets set for each target value. Therefore, the number of learning sets (number of sets) used to generate one set of model parameters is equal to the number of characteristic value sets, and is the sum of the number of characteristic value sets common to each one target value (this number is two or more as described above) across multiple target values.

[0037] Furthermore, each characteristic value can be either a scalar or a vector quantity. For example, SPUV, ENS, intensity, SNR, etc. are all scalar quantities. On the other hand, resolution may be treated as a vector quantity having the same number of dimensions as the observation region, or as an independent scalar quantity. For example, the resolution in a three-dimensional observation region may be treated as a three-dimensional vector containing the x-, y-, and z-direction resolutions as elements, or the x-, y-, and z-direction resolutions may be treated as separate scalar quantities. However, the x-, y-, and z-directions are directions of mutually orthogonal coordinate axes in a three-dimensional Cartesian coordinate system. The z-direction is the direction of depth in the sample, and the x- and y-directions are mutually orthogonal directions in a plane perpendicular to the z-direction. The process of determining this series of characteristic value sets corresponds to the process of step S102 described above.

[0038] In setting the characteristic value set, the training data generation unit 116 determines multiple possible values ​​within a predetermined range for each type of characteristic value. For example, the training data generation unit 116 determines three or more characteristic values ​​that are evenly distributed within a predetermined range for that type. Alternatively, the training data generation unit 116 may use a predetermined algebraic method to generate pseudorandom numbers uniformly distributed within a predetermined range, and then scale the generated pseudorandom numbers to a range corresponding to the type of characteristic value to determine the resulting value as the characteristic value. Examples of algebraic methods that can be used include the linear congruential method and the M-sequence method. For characteristic values ​​that are vector quantities, the training data generation unit 116 individually determines the values ​​of each element of the vector. Note that the range of the characteristic value may generally differ depending on the type of characteristic. However, a common range may be set for element values ​​or characteristic values ​​of the same type, such as x-direction resolution and y-direction resolution. The learning data generating unit 116 may be able to set the type and range of characteristic values ​​to be processed based on an operation signal input from the operation unit 160, for example.

[0039] The training data generator 116 generates a measurement signal having characteristics expressed by the characteristic values ​​in accordance with a predetermined propagation model for the incident wave, using some or all of the characteristic values ​​in the characteristic value set. The process of determining the scatterer distribution illustrated below corresponds to the process of step S104 above, and the process of determining an OCT speckle pattern as an example of a measurement signal from the scatterer distribution corresponds to the process of step S106 above.

[0040] For example, in the case where the characteristic values ​​constituting the characteristic value set include SPUVs, the training data generation unit 116 generates a tissue structure in which scatterers are stochastically distributed within a predetermined observation region using the determined SPUVs. More specifically, if the observation region is a three-dimensional region having a rectangular parallelepiped shape, the training data generation unit 116 calculates the number of scatterers by multiplying the determined SPUVs by the number of observation points within the occupied region of the sample located in the observation region. The occupied region of the sample is set in advance. If it is assumed that the sample occupies the entire observation region, the occupied region of the sample is not taken into account, and the training data generation unit 116 regards the number of observation points within the observation region as the number of observation points within the occupied region.

[0041] The training data generator 116 then sequentially generates uniform random numbers and determines the coordinate values ​​(x, y, z) of scatterers within the observation region by scaling every third generated random number with respect to the predetermined width, depth, and depth of the observation region. The training data generator 116 repeats the process of determining the positions of scatterers a number of times corresponding to the number of scatterers. A distribution of scatterers within the observation region (hereinafter referred to as the scatterer distribution f(x, y, z)) is generated as a tissue structure.

[0042] Taking the case where the characteristic value set includes spatial resolution as an example, the training data generation unit 116 determines a point spread function (PSF) corresponding to the spatial resolution. For example, the training data generation unit 116 determines a three-dimensional normal distribution having x-, y-, and z-directional resolutions as standard deviations in the x-, y-, and z-directions as shown in Equation (1), as PSF(x, y, z). Then, the training data generation unit 116 determines an OCT speckle pattern O(x, y, z) by convolving the point spread function PSF(x, y, z) with the scatterer distribution f(x, y, z), as shown in Equation (2). The OCT speckle pattern is an OCT signal that can be observed when incident light is scattered by a scatterer. In Equation (1), σx, σy, and σz represent the standard deviations in the x-, y-, and z-directions, respectively.

[0043]

[0044]

[0045] For example, in the case where the characteristic value set includes intensity, the training data generator 116 scales the measurement signal by multiplying the signal value at each observation point in the above-described OCT speckle pattern O(x, y, z) by a constant proportional to the intensity. The training data generator 116 updates the data indicating the signal value at each observation point after scaling as a new OCT speckle pattern (x, y, z).

[0046] Taking the case where the characteristic value set includes an SNR as an example, the training data generation unit 116 adjusts the measurement signal by adding a noise component value to the signal value for each observation point in the above-mentioned OCT speckle pattern O(x, y, z). When determining the noise component value, the training data generation unit 116 sequentially generates pseudorandom numbers whose distribution is a complex normal distribution and determines the noise component value for each observation point by multiplying each generated random number by a gain. The training data generation unit 116 determines a gain by which the noise component value is multiplied so that the ratio of the sum of the absolute squares of the signal values ​​to the sum of the absolute squares of the noise component values ​​in the observation area is equal to the SNR. Note that if scaling is performed on the measurement signal, the training data generation unit 116 adds the noise component value to the scaled measurement signal.

[0047] Note that multiple types of characteristic values ​​may be mutually dependent. In such cases, even if a certain type of characteristic value is not set, the training data generation unit 116 may derive that type of characteristic value from other types of characteristic values. Furthermore, the training data generation unit 116 may or may not use the derived characteristic value as a target value. For example, the training data generation unit 116 may determine the SPUV by dividing the set ENS by a coherence volume, which is the product of the spatial resolutions in the x, y, and z directions. Conversely, the training data generation unit 116 may determine the ENS by multiplying the SPUV by the coherence volume, which is the product of the spatial resolutions in the x, y, and z directions. Furthermore, the training data generation unit 116 may determine the cube root of the volume obtained by dividing the ENS by the ratio of the SPUV and the spatial resolutions in the z and x-y directions as the x-direction spatial resolution and the y-direction spatial resolution, respectively, and may multiply the x-direction spatial resolution by the ratio of the spatial resolutions to determine the z-direction spatial resolution. Alternatively, the learning data generator 116 may determine the SNR to be proportional to the intensity of the signal component by assuming the noise level, i.e., the sum of squares of the noise component values ​​within the tactile region, to be a constant value, or may calculate the signal value for each observation point by adding the signal component value and the noise component value scaled by that intensity. As described above, the SNR is given by the ratio of the sum of squares of the absolute values ​​of the signal value to the sum of squares of the absolute values ​​of the noise component value.

[0048] In the above description, an example was given in which an OCT speckle pattern within an observation region in a three-dimensional space was defined as a measurement signal. However, a similar technique can be used to define an OCT speckle pattern within an observation region in a two-dimensional space as a measurement signal. In this case, it is sufficient to omit the coordinate in one of the x, y, and z coordinate axes. For example, the point spread function PSF(x, y) and the OCT speckle pattern O(x, y) in a two-dimensional space such as the surface of a sample can be given by omitting the z coordinate, as shown in Equations (3) and (4), respectively.

[0049]

[0050]

[0051] The model training unit 118 reads training data from the storage unit 130 and determines a set of model parameters using a series of multiple training sets included in the read training data. The process of determining the model parameters corresponds to the process of step S108 described above. Model parameters refer to one or more parameters used in calculations according to a mathematical model, and are sometimes called parameter sets or hyperparameters. The model training unit 118 sets measurement signals for each training set included in the training data as input values ​​of the mathematical model. The model training unit 118 calculates model parameters so as to minimize the model error between the target value and an estimated value calculated as an output value for the measurement signals according to the mathematical model for the entire series of training data. As a loss function, which is an index indicating the magnitude of the model error, for example, the mean squared error (MSE) can be used. MSE is the simple average of the squares of the differences between the target value ci and the estimated value cip among N samples i, as shown in Equation (5). The MSE is an index indicating that the smaller the value, the smaller the model error, i.e., the higher the degree of approximation of the estimated value to the target value. The model learning unit 118 repeats a process of recursively updating the model parameters using, for example, the steepest descent method until the MSE becomes equal to or less than a predetermined determination threshold. The model learning unit 118 stores model parameter data indicating the determined model parameters in the storage unit 130. An example of the mathematical model according to this embodiment will be described later.

[0052]

[0053] The characteristic estimation unit 120 reads out the model parameter data stored by the model learning unit 118 and the measurement signal stored by the measurement signal acquisition unit 114 from the storage unit 130. The process of acquiring the measurement signal corresponds to the process of step S112 described above. The characteristic estimation unit 120 calculates, as an output value, an estimate of the target value to be calculated, using the model parameters indicated by the read model parameter data, in accordance with the mathematical model described above for the read measurement signal. The process of calculating the estimate corresponds to the process of step S114 described above. The characteristic estimation unit 120 associates the calculated estimate with the measurement signal and stores it in the storage unit 130.

[0054] The output processing unit 122 controls the output of various data based on an operation signal input from the operation unit 160. The operation signal instructs, for example, whether to display an OCT image or whether to display a target value. When an operation signal instructing the display of an OCT image is input, the output processing unit 122 reads the measurement signals acquired by the measurement signal acquisition unit 114 or the learning data generation unit 116 from the storage unit 130, converts the absolute values ​​of the complex amplitudes, which are signal values ​​for each observation point indicated by the measurement signals, into luminance values ​​within a predetermined range of values ​​for the pixels corresponding to the observation points, and outputs output image data having the converted luminance values ​​to the display unit 150 as display data. As a result, an OCT image based on the output image data is displayed on the display unit 150. The output processing unit 122 may select the measurement signals to be output based on the operation signal. When an operation signal instructing the display of a target value is input, the output processing unit 122 reads the estimated value of the target value calculated by the characteristic estimation unit 120 from the storage unit 130 and outputs display data indicating the read estimated value to the display unit 150. The output processing unit 122 may superimpose the estimated value on an OCT image based on the measurement signal used to calculate the estimated value and display it on the display unit 150 .

[0055] The output processing unit 122 may read setting screen data for guiding the setting of the above-mentioned setting values ​​(for example, the observation region, the target value, the type of characteristic value, the value range, etc.) from the storage unit 130, and output the read setting screen data as display data to the display unit 150. In this case, a setting screen based on the setting screen data is displayed on the display unit 150.

[0056] The storage unit 130 stores the above programs as well as various data used in the processes executed by the control unit 110 and various data acquired by the control unit 110. The storage unit 130 includes, for example, a non-volatile (non-temporary) storage medium such as a ROM (Read Only Memory), a flash memory, or a HDD (Hard Disk Drive). The storage unit 130 includes, for example, a volatile storage medium such as a RAM (Random Access Memory) or a register.

[0057] The input / output unit 140 is connected wirelessly or via a wire to other devices so as to input and output various types of data. The input / output unit 140 includes, for example, an input / output interface. The input / output unit 140 is connected to, for example, peripheral devices and a measurement system. The display unit 150 displays display information such as images, characters, and symbols based on display data input from the control unit 110. The display unit 150 may include, for example, a liquid crystal display, an organic electroluminescence display, or the like.

[0058] The operation unit 160 may be configured to include components such as buttons, knobs, dials, a mouse, and a joystick that receive user operations and generate operation signals in response to the received operations. The operation unit 160 outputs the acquired operation signals to the control unit 110. The operation unit 160 may be an input interface that receives operation signals wirelessly or via a wired connection from another device (for example, a portable device such as a remote controller).

[0059] (OCT System) Next, an OCT system 1 according to this embodiment will be described. FIG. 3 is a configuration diagram showing an example of the OCT system 1 according to this embodiment. The OCT system 1 is an observation system for observing the state of a sample using OCT. The OCT system 1 irradiates light onto the sample Sm, acquires interference light generated by interference between the light reflected from the sample Sm and the reference light reflected by the reference mirror 40, and acquires an OCT signal indicating the state of the surface and interior of the sample Sm as a measurement signal from the acquired interference light. The OCT system 1 converts the acquired measurement signal into an OCT image to visualize it.

[0060] The object to be observed as the sample Sm may be, for example, a living human or animal body, or a non-living body. The living body may be the fundus of the eye, blood vessels, teeth, subcutaneous tissue, etc. The non-living body may be an artificial structure such as an electronic component or a mechanical component, a natural structure such as stone or a mineral, or a substance without a specific shape.

[0061] The OCT system 1 includes a light source 10, a beam splitter 20, collimators 30a, 30b, 50a, and 50b, a reference mirror 40, galvanometer mirrors 60a and 60b, a spectrometer 70, and a measurement signal processing device 100. Among these components, the beam splitter 20, the collimators 30a, 30b, 50a, and 50b, the reference mirror 40, the galvanometer mirrors 60a and 60b, and the spectrometer 70 constitute an optical system called an interferometer. The interferometer illustrated in FIG. 1 is a Michelson interferometer. More specifically, the light source 10, the spectrometer 70, the collimators 30a, and the collimators 50a are each connected to the beam splitter using optical fibers. The OCT system 1 is, for example, a Fourier-domain OCT (FD-OCT).

[0062] The light source 10 is a broadband light source such as an ultrashort pulse laser or an SLD (Superluminescent Diode). The light source 10 emits, for example, a probe light having a near-infrared wavelength (e.g., 800 to 1000 nm) and low coherence. The light emitted from the light source 10 is guided inside an optical fiber and enters the beam splitter 20. The beam splitter 20 splits the incident light into light guided toward the collimator 30a (hereinafter referred to as reference light) and light guided toward the collimator 50a (hereinafter referred to as measurement light). The beam splitter 20 is, for example, a cube beam splitter.

[0063] The collimator 30a converts the reference light guided from the beam splitter 20 into parallel light and emits the parallel light toward the collimator 30b. The collimator 30b collects the parallel light incident from the collimator 30a and emits the collected reference light toward the reference mirror 40. The collimator 30b receives the reference light reflected by the reference mirror 40, converts it into parallel light, and emits the converted parallel light toward the collimator 30a. The collimator 30a collects the parallel light incident from the collimator 30b and guides it toward the beam splitter 20.

[0064] On the other hand, the collimator 50a converts the measurement light guided from the beam splitter 20 into parallel light and emits the parallel light toward the galvanometer mirror 60a. The parallel light incident from the collimator 50a is reflected by the surfaces of the galvanometer mirrors 60a and 60b and emitted toward the collimator 50b. The collimator 50b focuses the parallel light incident from the collimator 50a via the galvanometer mirrors 60a and 60b and irradiates the sample Sm with the focused measurement light. The measurement light irradiated to the sample Sm is reflected by a reflecting surface of the sample Sm and enters the collimator 50b. The reflecting surface is not limited to, for example, the boundary surface between the sample Sm and the surrounding environment (e.g., the atmosphere), but can also be a boundary surface separating materials or tissues with different refractive indices within the sample Sm. Hereinafter, the light reflected by the reflecting surface of the sample Sm and incident on the collimator 50b is referred to as reflected light.

[0065] The collimator 50b emits the incident reflected light toward the galvanometer mirror 60b. The light is reflected by the surfaces of the galvanometer mirrors 60b and 60a, respectively, and emitted toward the collimator 50a. The collimator 50a collects the parallel light incident from the collimator 50a via the galvanometer mirrors 60a and 60b, and guides the collected reflected light toward the beam splitter 20.

[0066] The beam splitter 20 guides the reference light reflected by the reference mirror 40 and the light reflected by the sample Sm to the spectrometer 70 via an optical fiber. The spectrometer 70 includes a diffraction grating and a light receiving element. The diffraction grating separates the reference light and the reflected light guided from the beam splitter 20. The separated reference light and reflected light interfere with each other to form interference light. The light receiving element is disposed on an imaging surface onto which the interference light is irradiated. The light receiving element detects the irradiated interference light and generates a measurement signal corresponding to the detected interference light. The light receiving element outputs the generated measurement signal to the measurement signal processing device 100. It is desirable that the distance between adjacent observation points in the observation region be equal to or less than the spatial resolution of the optical system.

[0067] (Mathematical Model) Next, an example of a mathematical model according to this embodiment will be described. FIG. 4 is an explanatory diagram showing an example of a mathematical model according to this embodiment. FIG. 5 is a table showing the configuration of the mathematical model exemplified in FIG. 4. In FIG. 5, each column shows the layer number, type, number of kernels, kernel size, stride, activation function, and axis. However, - indicates that there is no corresponding item due to the layer type.

[0068] The mathematical model shown in Figures 4 and 5 is a CNN. In the example shown in Figures 4 and 5, the input value to the CNN is a three-dimensional OCT signal, and the output value from the CNN is used to calculate a one-dimensional target value (scalar). Examples of one-dimensional characteristic values ​​include diffuser density, signal intensity, ENS, SNR, and resolution in each direction (i.e., x, y, and z direction resolutions).

[0069] A CNN is a type of artificial neural network and includes one input layer, multiple intermediate layers, and an output layer. The CNN illustrated in FIG. 4 includes an input layer 118a, a feature extraction layer group 118b, and an estimation layer group 118c. In FIG. 5, the 0th layer corresponds to the input layer 118a, the 1st to 16th layers correspond to the feature extraction layer group 118b, and the 17th and 18th layers correspond to the estimation layer group 118c. The layers in the feature extraction layer group 118b and the first layer in the estimation layer group 118c correspond to intermediate layers, and the last layer in the estimation layer group 118c corresponds to the output layer. Each layer has one or more nodes (also referred to as nodes, neurons, etc.). Each node outputs a function value of a predetermined function for an input value as an output value. However, when dealing with three-dimensional measurement signals, the resolution takes the form of a three-dimensional vector value. Therefore, when calculating the resolution as a three-dimensional output value, a CNN with an output layer including three nodes may be used.

[0070] The input layer 118a outputs signal values ​​for each observation point indicated by the measurement signal input as an input value to the next layer. Each node in the input layer 118a receives a signal value from the corresponding observation point, and the input signal value is output to the corresponding node in the next layer. The number of kernels in FIG. 5 indicates the number of kernels used to process (e.g., calculate) each input value. A kernel refers to a processing unit for calculating one output value at a time. A kernel is also called a filter. The kernel size indicates the number of input values ​​used in one kernel processing. For example, the 0th layer in FIG. 5 has a kernel with 32 x 32 x 32 nodes constituting the input layer 118a, and indicates that one output value is given for each input value to each node.

[0071] The feature extraction layer group 118b calculates feature extraction values ​​indicating the features of multidimensional input values. In the example shown in FIG. 4 , the feature extraction layer group 118b is formed by alternately stacking one or more convolutional layers and pooling layers two or more times. More specifically, the feature extraction layer group 118b is formed by stacking two adjacent convolutional layers Conv11 and Conv12, one pooling layer Pool13, two convolutional layers Conv21 and Conv22, one pooling layer Pool23, two convolutional layers Conv31 and Conv32, one pooling layer Pool33, and convolutional layers Conv41 and Conv42 in that order. However, one normalization layer is sandwiched between the layer following each of the two convolutional layers. The feature extraction layer group 118b includes a smoothing layer as the final layer.

[0072] A convolutional layer is a layer that calculates convolution values ​​for each kernel by performing a convolution operation on input values ​​input to each of multiple nodes from the previous layer, calculates a function value of a predetermined activation function for a correction value obtained by adding the calculated convolution value and a bias value as an output value, and outputs the calculated output value to the next layer. In the convolution operation, each node receives one or more input values ​​from the previous layer, and an independent convolution coefficient is used for each input value. The convolution coefficients, bias values, and activation function parameters form one set of model parameters.

[0073] For example, in the first and second layers of FIG. 5 , the number of kernels is set to 1 and 20, respectively. The kernel size of both the first and second layers is 32 × 32 × 32. The number of kernels of 20 indicates that there are 20 convolution coefficients, and output values ​​corresponding to each of the 20 kernels are obtained for each node, which are then input to the corresponding node in the next layer. Furthermore, "ReLU" in the activation function column indicates that the activation function is a rectified linear unit (RLU). A RLU is a function that sets a threshold (e.g., 0) as the output value for input values ​​below that threshold, and outputs input values ​​exceeding the threshold as is. Therefore, this threshold can also be part of a set of model parameters. Furthermore, for convolutional layers, whether or not to reference input values ​​from nodes in the previous layer and whether or not to output output values ​​to nodes in the next layer can also be part of a set of model parameters. Therefore, unlike the fully connected layer described below, each node in a convolutional layer is not necessarily connected to all nodes in the previous layer so that it receives input values, nor is it necessarily connected to all nodes in the next layer so that it outputs output values.

[0074] A normalization layer is a layer that normalizes input values ​​input to multiple nodes within a predetermined range using a common normalization parameter (batch normalization) and outputs the normalized values ​​as output values. The axis item of the third layer in Figure 5 contains a code indicating the type and range of parameters used as the axis for normalization. The -1 set in this item indicates that the output values ​​from all nodes are normalized within a range from 0 to 1 by multiplication using a common multiplication value.

[0075] A pooling layer is a layer having nodes that determine a representative value from input values ​​received from multiple nodes in the previous layer and output the determined representative value as an output value to the next layer. The representative value is, for example, a value that statistically represents multiple input values, such as the maximum value, average value, or mode. Note that the "2" set in the stride field for the fourth layer in Figure 5 indicates the range of adjacent nodes in the previous layer that reference the input value for one node. In the example shown in Figure 5, 2 x 2 x 2 spatially adjacent previous nodes in three-dimensional space are referenced. Therefore, a pooling layer is a layer that downsamples input values ​​from the previous layer to a lower dimension and provides output values ​​to the next layer.

[0076] The estimation layer group 118c is a layer that calculates estimated values ​​as output values ​​from the feature extraction values ​​input from the feature extraction layer group 118b. The estimation layer group 118c is formed by stacking a flattening layer (see FIG. 5, but omitted in FIG. 4) and two fully connected layers FC01 and FC02. The flattening layer unrolls input values ​​of multidimensional samples (three or more dimensions) into output values ​​of two-dimensional samples. For example, the 16th layer in FIG. 5 receives as input values ​​the 80 three-dimensional 4×4×4 element values ​​for each observation point normalized and output by the 15th layer, and linearly rearranges the three-dimensional array of 5,120 input values ​​into a single vector having 5,120-dimensional element values. Therefore, the flattening layer does not perform any substantial computation. The fully connected layer is a layer that performs a convolution operation on each of multiple nodes for input values ​​received from the previous layer to calculate a convolution value, calculates an output value by adding the calculated convolution value and a bias value, and outputs the calculated output value to the next layer. In other words, the fully connected layer is a layer that performs a convolution operation on all of the multiple input values ​​received from the previous layer using parameter sets (kernels) whose number is less than the number of input values, and outputs the calculated value. A predetermined activation function may be set for each node of the fully connected layer, and the function value of the predetermined activation function for the calculated correction value may be output to the next layer as an output value. However, in the 17th layer of FIG. 5 , a normalized linear unit is set as the activation function for each of the 128 kernels, and one output value is calculated by referring to the input values ​​from the 40 nodes of the 16th layer as input values. In the 18th layer of FIG. 5 , one convolution value obtained by performing a convolution operation on the input values ​​from the 128 nodes of the 17th layer is calculated as an output value. Therefore, in a fully connected layer, the convolution coefficients, bias values, and activation function parameters are part of one set of model parameters. By using a fully connected layer as one or more final layers, including the output layer, the degrees of freedom are reduced while taking into consideration all components that significantly affect the characteristic values ​​provided by the immediately preceding layer, and the final characteristic values ​​can be derived.

[0077] The number of layers of the CNN, the type of each layer, the number of nodes in each layer, and so forth are not limited to those shown in FIGS. 4 and 5 . The CNN according to this embodiment may be configured to calculate an estimate of a predetermined target value as an output value for a measurement signal having signal values ​​for multiple observation points as input values. The estimate is not limited to a one-dimensional scalar value as described above, but can also be a vector or matrix of two or more dimensions. For example, the characteristic value may be a three-dimensional resolution vector having x-direction resolution, y-direction resolution, and z-direction resolution as elements. Furthermore, the measurement signal is not limited to a three-dimensional signal in which observation points are distributed in three-dimensional space, but may also be a signal of lower dimension than a two-dimensional signal or a high-dimensional signal of four or more dimensions. However, as exemplified in FIGS. 4 and 5 , the CNN according to this embodiment preferably includes an intermediate layer configured by alternately stacking one or more convolutional layers and pooling layers for two or more cycles. This is because the repetition of the convolutional layers narrows down the components that significantly affect the characteristic value. Note that the pooling layer may be omitted in this repetition of the convolutional layers.

[0078] As described above, the characteristic estimation unit 120 uses a mathematical model such as a CNN to calculate an estimated value for each block consisting of multiple adjacent observation points that form an OCT signal. Therefore, an OCT signal that generates one frame of an OCT image typically corresponds to multiple blocks. Therefore, when an operation signal indicating display of a target value is input, the output processing unit 122 may convert the estimated value calculated for each block into a luminance value within a predetermined range of the pixel corresponding to that block. The output processing unit 122 outputs display data indicating the converted luminance value to the display unit 150. As a result, a target value image that represents the distribution of target values ​​for each block using luminance is displayed on the display unit 150.

[0079] (Calculation Example of Characteristic Values) Next, a calculation example of characteristic values ​​calculated in the estimation step (S02) using the model parameters obtained by executing the learning step (S01) will be described. In the following calculation example, a two-dimensional sample surface (en-face surface) with 32 x 32 observation points is used as the observation region. Each characteristic value set includes x-direction resolution, y-direction resolution, SPUV, and SNR as characteristic values. As a measurement signal, an OCT speckle pattern in which scatterers are stochastically distributed in the SPUV and which has set x-direction resolution, y-direction resolution, and SNR as its characteristics is used. Then, the model learning unit 118 calculates CNN model parameters for calculating an estimate of the target value using the measurement signal and multiple sets of training data having any of the x-direction resolution, y-direction resolution, SPUV, and SNR as the target value. That is, a total of four sets of model parameters are calculated for calculating estimated values ​​of the x-direction resolution, y-direction resolution, SPUV, and SNR from one common measurement signal given by each set of characteristic values. Note that, for example, the value ranges of the x-direction resolution and y-direction resolution may be 3 to 20 μm, the value range of the SNR may be 0 to 50 dB, and the SPUV may be 0 to 0.0497 / μm2. The characteristic values ​​of each type may be determined using uniform random numbers so as to be randomly distributed within their value ranges.

[0080] Alternatively, a learning set based on each of 50,000 characteristic value sets may be used to calculate one set of model parameters. The 50,000 characteristic value sets may include multiple characteristic value sets each sharing one target value (e.g., SPUV 0.03 / μm²), and a 32-pixel x 32-pixel OCT speckle pattern may be generated from each of the characteristic value sets. While depending on the specifications of various devices and the scanning method of the OCT signal, the size of one pixel corresponds to approximately 1 μm to several μm. The characteristic estimation unit 120 may then use an OCT speckle pattern, whose target values ​​indicating the characteristics are known, as a measurement signal and calculate estimates of the respective target values ​​using each of the four sets of model parameters calculated by the model training unit 118, according to the CNN illustrated in FIG. 5 . However, the characteristic estimation unit 120 may use an OCT speckle pattern separate from the OCT speckle pattern used by the model training unit 118 to calculate the model parameters.

[0081] In the above example, the x-direction resolution, y-direction resolution, SPUV, SNR, and ENS are mainly used as characteristic values, but the present invention is not limited to this. The characteristic value set may also include, for example, aberration coefficients that indicate the mode and degree of wavefront aberration of the optical system. Aberrations cause phase changes in the xy plane for measurement signals, and therefore can be considered as a type of factor that causes fluctuations in the measurement signals. For example, Zernike coefficients A nm can be used as aberration coefficients. For example, the 0th,0th order Zernike coefficient A00 is a constant term, the 1st,0th order Zernike coefficient A10 is the tilt component in the x direction, the 1st,1st order Zernike coefficient A11 is the tilt component in the y direction, the 2nd,0th order Zernike coefficient A20 is the astigmatism in the 0° direction and the 90° direction, the 2nd,1st order Zernike coefficient A21 is the focus shift, the 2nd,2nd order Zernike coefficient A22 is the astigmatism in the 45° direction, the 3rd,1st order Zernike coefficient A31 is the coma aberration of the x component, and the 3rd,2nd order Zernike coefficient A32 is the coma aberration of the y component, etc. These are coefficients that indicate the degree of each of these.

[0082] When the incident direction of light is the z direction, the aberration W(fx, fy) for a frequency (fx, fy) on a two-dimensional plane xy is expressed by the Zernike polynomial shown in equation (6). In equation (6), ρ indicates the frequency in the radial direction from the optical axis directed in the z direction, and θ indicates the frequency in the azimuthal direction. Rn n-2m(ρ) indicates a radial polynomial of the nth and n-mth degrees, and is defined by equation (7). In equation (7), ...! indicates the factorial of the integer .... n is an integer equal to or greater than 0, and m is an integer equal to or greater than 0 and equal to or less than n. However, the maximum degree k is a predetermined integer equal to or greater than 0. That is, the aberration W(fx, fy) is given by the sum of the products of the nth and mth Zernike coefficients Anm, the nth and n-2mth radial polynomials at that position, and the cosine value cos|n-2m|θ (where n-2m is 0 or more) or the sine value sin|n-2m|θ (where n-2m is less than 0).

[0083]

[0084]

[0085] Therefore, when generating a measurement signal that includes the influence of aberration, the training data generation unit 116 performs a two-dimensional Fourier transform on the xy plane on the three-dimensional point spread function PSF(x, y, z) to convert it into a transformation coefficient p(fx, fy; z) in the frequency domain (fx, fy). Then, as shown in equation (8), the training data generation unit 116 multiplies the transformation coefficient p(fx, fy; z) by a phase factor eiaW(fx, fy) that indicates the phase change caused by the aberration to obtain a transformation coefficient after the phase change, and performs a two-dimensional inverse Fourier transform on the fx-fy plane on the transformation coefficient after the phase change, thereby calculating a point spread function PSF′(x, y, z) that includes the phase change due to the aberration in the three-dimensional spatial domain (x, y, z). In equation (8), a is a predetermined real value that indicates the degree of phase change due to the aberration. Ffx,fz-1[...] denotes the two-dimensional inverse Fourier transform of... on the fx-fy plane.

[0086]

[0087] The training data generator 116 can generate, as a measurement signal, an OCT speckle pattern O'(x, y, z) that has undergone a phase change due to aberration by performing a convolution operation on the scatterer distribution f(x, y, z) with the PSF'(x, y, z) that includes aberration. Thus, by using the measurement signal that has undergone a phase change due to aberration as an input value of the training data, the training data generator 116 can calculate model parameters for calculating an aberration coefficient as a target value, or model parameters for robustly calculating other types of characteristic values ​​as target values ​​against variations in the aberration coefficient.

[0088] The characteristic value set may also include, for example, a dispersion coefficient related to chromatic dispersion of the optical system. Chromatic dispersion causes a phase change in the measurement signal in the z direction according to the wavelength, and can therefore be considered a type of factor that causes fluctuations in the measurement signal. Generally, chromatic dispersion occurs due to the frequency dependence of the propagation constant in the optical system. The propagation constant is a physical quantity equivalent to an effective wave number. Taylor expansion of the propagation constant β(ω) with respect to frequency ω around the center frequency ω0 of the light source is expressed as in Equation (9).

[0089]

[0090] In equation (9), β(ω) represents the propagation constant at the center frequency ω of the incident light. a1 and a2 represent the first and second derivatives at the center frequency ω, respectively. The first derivative a1 corresponds to the reciprocal of the group velocity, and the second derivative a2 corresponds to the group velocity dispersion. Therefore, for example, the group velocity and group velocity dispersion can be used as the dispersion coefficient.

[0091] Therefore, when generating a measurement signal that includes the influence of chromatic dispersion, the training data generation unit 116 performs a one-dimensional Fourier transform in the z direction on the three-dimensional point spread function PSF(x, y, z) to convert it into a transformation coefficient q(ω; x, y) in the frequency domain (ω). Then, as shown in equation (10), the training data generation unit 116 multiplies the transformation coefficient q(ω; x, y) by a phase factor eibβ(ω) that indicates the phase change due to chromatic dispersion, and performs a one-dimensional inverse Fourier transform in the ω direction on the transformation coefficient after the phase change, thereby calculating the point spread function PSF″(x, y, z) that has undergone a phase change due to chromatic dispersion in the three-dimensional spatial domain (x, y, z). Note that in equation (10), b is a predetermined real value that indicates the degree of phase change due to chromatic dispersion. Fω−1[...] indicates a one-dimensional inverse Fourier transform in the ω direction of ....

[0092]

[0093] The training data generation unit 116 can generate, as a measurement signal, an OCT speckle pattern O''(x, y, z) that has undergone a phase change due to wavelength dispersion by performing a convolution operation on the above-mentioned scatterer distribution f(x, y, z) with a PSF''(x, y, z) that includes a phase change due to wavelength dispersion. Thus, by using the measurement signal that has undergone a phase change due to wavelength dispersion as an input value of the training data, the training data generation unit 116 can calculate model parameters for calculating a dispersion coefficient as a target value, or model parameters for robustly calculating other types of characteristic values ​​as target values ​​against fluctuations in the dispersion coefficient. Note that the characteristic value set may include either an aberration coefficient or a dispersion coefficient, or both, as the dispersion coefficient.

[0094] In the above description, the measurement signal to be processed by the measurement signal processing device 100 is an OCT signal, but this is not limiting. This embodiment can also be applied to measurement signals that indicate the state of a sample using a measurement principle other than OCT or signals for analyzing that state. The measurement signal is not limited to a raw signal obtained by measurement or a signal generated by simulating a raw signal obtained by measurement, but may also be a signal obtained by performing predetermined post-processing on such a signal, such as output display data primarily intended for visualization, such as an OCT image.

[0095] The measurement signal may be, for example, a speckle interferometry signal, a photoacoustic measurement signal, an ultrasonic tomography signal, or the like. The speckle interferometry signal is a measurement signal acquired using speckle interferometry. Speckle interferometry involves the steps of irradiating a sample in a predetermined reference state with a coherent wave such as laser light as an incident wave and detecting, as a reference wave, an interference wave resulting from interference between the incident wave and a reflected wave from the sample; further detecting, using a similar technique, an interference wave generated by irradiating a sample whose state, such as position, orientation, or shape, has changed from the reference state; and acquiring a speckle interferometry signal indicating interference fringes obtained by interfering the detected interference wave with the reference wave. The speckle interferometry signal is primarily used to analyze the amount of movement, change in orientation, or deformation of the sample as a characteristic value. The incident wave is not limited to visible light, such as laser light, but may also be X-rays, infrared rays, ultrasound, or the like.

[0096] A photoacoustic measurement signal is a measurement signal obtained using a photoacoustic method. The photoacoustic method involves irradiating a sample with laser light, which is visible light or infrared light (including near-infrared light), as an incident wave and detecting the sound waves emitted from the sample as a photoacoustic measurement signal. Here, the molecules constituting the sample absorb the incident wave and generate heat, which causes the sample to expand due to the generated heat, generating sound waves. The photoacoustic measurement signal is used to determine the distribution of optical absorbers contained in the sample. Therefore, characteristic values ​​derived from the distribution of optical absorbers can be used to analyze, for example, the density of optical absorbers and the active sites of the sample. The generated sound waves are not limited to audible sounds but can also be ultrasonic waves.

[0097] An ultrasonic tomography signal is a measurement signal obtained by ultrasonic tomography. Ultrasonic tomography includes the steps of irradiating a sample with ultrasonic waves as an incident wave and detecting an interference wave formed by interference between the incident wave and a reflected wave emitted from the sample as an ultrasonic tomography signal. The ultrasonic tomography signal is used to analyze characteristic values ​​derived from tissue density, its distribution, structure, etc. The above tomography method is not limited to a method using visible light or ultrasonic waves as an incident wave, but may also use other types of waves such as infrared light, ultraviolet light, or X-rays.

[0098] As described above, the measurement signal processing device 100 according to this embodiment includes a model learning unit 118 that determines, for each learning set including a measurement signal and at least one predetermined characteristic value indicating the characteristics of the measurement signal as a target value, model parameters for calculating an estimated value calculated using a predetermined mathematical model for the measurement signal so as to minimize the difference between the target value and the estimated value. The measurement signal processing device 100 also includes a training data generation unit 116 that determines characteristic value sets, which are sets of multiple characteristic values ​​indicating the characteristics of the measurement signal including the target value, so that the characteristic value sets are common to multiple characteristic value sets for each target value, and generates, for each characteristic value set, a learning set including the measurement signal having characteristics indicated by the multiple characteristic values ​​and the target value. The measurement signal may be any of an OCT signal, a speckle interferometry signal, a photoacoustic measurement signal, an ultrasound tomography signal, etc. One target value may include two or more characteristic values. In this case, model parameters for estimating two or more characteristic values ​​from the measurement signal can be generated. Using the generated model parameters, it is possible to estimate two or more characteristic values ​​from the measurement signal with practical accuracy, which was previously difficult or impossible. The two or more types of characteristic values ​​may include, for example, the scatterer density and the resolution of the measurement system.

[0099] This configuration generates measurement signals based on multiple types of characteristic values ​​so that each characteristic value is common across multiple learning sets, and model parameters can be determined so that the estimated value calculated from the generated measurement signals approximates the target value. Therefore, by using the determined model parameters, the target value can be calculated from the measurement signals according to a mathematical model, robust against fluctuations in characteristic values ​​different from the target value. For example, if the characteristic value is the scatterer density of a sample and the parameters of the multiple characteristic values ​​include the intensity or signal-to-noise ratio of the measurement signal, the scatterer density of the sample can be estimated independently of fluctuations in the intensity or signal-to-noise ratio. The cell nucleus density of a biological tissue sample, an example of a scatterer density, can be more accurately estimated using the measurement signals. This reduces the burden of human judgment and erroneous judgments in pathology research and diagnosis.

[0100] The training data generation unit 116 may determine the tissue structure of the sample (e.g., a tissue structure in which scatterers are distributed at a set scatterer density) based on at least one of multiple types of characteristic values ​​(e.g., scatterer density), and generate a measurement signal for a wave incident on the sample based on the determined tissue structure.

[0101] With this configuration, measurement signals are generated based on tissue structures determined using set characteristic values, making it possible to quantitatively indicate the characteristics of tissue structures under different conditions and generate a variety of measurement signals.

[0102] Alternatively, the training data generator 116 may use random numbers to determine the characteristic values ​​so that each of the multiple types falls within a predetermined range. This configuration allows for efficient generation of multiple types of randomly distributed characteristic values ​​across a large number of sets of characteristic values. Furthermore, the distribution of these characteristic values ​​can be less biased than with multiple types of artificially determined characteristic values. This allows for the acquisition of model parameters that enable calculation of the target value with higher accuracy.

[0103] Furthermore, the multiple types of characteristic values ​​may include a spatial resolution, an aberration coefficient, or a dispersion coefficient related to chromatic dispersion in the measurement system for the sample, thereby reducing or eliminating a decrease in the estimation accuracy of the characteristic values ​​due to fluctuations in the spatial resolution, the aberration coefficient, or the chromatic dispersion that depend on the measurement system itself or the sample to be measured.

[0104] The mathematical model also includes a neural network including an input layer that outputs signal values ​​for each observation point of an input measurement signal to a first intermediate layer, multiple intermediate layers, and an output layer that outputs an estimated value based on the output value input from the last intermediate layer. The multiple intermediate layers may be configured by repeatedly stacking one or more convolutional layers that perform convolution processing on input values ​​input from the previous layer and output calculated values ​​obtained from the next layer. This configuration allows features that contribute to the fluctuation of the target value to be extracted from the input measurement signal with each layer, thereby improving the accuracy of estimating the target value. Furthermore, at least one or more final layers including the output layer may be fully connected layers that output calculated values ​​obtained by performing convolution processing on all of the multiple input values ​​input from the previous layer using parameter sets whose number is less than the number of the input values. In this way, the degrees of freedom can be reduced while fully considering all components that significantly affect the characteristic value provided from the previous layer, thereby ultimately deriving the characteristic value.

[0105] The measurement signal processing device 100 may include a characteristic estimating unit 120 that calculates an estimate of the target value for the measurement signal using the model parameters based on the mathematical model. With this configuration, the measurement signal processing device 100 can calculate an estimate of the target value from the input measurement signal with high accuracy using the model parameters acquired by the device itself.

[0106] The embodiments of the present invention have been described in detail above with reference to the drawings, but the specific configuration is not limited to that described above, and various design changes can be made within the scope of the gist of the present invention.

[0107] For example, the measurement signal processing device 100 may be realized as a part of the OCT system 1, or may be a single device independent of the optical system. In this case, the measurement system control unit 112 may be omitted from the control unit 110 of the measurement signal processing device 100. The measurement signal acquisition unit 114 is not limited to the optical system, and may acquire measurement signals from other devices such as a data storage device or a PC.

[0108] The measurement signal processing device 100 may also include a display unit 150 and an operation unit 160, or one or both of them may be omitted. The measurement signal processing device 100 may also omit one or both of the characteristic estimation unit 120 and the output processing unit 122. When the characteristic estimation unit 120 is omitted, the model learning unit 118 may output the calculated model parameters to another device, such as a data storage device, a PC, or another measurement signal processing device. The output destination device may have the same function as the characteristic estimation unit 120, that is, the function of calculating characteristic values ​​for the measurement signal according to a predetermined mathematical model using the model parameters acquired from the measurement signal processing device 100.

[0109] In the above description, the training data generator 116 generates, as information indicating the tissue structure, a scatterer distribution f(x, y, z) that indicates a situation in which scatterers are stochastically and randomly distributed within the observation region at a scatterer density given as a characteristic value. However, this is not limiting. The training data generator 116 may acquire tissue structure information indicating the tissue structure of the sample from another device, and the tissue structure information may include multiple tissue structures and known characteristic values ​​indicating the characteristics (e.g., SPUV, ENS, etc.) possessed by each tissue structure. When acquiring tissue structure information, the training data generator 116 does not need to determine the tissue structure of the sample, and these tissue structures and characteristic values ​​may be used as part of a characteristic value set. The tissue structure may be a structure in which scatterers are regularly arranged within the observation region, for example, a lattice spatially arranged at a predetermined period. Furthermore, the scattering intensity or scattering cross section of each scatterer may be included in the characteristic value set as a characteristic value related to the tissue structure.

[0110] Although a case where a CNN is used as the mathematical model has been described as an example, this is not limiting. The mathematical model may be other types of neural networks, such as a recurrent neural network or a probabilistic neural network. Other types of mathematical models, such as a random forest or multiple regression model, may be used as long as they can acquire the relationship between the measurement signal and the target value. A multiple regression model is preferable as long as the relationship between the measurement signal as the input value and the target value as the output value is linear or can be approximated to a linear relationship. Furthermore, the activation function in the CNN is not limited to the normalized linear unit, and a sigmoid function, a softmax function, or the like may be used. The model error index is not limited to the MSE, and cross entropy, or the like, may be used.

[0111] Furthermore, part or all of the measurement signal processing device 100 in the above-described embodiment may be realized as an integrated circuit such as an LSI (Large Scale Integration). Each functional block of the measurement signal processing device 100 may be individually implemented as a processor, or part or all of the blocks may be integrated into a processor. The integrated circuit implementation method is not limited to LSI, and may be implemented using a dedicated circuit or a general-purpose processor. Furthermore, if an integrated circuit implementation technology that can replace LSI emerges due to advances in semiconductor technology, an integrated circuit based on that technology may be used.

[0112] [Embodiment 2] Another embodiment of the present invention will be described below. For ease of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0113] OCT signals inevitably produce a pattern called speckle. It has been known that the contrast of this speckle is related to the number of scatterers within a range known as the "coherence volume of the OCT probe light." This coherence volume is roughly a volume in three-dimensional space determined by the resolution of the OCT. This suggests that if the OCT resolution is known, the scatterer density (number of scatterers per unit volume) can be estimated from the speckle contrast. However, because the resolution of OCT varies depending on the structure of the sample itself, it is impossible to know in advance the resolution when actually measuring a given tissue.

[0114] On the other hand, conventional techniques (such as those in Patent No. 6480104) allow for the estimation of the actual resolution within tissue by analyzing the speckle pattern. These findings suggest that the scattering density within a sample can be determined by appropriately analyzing the speckle pattern. However, in practice, there is still room for improvement in the combined method described above due to various external factors, such as the signal-to-noise ratio of the image and the influence of aberrations.

[0115] In Patent Document 1, optical wave simulation using Fourier optics is used to generate "OCT patterns (speckle patterns) generated when measuring samples with various scatterer densities," and these patterns are used to train a DCNN. This DCNN successfully evaluated the scatterer density in biological samples from OCT images actually obtained from the biological specimen. However, at the same time, this conventional technique has been found to have a systematic bias in the estimated scatterer density when the signal-to-noise ratio is low (Prior Art: Thitiya Seesan, Ibrahim Abd El-Sadek, Pradipta Mukherjee, Lida Zhu, Kensuke Oikawa, Arata Miyazawa, Larina Tzu-Wei Shen, Satoshi Matsusaka, Prathan Buranasiri, Shuichi Makita, and Yoshiaki Yasuno, "Deep convolutional neural network-based scatterer density and resolution estimators in optical coherence tomography," Biomed. Opt. Express 13, 168-183 (2022)).

[0116] In this embodiment, in order to avoid the influence of this systematic bias, a method for generating a pseudo optical interference signal using a new noise model and a method for training a machine model using the pseudo optical interference signal will be described.

[0117] According to the inventors' investigation, the systematic bias in the estimated scatterer density when the signal-to-noise ratio of noise is low is thought to be caused by not taking into account the correlation of noise depending on the incident direction of the measurement target (measurement pixel). Therefore, a new noise model was constructed that takes into account the correlation of noise depending on the incident direction.

[0118] (Noise model reflecting noise characteristics) This noise model includes a shot noise component, a relative intensity noise component, and a non-optical noise component in a local region. Based on this noise model, the following equation (11) expresses the total noise contained in the optical interference signal in the wavenumber domain.

[0119]

[0120] In equation (11), k represents the wave number of light. Na(k) on the left side represents total noise (Gaussian noise). The first term on the right side represents the shot noise (light source noise) component, Nsh(k) is the shot noise, S(k) is the spatial spectral intensity distribution of the light source, and c1 is a coefficient. The second term on the right side represents the relative intensity noise (noise due to fluctuations in light intensity) component, NRIN(k) is the relative intensity noise, and c2 is a coefficient. The third term on the right side represents the non-optical noise component, Ndet(k) is the non-optical noise, and c3 is a coefficient. Nsh(k), NRIN(k), and Ndet(k) all represent random variables that follow a normal distribution with mean zero.

[0121] As can be seen from equation (11), the shot noise component is a noise component obtained by multiplying a random variable following a normal distribution with zero mean by a coefficient including the square root of the spatial spectral intensity distribution of the light source. Therefore, it varies in proportion to the square root of the spectral intensity distribution. The relative intensity noise is a noise component obtained by multiplying a random variable following a normal distribution with zero mean by a coefficient including the spatial spectral intensity distribution of the light source. Therefore, it varies in proportion to the spectral intensity distribution. The non-optical noise is a noise component obtained by multiplying a random variable following a normal distribution with zero mean by a coefficient not including the spatial spectral intensity distribution of the light source. Therefore, it is unrelated to the spectral intensity distribution. Equation (12), which represents the total noise in the depth direction, can be obtained by Fourier transforming equation (11). The "depth direction" refers to the direction in which light is incident. The direction perpendicular to the depth direction is conveniently referred to as the "lateral direction." Equation (11) is derived based on the physical laws of optical interference. However, equation (11) represents the noise superimposed on the signal obtained by SD-OCT (Spectral-Domain OCT, a method of obtaining an interference signal for each wavelength using a spectrometer using a broadband wavelength light source, i.e., a light source that can simultaneously emit light of a wide band of wavelengths). F and the tilde (symbol "~") in equation (12) represent Fourier transform. * in equation (12) represents a convolution operation. The Fourier transform is an operation performed to obtain an OCT signal. Note that F[S(k)] in equation (12) corresponds to the resolution.

[0122]

[0123] Table 1 compares the spatial correlation distance of each noise component expressed by equation (12) with that of the prior art. The left column of Table 1 lists each noise component, the middle column lists the noise model of the prior art, and the right column lists the noise model of this embodiment. Such a noise model makes it possible to generate pseudo-noise that is closer to reality.

[0124]

[0125] (Method for generating optical interference signal) Next, a method for generating an artificial optical interference signal using the above noise model will be described. The artificial optical interference signal is an optical interference signal derived by program calculation, which indicates the intensity of scattered light predicted when light is incident on a pseudo scatterer (pseudo scatterer). In the following description, unless otherwise specified, the artificial optical interference signal will be simply referred to as an "optical interference signal." The optical interference signal includes a pseudo optical interference signal predicted from the physical principles of optical interference and pseudo noise contained therein. In this embodiment, pseudo noise based on the above noise model is used as the pseudo noise.

[0126] 6 is a flowchart showing the flow of an optical interference signal generation method S3 for generating an optical interference signal. As shown in the figure, the optical interference signal generation method S3 includes steps S31 to S34. Step S31 is a characteristic value setting step for setting characteristic values ​​of pseudo-scatterers. Step S31 may be executed by a characteristic value setting unit 211, which will be described later. The pseudo-scatterers are virtual scatterers for which at least one characteristic value of density, signal-to-noise ratio, spatial resolution (depth direction and / or lateral direction), speckle contrast, and effective number of scatterers is set. Therefore, the characteristic value to be set is at least one of the density, signal-to-noise ratio, spatial resolution (depth direction and / or lateral direction), speckle contrast, and effective number of scatterers of the pseudo-scatterers. These characteristic values ​​affect the measured optical interference signal. Some of these characteristic values ​​are interrelated. For example, density and effective number of scatterers are interrelated, and signal-to-noise ratio and spatial resolution are interrelated.

[0127] For example, when setting the density, the coordinate values ​​of the individual scatterers are set based on uniform random numbers as described in embodiment 1. The speckle contrast is the intensity distribution of scattered light including noise.

[0128] Step S32 is a signal generation step for generating a pseudo optical interference signal from the pseudo scatterer having the set characteristic value. Step S32 may be executed by the signal generation unit 212, which will be described later. As described above, the pseudo optical interference signal is an optical interference signal predicted from the physical principles of optical interference.

[0129] Step S33 is a noise generation step in which pseudo-noise is generated using a noise model representing noise in the pseudo-optical interference signal. Step S33 may be performed by the noise generator 213, which will be described later. The above-described noise model is applied to each local region of the target to generate pseudo-noise. Then, by superimposing the pseudo-noise on the pseudo-optical interference signal of that local region, an artificial optical interference signal that simulates the optical interference signal actually measured can be generated. A local region is a region that is considered to have a uniform composition. For example, it is preferable that the local region does not include a boundary where the tissue composition changes significantly. In a typical OCT measurement device, the local region is, for example, a region of 16 pixels by 16 pixels in the depth direction and lateral direction.

[0130] Step S34 is a superposition step in which the pseudo optical interference signal and the pseudo noise are superposed. Step S34 may be executed by the superposition unit 214, which will be described later. This makes it possible to generate an optical interference signal containing pseudo noise that accurately reflects the characteristics of actual noise. Furthermore, by training a machine model using the generated optical interference signal, it is possible to generate a machine model that can accurately estimate characteristic values.

[0131] The optical interference signal generating method S3 may further include step S35. Step S35 is an image generating step of generating an image of a pseudo speckle pattern from a signal obtained by superimposing a pseudo optical interference signal and pseudo noise. Step S35 may be executed by the image generating unit 215 described later. The generated image accurately reflects the characteristics of the noise contained in the actually measured OCT signal, and therefore, by using it as training data for training a mechanical model described later, it is possible to generate a mechanical model capable of accurately estimating characteristic values.

[0132] (Optical Interference Signal Generating Apparatus) Next, an apparatus for generating an optical interference signal using the above-described noise model will be described. Fig. 7 is a block diagram showing the configuration of an optical interference signal generating apparatus 200 according to this embodiment. The optical interference signal generating apparatus 200 includes a control unit 210, a storage unit 220, an input / output unit 230, a display unit 240, and an operation unit 250. The storage unit 220, the input / output unit 230, the display unit 240, and the operation unit 250 are similar to the storage unit 130, the input / output unit 140, the display unit 150, and the operation unit 160 described in the first embodiment, and therefore description thereof will be omitted.

[0133] The control unit 210 includes a characteristic value setting unit 211, a signal generating unit 212, a noise generating unit 213, a superimposing unit 214, and an output processing unit 216. The control unit 210 may also include an image generating unit 215.

[0134] The characteristic value setting unit 211 sets at least one characteristic value from among the density of pseudo-scatterers, the signal-to-noise ratio, the spatial resolution, the speckle contrast, and the number of effective scatterers. Specifically, the user inputs each characteristic value on an input screen. For example, the user sets the density of scatterers and inputs it via the operation unit 250. The characteristic value setting unit 211 acquires the density input by the user and stores it in the storage unit 220. Note that the storage unit 220 stores data on each characteristic value. When an optical interference signal and pseudo-noise are generated, they are generated using one or more of these characteristic values. If the user does not set a new value, the already stored data is used. Alternatively, the user may specify characteristic values ​​to be used in the calculation or characteristic values ​​not to be used.

[0135] The signal generator 212 generates a pseudo optical interference signal from a pseudo scatterer having a set characteristic value, identifies the structure of the pseudo target scatterer according to the characteristic value stored in the memory 220, and calculates the intensity of scattered light generated by optical interference by referring to the intensity spectrum distribution of the light source.

[0136] The noise generator 213 generates pseudo noise using a noise model that represents noise in the pseudo optical interference signal. As the noise model, the calculation formula expressed by Equation (12) is stored in the storage unit 220.

[0137] The superimposing unit 214 superimposes the pseudo optical interference signal and the pseudo noise. As a result, an artificial optical interference signal (OCT signal) from the pseudo scatterer is generated. This optical interference signal is stored in the storage unit 220 together with the set characteristic values.

[0138] The image generator 215 generates an image of a pseudo speckle pattern from a signal obtained by superimposing the pseudo optical interference signal and the pseudo noise. The generated image data is stored in the storage unit 220. The optical interference signal and / or the image data stored in the storage unit 220 is used to train a machine model.

[0139] The optical interference signal generating device 200 having the above configuration can generate an optical interference signal containing pseudo-noise that accurately reflects the characteristics of actual noise. Furthermore, as will be described later, by training a machine model using the generated optical interference signal, it is possible to generate a machine model that can accurately estimate characteristic values.

[0140] The image generation unit 215 generates an image of a pseudo speckle pattern from a signal obtained by superimposing the pseudo optical interference signal and pseudo noise. By using the generated image as training data for training a mechanical model (described later), it is possible to generate a mechanical model capable of accurately estimating characteristic values. The specific details of the information processing executed by each of the above units are as described in the optical interference signal generation method S3.

[0141] The output processing unit 216 controls the output of various data based on an operation signal input from the operation unit 250. For example, the output processing unit 216 generates data displaying a list of characteristic values ​​input by the user, data displaying an image of a pseudo speckle pattern, etc. in response to a user operation, and outputs the data to the display unit 240.

[0142] The optical interference signal generation method and device described above used the noise model for the SD-OCT system. However, the same noise model as equation (11) also applies to the swept-source OCT (SS-OCT) system, which uses a wavelength-swept light source (a light source that emits broadband wavelength light by electronically or mechanically changing the wavelength of single-wavelength light such as a laser), and therefore the same noise model as that for SD-OCT can be used.

[0143] The target of the optical interference signal generated using the optical interference signal generating method S3 and the optical interference signal generating device 200 (i.e., the target for measuring the scattering body) may be any target from which coherent scattered light can be measured when light ranging from far-infrared light to visible light is incident. Such a target may be, for example, a living human or animal body, or a non-living body. The living body may be ocular tissue such as the fundus or cornea, blood vessels, teeth, subcutaneous tissue, etc. The non-living body may be an artificial structure such as an electronic component or a mechanical component (e.g., a resin containing a scattering body), a natural structure such as stone or a mineral, or a substance without a specific shape.

[0144] (Method for training a mechanical model) Next, a method for training a mechanical model using an optical interference signal containing noise using a new noise model and characteristic values ​​of a scatterer from which the optical interference signal is obtained will be described. The mechanical model to be trained is a mechanical model that receives an optical interference signal obtained by actually measuring an object as input and outputs at least one characteristic value of the object. Note that an acquired OCT image may also be used as the optical interference signal.

[0145] 8 is a flowchart showing the flow of the machine model learning method S4. As shown in the figure, the machine model learning method S4 includes steps S41 to S43.

[0146] Step S41 is a characteristic value setting step for setting a characteristic value of the pseudo-scatterer. Step S41 may be executed by the characteristic value setting unit 311, which will be described later. The characteristic value is, for example, at least one of density, signal-to-noise ratio, spatial resolution, speckle contrast, and the number of effective scatterers. The characteristic value to be set is preferably a characteristic value to be estimated (characteristic value of interest). For example, in OCT measurement of the fundus, the number of cell nuclei in a layer of interest is useful as diagnostic information. In other words, there is a demand for estimating the number of cell nuclei (i.e., cells). Therefore, it is desirable to estimate the density of scatterers corresponding to the number of cell nuclei. Therefore, density can be used as an example of the characteristic value to be set.

[0147] Step S42 is a superimposed signal generation step for generating a superimposed optical interference signal by superimposing a pseudo optical interference signal from a pseudo scatterer having a set characteristic value and a pseudo noise generated using a noise model representing the noise of the pseudo optical interference signal. Note that a synthesized OCT image may be generated as the superimposed optical interference signal. The noise model is as described above. Step S42 may be executed by the superimposed signal generation unit 312, which will be described later.

[0148] Note that steps S41 and S42 may be executed by the aforementioned optical interference signal generating apparatus 200. That is, steps S41 and S42 may be executed by the characteristic value setting unit 211, the signal generating unit 212, the noise generating unit 213, and the superimposing unit 214 of the optical interference signal generating apparatus 200.

[0149] Step S43 is a learning step in which, using the set characteristic values ​​and the generated superimposed optical interference signal, a mechanical model is trained, which takes an optical interference signal obtained by actually measuring the object as an input and outputs at least one characteristic value of the object. Step S43 may be executed by the learning unit 313, which will be described later. Note that, when an OCT image is used as the superimposed optical interference signal, the actually measured OCT image is used as the input.

[0150] By the above method, a trained machine model is generated that takes as input an optical interference signal obtained by actually measuring an object to be measured, trained using set characteristic values, a pseudo-optical interference signal from an object having the characteristic values, and an optical interference signal onto which pseudo-noise generated using a noise model representing the noise contained in the pseudo-optical interference signal is superimposed, and outputs at least one of the characteristic values ​​of the object to be measured.

[0151] The learning method is, for example, to generate optical interference signals when one characteristic value of interest is fixed and the other characteristic values ​​are changed to various values, and to define a learning set that includes the combination of characteristic values ​​and the optical interference signal. A large number of such learning sets are generated. Alternatively, a characteristic value of interest is changed and fixed, and optical interference signals are generated when the other characteristic values ​​are changed to various values, and to define a learning set that includes the combination of characteristic values ​​and the optical interference signal. A large number of such learning sets are generated. The learning sets generated in this way are trained by a machine model.

[0152] The method for training the machine model is as described in embodiment 1. In this way, it is possible to generate a machine model that accurately estimates a characteristic value of interest. For example, by training the machine model using the scatterer density as the characteristic value of interest, it is possible to generate a machine model that inputs an actually measured optical interference signal (or OCT image) and accurately estimates (outputs) the scatterer density of interest.

[0153] Furthermore, the machine model may be trained to estimate not one type of characteristic value but multiple types. For example, the machine model may be trained to estimate characteristic values ​​related to each other. The training set used for training is data generated within the range of characteristic values ​​that the measurement object typically has. The machine model may also include training data that assumes an abnormality in the measurement object.

[0154] A single machine model may be trained to estimate a single characteristic value, thereby generating a specialized machine model. Alternatively, a single machine model may be trained to estimate multiple related characteristic values ​​simultaneously or individually, thereby generating a general-purpose machine model. Furthermore, by appropriately combining the order of training, the machine model can be trained efficiently. For example, a machine model that has been sufficiently trained to estimate density may be further trained to estimate SNR.

[0155] (Machine Model Learning Apparatus) Next, a machine model learning apparatus will be described. FIG. 9 is a block diagram showing the configuration of a machine model learning apparatus 300. As shown in the figure, the learning apparatus 300 includes a control unit 310, a storage unit 320, an input / output unit 330, a display unit 340, and an operation unit 350. The learning apparatus 300 is connected to a machine model 360, which is the object of learning, so that information can be communicated therewith, for example, via the Internet. The storage unit 320, the input / output unit 330, the display unit 340, and the operation unit 350 are similar to the storage unit 130, the input / output unit 140, the display unit 150, and the operation unit 160 described in the first embodiment, and therefore description thereof will be omitted.

[0156] The control unit 310 includes a characteristic value setting unit 311, a superimposed signal generating unit 312, a learning unit 313, and an output processing unit 314. The characteristic value setting unit 311 sets the characteristic values ​​of the pseudo scatterers.

[0157] The superimposed signal generator 312 generates a superimposed optical interference signal (or image) by superimposing a pseudo optical interference signal from a pseudo scatterer having a set characteristic value and a pseudo noise generated using a noise model representing the noise of the pseudo optical interference signal. As described above, a synthesized OCT image may be generated as the superimposed optical interference signal.

[0158] The learning unit 313 uses the set characteristic values ​​and the generated superimposed optical interference signal (or image data) to train a mechanical model 360 that takes the optical interference signal measured from the object as input and outputs at least one of the characteristic values ​​of the object.

[0159] The output processing unit 314 controls the output of various data based on operation signals input from the operation unit 350. For example, the output processing unit 314 generates information such as set characteristic values, generated superimposed optical interference signals (or OCT image data), and learning progress and completion status in response to user operations, and outputs the information to the display unit 340.

[0160] The learning device 300 may be connected to the aforementioned optical interference signal generation device 200 so that information can be communicated therewith, for example, via the Internet. The characteristic value setting unit 311 may acquire characteristic values ​​set by the characteristic value setting unit 211 of the optical interference signal generation device 200. The superimposed signal generation unit 312 may acquire the optical interference signal superimposed by the superimposing unit 214 of the optical interference signal generation device 200. In other words, the learning device 300 may acquire the optical interference signal stored in the storage unit 220 of the optical interference signal generation device 200.

[0161] Next, the evaluation results of the noise model of this embodiment will be described. FIG. 10 is a graph showing the results of simulating scattered light from a pseudo target containing pseudo scatterers. The horizontal axis of the graph represents the simulated concentration of pseudo scatterers (lipids), and the vertical axis represents the density of the scatterers evaluated from the pseudo-scattered light predicted from the concentration of the pseudo scatterers. The solid dots and solid lines represent the evaluation results when the noise model of this embodiment is used, while the triangles and dotted lines represent the evaluation results when the noise model of the prior art is used. The evaluation results using the noise model of this embodiment yield a straight line passing through the origin of the graph. On the other hand, the evaluation results using the noise model of the prior art yield a straight line that does not pass through the origin of the graph. While the theoretical result should be a straight line passing through the origin of the graph, the prior art technique exhibits large errors when the concentration is low. In contrast, the graph of this embodiment yields a straight line close to the theoretical result. This indicates that the noise model of this embodiment is able to accurately simulate actual noise.

[0162] 11 is a graph showing the evaluation results of the scatterer density when using the noise model of the present embodiment and the conventional technology when assuming microparticles as pseudo-scatterers. The horizontal and vertical axes are the same as in FIG. 10, and ● and △ indicate the cases when the noise model of the present embodiment is used and the noise model of the conventional technology, respectively. As shown in the figure, when the noise model of the present embodiment is used, results close to the theoretical values ​​are obtained.

[0163] FIG. 12 is a graph showing the results of evaluating a tumor spheroid (a spherical culture sample of cancer cells) using the noise model of this embodiment. The measured sample was a human breast cancer-derived spheroid (MCF-7 spheroid). The horizontal axis of FIG. 12 represents elapsed time, and the vertical axis represents the density of the scatterers (tumor spheroid). Specifically, the average scatterer density within the spheroid was calculated from the scatterer density of the spheroid and plotted against time. The number of cells in tumor spheroids decreases over time due to the breakdown of cell nuclei caused by the progression of necrosis. In other words, the scatterer density decreases over time. As shown in FIG. 12 , the conventional technology (△) was unable to accurately estimate the decrease in scatterer density after 15 hours due to a decrease in the signal-to-noise ratio associated with the decrease in scatterer density. In contrast, the present embodiment (●) accurately observed a decrease in scatterer density up to the full evaluation time (28 hours). This demonstrates that the noise model of this embodiment can accurately analyze scattered light.

[0164] [Embodiment 3] Embodiment 3 of the present invention will be described in detail below. For ease of explanation, components having the same functions as those described in Embodiments 1 and 2 above will be denoted by the same reference numerals, and their descriptions will not be repeated. Embodiment 3 describes a method or apparatus for evaluating or diagnosing an OCT signal or OCT image acquired by an OCT device. In this embodiment, "evaluation" refers to estimating a condition that is not primarily a pathological condition, such as age-related changes, and "diagnosis" refers primarily to medical estimation of a disease, including risk diagnosis of an early stage of the disease. Hereinafter, "evaluation or diagnosis" will be simply referred to as "evaluation." Evaluation may be performed, for example, using a trained machine model. In this embodiment, a case where an OCT signal or OCT image of human eye tissue is used will be described as an example, but the subject is not limited to humans. Examples of eye tissue include the retina and cornea.

[0165] (Ocular Tissue Evaluation Device 400) FIG. 13 is a block diagram showing the configuration of the ocular tissue evaluation device 400 according to this embodiment. As shown in the figure, the ocular tissue evaluation device 400 includes a control unit 410, a storage unit 420, an input / output unit 430, a display unit 440, and an operation unit 450. Some or all of the functions of the control unit 410 may be implemented as a computer including, for example, a processor such as a CPU (Central Processing Unit) and a memory. Some or all of the control unit 410 is not limited to general-purpose hardware such as a processor, and may be configured to include dedicated hardware such as an LSI (Large Scale Integration) or an ASIC (Application Specific Integrated Circuit). The storage unit 420, the input / output unit 430, the display unit 440, and the operation unit 450 have the same functions as the storage unit, the input / output unit, the display unit, and the operation unit described in the first or second embodiment, and therefore will not be described here.

[0166] The control unit 410 includes an acquisition unit 411, a generation unit 412, a comparison unit 413, and an output processing unit 414. The acquisition unit 411 acquires an interference signal between incident light and scattered light of the incident light when light in the wavelength range from infrared light to visible light is incident on a target ocular tissue. In this embodiment, the infrared light may include mid-infrared light or near-infrared light. The interference signal may be an OCT signal or an OCT image of the ocular tissue. The acquisition unit 411 may acquire an interference signal stored in the storage unit 420. Alternatively, the acquisition unit 411 may acquire an interference signal input by a user via the operation unit 450. Alternatively, the acquisition unit 411 may acquire an interference signal transmitted from an external source via the Internet or the like via the input / output unit 430. In this case, the input / output unit 430 may have a wired or wireless information communication function. For example, the ocular tissue evaluation device 400 may be configured to be connected to a hospital or the like and receive OCT images acquired at the hospital or the like to evaluate the ocular tissue.

[0167] The generator 412 generates a local signal intensity pattern by numerically processing the local signal intensity of the interference signal. The local signal intensity pattern may be an average of the local signal intensity. The average of the local signal intensity is the so-called OCT signal intensity. The local signal intensity pattern may also be an evaluation quantity obtained by statistically processing the local signal intensity. Examples of higher-order statistical quantities include standard deviation, variance, skewness, and kurtosis. The local signal intensity pattern may also be a secondary quantity obtained by processing the local intensity pattern using some kind of program. In this case, the program may be, for example, a trained mechanical model trained using the local signal intensity or image data that visualizes the signal intensity. In this case, the signal intensity pattern is a characteristic value (e.g., scatterer density) obtained by inputting the local signal intensity or image data that visualizes the signal intensity into the trained mechanical model. Alternatively, the program may be, for example, a program that derives resolution from speckle contrast and speckle size and then calculates the scatterer density from the resolution. In this way, the secondary quantity may be, for example, the scatterer density, local resolution, etc. The secondary quantity may be, for example, any of the characteristic values ​​described in the second embodiment.

[0168] The above-mentioned mechanical model may be a mechanical model trained using a noise model including a shot noise component, a relative intensity noise component, and a non-optical noise component. As described in the second embodiment, the shot noise component for the direction of light incident on the ocular tissue is a noise component obtained by Fourier transforming a random variable following a mean normal distribution multiplied by a coefficient including the square root of the spatial spectral intensity distribution of the light source. Furthermore, the relative intensity noise component for the direction of light incident on the ocular tissue is a noise component obtained by Fourier transforming a random variable following a mean normal distribution multiplied by a coefficient including the spatial spectral intensity distribution of the light source. Furthermore, the non-optical noise component is a noise component obtained by multiplying a random variable following a mean normal distribution multiplied by a coefficient not including the spatial spectral intensity distribution of the light source.

[0169] The term "local" may refer to a region having a length of about 0.1 to 100 times the OCT resolution for each side (each direction) of the region to be measured, or a region having a length of 0.5 to 50 times, or 1 to 30 times, etc. In other words, the local signal intensity may be the signal intensity of a region in which the length in each axis direction of a three-dimensional coordinate system, with the incident direction of light as one axis, is 0.1 to 100 times, or 0.5 to 50 times, or 1 to 30 times the resolution in that axis direction.

[0170] The comparison unit 413 compares the target signal intensity pattern with a normal signal intensity pattern obtained by numerically processing the local signal intensities of interference signals obtained by irradiating light onto normal ocular tissue. The normal signal intensity pattern is the local signal intensity pattern of the ocular tissue of a normal person. A normal person refers to a person whose data obtained by examining their ocular tissue falls within a range generally considered medically normal. The comparison unit 413 compares the signal intensity pattern with the normal signal intensity pattern for each layer included in, for example, the retina or cornea.

[0171] The comparison unit 413 may, for example, compare the "amount" obtained from an individual evaluation subject (patient) with the "amount" previously obtained from a control group (normal individuals). The data obtained from the control group may be, for example, data obtained from a normal individual database. More specifically, for example, the ratio of the difference between the average of the normal individual database and the "amount" obtained from the individual patient to the standard deviation of the normal individual database may be calculated. The comparison unit 413 may test this ratio using, for example, a statistical test method (such as a t-test or a signed rank test). The comparison unit 413 may perform evaluation based on the test results.

[0172] Furthermore, the comparison unit 413 may compare, for example, a secondary quantity obtained by a trained machine model (e.g., the density of scatterers estimated by the machine model) with a corresponding secondary quantity of a normal person (normal scatterers, i.e., the density of eye cells, etc.). The trained machine model may be a machine model trained by the method described in embodiment 2.

[0173] FIG. 14 is a block diagram showing the configuration of an ocular tissue evaluation device 400A, another example of this embodiment. The ocular tissue evaluation device 400A includes a comparison unit 413A instead of the comparison unit 413 included in the ocular tissue evaluation device 400. The comparison unit 413A includes a trained machine model 4131. The comparison unit 413A inputs the local signal intensity pattern of the target generated by the generation unit 412 to the machine model 4131 and outputs a secondary quantity (e.g., estimated scatterer density). The comparison unit 413A then compares the secondary quantity (e.g., scatterer density) output by the machine model 4131 with a secondary quantity (e.g., scatterer density) of a normal person. The comparison unit 413A may further evaluate the result of the comparison as normal or abnormal. Although FIG. 14 illustrates the machine model 4131 as being incorporated into the ocular tissue evaluation device 400A, the machine model 4131 may be located in a different position from the ocular tissue evaluation device 400A. In this case, the mechanical model 4131 and the ocular tissue evaluation device 400A are connected to each other so as to be able to communicate information via the Internet or the like.

[0174] The comparison unit 413 may divide the target region into sectors. For example, the comparison unit 413 may compare the signal intensity pattern with the normal signal intensity pattern in each of the regions divided into en face sectors for the retina or cornea, or in a combined region of these. Such division may be performed for each layer of the retina or cornea. Furthermore, evaluation may be performed by combining multiple divided regions.

[0175] The comparison unit 413, 413A may perform at least one of the following: evaluating glaucoma using a signal intensity pattern of at least one of the optic nerve fiber layer (NFL), ganglion cell layer (GCL), inner plexiform layer (IPL), inner nuclear layer (INL), outer plexiform layer (OPL), and outer nuclear layer (ONL) of the retina; evaluating age-related macular degeneration using a signal intensity pattern of at least one of the INL, OPL, ONL, photoreceptor (related) layer, retinal pigment epithelium (layer), and choriocapillaris (layer); and evaluating retinitis pigmentosa using a signal intensity pattern of at least one of the INL, OPL, ONL, and photoreceptor (related) layer. Specific methods will be described later. The inner nuclear layer (INL), outer plexiform layer (OPL), and outer nuclear layer (ONL) are layers of nerve cells (second neurons) that connect photoreceptors and ganglion cells, and are layers that may experience changes in glaucoma, photoreceptor cell disease, and age-related macular degeneration.

[0176] The comparison unit 413, 413A may perform at least one of evaluation of corneal degeneration or corneal opacity, or evaluation of corneal cell density, using a signal intensity pattern of at least one of the corneal epidermis, Bowman's layer, stroma, Descemet's membrane, and corneal endothelium. A specific method for this will be described later.

[0177] The output processing unit 414 has the same functions as the output processing unit described in embodiment 1 or 2. That is, the output processing unit 414 generates a display image including various display contents based on a user instruction, and outputs and displays the image on the display unit 440. For example, the output processing unit 414 generates an input screen for accepting a user instruction, a display screen for displaying the contents generated or compared by the generation unit 412 and the comparison units 413 and 413A, and outputs the screens to the display unit 440.

[0178] The above-described configuration of the ocular tissue evaluation device 400 or ocular tissue evaluation device 400A enables the state of ocular tissue to be evaluated using OCT images with higher accuracy than conventional techniques. In particular, by using a machine model trained using a new noise model, noise signals in OCT images can be reflected with higher accuracy, enabling the state of ocular tissue to be evaluated using OCT images with higher accuracy than conventional techniques.

[0179] (Ocular tissue evaluation method) Next, an ocular tissue evaluation method using the ocular tissue evaluation device 400 or the ocular tissue evaluation device 400A will be described. Fig. 15 is a flowchart showing the flow of the ocular tissue evaluation method S5 according to this embodiment. As shown in the figure, the ocular tissue evaluation method S5 includes steps S51 to S53.

[0180] Step S51 is an acquisition step of acquiring an interference signal between incident light and scattered light of the incident light when light in the wavelength range from infrared light to visible light is incident on the target eye tissue. The acquisition step is executed by, for example, the acquisition unit 411.

[0181] Step S52 is a generating step of generating a local signal strength pattern by performing numerical processing on the local signal strength of the interference signal. The generating step is executed by the generating unit 412, for example.

[0182] Step S53 is a comparison step in which the target signal intensity pattern is compared with a normal signal intensity pattern obtained by numerically processing the local signal intensity of an interference signal obtained by irradiating light onto normal ocular tissue. The comparison step is performed by, for example, the comparison unit 413 or the comparison unit 413A. The contents performed by each of the above units are as described above.

[0183] The above-described ocular tissue evaluation method enables the state of ocular tissue to be evaluated using OCT images with higher accuracy than conventional techniques. In particular, by using a machine model trained using a new noise model, noise signals in OCT images can be reflected with higher accuracy, enabling the state of ocular tissue to be evaluated using OCT images with higher accuracy than conventional techniques.

[0184] (Evaluation of the Retina) Next, as a specific example, an example of evaluating the human retina will be described. The human retina is composed of retinal ganglion cells (GCs). These retinal ganglion cells have the following three layers: Nerve fiber layer (NFL): A layer containing the axons of GCs. Ganglion cell layer (GCL): A layer containing the cell bodies of GCs. Inner plexiform layer (IPL): A layer containing synapses (connective tissue) between GCs and other nerve cells. These three layers are collectively called the ganglion cell complex (GCC).

[0185] In conventional technology, the thickness of GCC (or NFL) can be determined by image analysis of OCT images. According to the inventors' findings, the thickness of GCC or NFL has a relatively small correlation with age. On the other hand, it has been found that the thickness of GCC or NFL has a stronger correlation with age than the correlation with age. In other words, estimating the GCC cell density makes it possible to more accurately evaluate the degree of aging of ocular tissues. Furthermore, the evaluation method according to this embodiment is believed to be capable of evaluating not only age-related changes in ocular tissues but also diseases accompanied by changes in cell number, such as glaucoma, age-related macular degeneration, and retinitis pigmentosa. Photoreceptor cell diseases such as retinitis pigmentosa can be evaluated by estimating the scatterer density in the photoreceptor (associated) layer. Furthermore, age-related macular degeneration can be evaluated by estimating the scatterer density in the photoreceptor (associated) layer, retinal pigment epithelium (layer), and choriocapillaris (layer). The scatterer density of each of these layers can be estimated using a machine model trained by the method described in embodiment 2. As described above, the signal intensity patterns of the inner nuclear layer (INL), outer plexiform layer (OPL), or outer nuclear layer (ONL) may also be used.

[0186] Below, we explain the correlation between measured OCT signal intensity and age in GCC (NFL, GCL, IPL), and the correlation between age and GCC cell density (scatterer density) estimated using a trained machine model. Figure 16 is a graph showing the correlation between measured NFL OCT signal intensity and age, and the correlation between estimated NFL cell density and age. Twenty-four data sets were analyzed in each case. The left side of the graph shows the correlation between OCT signal intensity (dB) and age (years), while the right side shows the correlation between NFL cell density (number of cells per cubic micrometer) and age (years). The correlation between OCT signal intensity and age was significant at a 5% significance level (p = 0.025) for OCT signal intensity. Furthermore, the correlation between cell density was significant at a 5% significance level (p = 0.030) for cell density. This demonstrates that both methods are capable of estimating the age of a subject's retina.

[0187] 17 is a graph showing the correlation between age and measured OCT signal intensity or estimated cell density in GCLs using a method similar to that shown in FIG. 16 . The correlation between OCT signal intensity and age was significant at a 5% significance level (p=0.025) for OCT signal intensity. Furthermore, the correlation was significant at a 0.1% significance level (p=0.0009) for cell density. It can be seen that, in the case of GCLs, retinal age can be accurately assessed by using cell density.

[0188] Figure 18 is a graph showing the correlation between age and measured OCT signal intensity or estimated cell density in IPL using a method similar to that shown in Figure 16. The correlation between OCT signal intensity and age was not significant at the 5% significance level (p=0.079) for OCT signal intensity, and was significant at the 5% significance level (p=0.037) for cell density.

[0189] FIG. 19 is a diagram showing the en face sector of the retina. As shown in the figure, a circular area of ​​approximately 6 mm in diameter in the acquired right eye fundus image is divided into eight regions. Specifically, the image is divided by 1 mm and 3 mm circumferential circles and two orthogonal radial line segments. Data within the central 1 mm circle is excluded because it is near the fovea. This is because the fovea region does not contain many GCC cells. The eight regions are, starting from the outer upper sector in a clockwise direction, the superior outer (outer upper), nasal outer (outer nasal), inferior outer (outer lower), and temporal outer (outer temporal), and starting from the inner upper sector in a clockwise direction, the superior inner (inner upper), nasal inner (inner nasal), inferior inner (inner lower), and temporal inner (inner temporal). Note that the size and division method of the en face sector regions are not limited to those described above and can be set as appropriate. For example, a method of dividing a circular region with a diameter of 6 mm into circles of 3 mm and 1 mm is a standard division method for diabetic retinopathy diagnosis.

[0190] 20 is a graph showing the correlation coefficient r between OCT signal intensity or estimated cell density and age in each sector of the NFL. As shown in the figure, estimated cell density tends to have a higher correlation with age than OCT signal intensity.

[0191] 21 is a graph showing the correlation coefficient r between OCT signal intensity or estimated cell density and age in each sector of the IPL, similar to Fig. 20. Here too, estimated cell density tends to have a higher correlation with age than OCT signal intensity.

[0192] When using the en face sector, a comprehensive evaluation may be performed based on the evaluation results for each region, or a combination of different regions may be evaluated for different diseases. For example, the entire inner ring (approximately 6 mm in diameter) corresponds to the macular region, so in the case of age-related macular degeneration, the entire inner ring may be comprehensively evaluated. Furthermore, the entire inner ring corresponds to the central part of the macular region, so it can be used to evaluate age-related macular degeneration in the central part.

[0193] (Corneal Evaluation) When evaluating the cornea, data from the epidermis, Bowman's layer, stroma, Descemet's membrane, and corneal endothelium are used. Alternatively, data from multiple layers may be combined for evaluation. The en face sector may be the same as the retina if the diameter is not limited. For example, a circle with an outer diameter of 8 mm corresponds to the size of the pupil in a dark environment. A circle with a diameter of 2 mm corresponds to the size of the pupil in a bright environment. Furthermore, as with the retina, the cornea may be divided into four sectors: upper, lower, nasal, and temporal. Examples of data to be estimated include corneal transparency and cell density. The evaluation value (estimated value) of corneal transparency can be used to evaluate corneal degeneration, corneal opacity (bacterial infection, viral infection), etc. Furthermore, the evaluation value (estimated value) of corneal cell density, particularly the evaluation value (estimated value) of corneal endothelium cell density, can be used to evaluate the degree of aging. In other words, the higher the cell density, the younger and healthier the cornea can be evaluated.

[0194] [Example of implementation by software] The functions of the optical interference signal generating device 200, the learning device 300, and the ocular tissue evaluation devices 400 and 400A (hereinafter referred to as "devices") can be realized by a program that causes a computer to function as the device, and a program that causes a computer to function as each control block of the device (particularly each part included in the control units 210, 310, 410, and 410A).

[0195] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The functions described in each of the above embodiments are realized by executing the program using the control device and storage device.

[0196] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0197] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0198] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

[0199] [Summary] (Aspect 1) An optical interference signal generating method according to one embodiment of the present invention includes a characteristic value setting step of setting characteristic values ​​of a pseudo-scatterer, a signal generating step of generating a pseudo-optical interference signal from the pseudo-scatterer having the set characteristic values, a noise generating step of generating pseudo-noise using a noise model representing noise in the pseudo-optical interference signal, and a superimposing step of superimposing the pseudo-optical interference signal and the pseudo-noise.

[0200] (Aspect 2) The optical interference signal generation method according to aspect 1, wherein the noise model includes a shot noise component in a local region, a relative intensity noise component, and a non-optical noise component.

[0201] (Aspect 3) The optical interference signal generating method according to Aspect 2, wherein the shot noise component in the direction of light incident on the pseudo-scatterer is a noise component obtained by Fourier transform of a random variable that follows a normal distribution with mean zero multiplied by a coefficient including the square root of the spatial spectral intensity distribution of the light source.

[0202] (Aspect 4) The optical interference signal generating method according to aspect 2, wherein the relative intensity noise component for the direction of light incident on the pseudo-scatterer is a noise component obtained by Fourier transforming a random variable that follows a normal distribution with mean zero multiplied by a coefficient including the spatial spectral intensity distribution of the light source.

[0203] (Aspect 5) The optical interference signal generating method according to aspect 2, wherein the non-optical noise component is a noise component obtained by multiplying a random variable following a normal distribution with a mean of zero by a coefficient that does not include the spatial spectral intensity distribution of the light source.

[0204] (Aspect 6) The optical interference signal generating method according to any one of aspects 1 to 5, further comprising an image generating step of generating an image of a pseudo speckle pattern from a signal obtained by superimposing the pseudo optical interference signal and the pseudo noise.

[0205] (Aspect 7) The optical interference signal generating method according to any one of Aspects 1 to 6, wherein the light incident on the pseudo-scatterer is wavelength swept light or broadband wavelength light.

[0206] (Aspect 8) An optical interference signal generating method according to any one of aspects 1 to 7, wherein in the characteristic value setting step, at least one of the density of the pseudo-scatterers, the signal-to-noise ratio, the spatial resolution, the speckle contrast, and the number of effective scatterers is set as the characteristic value.

[0207] (Aspect 9) An optical interference signal generating device according to one embodiment of the present invention includes a characteristic value setting unit that sets characteristic values ​​of a pseudo-scatterer, a signal generating unit that generates a pseudo-optical interference signal from the pseudo-scatterer having the set characteristic values, a noise generating unit that generates pseudo-noise using a noise model that represents noise in the pseudo-optical interference signal, and a superimposing unit that superimposes the pseudo-optical interference signal and the pseudo-noise.

[0208] (Aspect 10) A method for training a machine model according to one embodiment of the present invention includes: a characteristic value setting step of setting a characteristic value of a pseudo-scatterer; a superimposed signal generation step of generating a superimposed optical interference signal by superimposing a pseudo-optical interference signal from the pseudo-scatterer having the set characteristic value and a pseudo-noise generated using a noise model representing noise in the pseudo-optical interference signal; and a learning step of using the set characteristic value and the generated superimposed optical interference signal to train a machine model that receives an optical interference signal obtained by actually measuring an object as an input and outputs at least one of the characteristic values ​​of the object.

[0209] (Aspect 11) A machine model learning device according to one embodiment of the present invention includes: a characteristic value setting unit that sets characteristic values ​​of a pseudo-scatterer; a superimposed signal generation unit that generates a superimposed optical interference signal by superimposing a pseudo-optical interference signal from the pseudo-scatterer having the set characteristic value and a pseudo-noise generated using a noise model that represents noise in the pseudo-optical interference signal; and a learning unit that uses the set characteristic value and the generated superimposed optical interference signal to learn a machine model that receives an optical interference signal obtained by actually measuring an object as an input and outputs at least one of the characteristic values ​​of the object, using the set characteristic value and the generated superimposed optical interference signal.

[0210] (Aspect 12) A trained machine model according to one embodiment of the present invention is a trained machine model that takes as input an optical interference signal obtained by actually measuring an object to be measured, and outputs at least one of the characteristic values ​​of the object to be measured, and that has been trained using a set characteristic value, a pseudo-optical interference signal from an object having the characteristic value, and an optical interference signal onto which pseudo-noise generated using a noise model representing the noise contained in the pseudo-optical interference signal has been superimposed.

[0211] (Aspect 13) A program according to one embodiment of the present invention is a program that causes a computer to execute a characteristic value setting process that sets characteristic values ​​of a pseudo-scatterer, a signal generation process that generates a pseudo-optical interference signal from the pseudo-scatterer having the set characteristic values, a noise generation process that generates pseudo-noise using a noise model that represents noise in the pseudo-optical interference signal, and a superposition process that superimposes the pseudo-optical interference signal and the pseudo-noise.

[0212] (Aspect 14) An ocular tissue evaluation method according to one embodiment of the present invention includes: an acquisition step of acquiring an interference signal between incident light and scattered light of the incident light when light in a wavelength range from infrared light to visible light is incident on a target ocular tissue; a generation step of generating a local signal intensity pattern by numerically processing the local signal intensity of the interference signal; and a comparison step of comparing the signal intensity pattern of the target with a normal signal intensity pattern obtained by numerically processing the local signal intensity of the interference signal obtained by incidenting the light on normal ocular tissue.

[0213] (Aspect 15) The ocular tissue evaluation method according to Aspect 14, wherein the signal intensity pattern is a scatterer density of the ocular tissue obtained by inputting the local signal intensity or image data that visualizes the signal intensity into a trained machine model.

[0214] (Aspect 16) The ocular tissue evaluation method according to Aspect 14, wherein the signal intensity pattern is an evaluation quantity obtained by statistically processing the local signal intensities.

[0215] (Aspect 17) The ocular tissue evaluation method according to Aspect 15, wherein the mechanical model is a mechanical model trained using a noise model including a shot noise component, a relative intensity noise component, and a non-optical noise component.

[0216] (Aspect 18) The ocular tissue evaluation method according to Aspect 17, wherein the shot noise component for the direction of light incident on the ocular tissue is a noise component obtained by Fourier transform of a product of a random variable following a mean normal distribution and a coefficient including the square root of the spatial spectral intensity distribution of a light source, the relative intensity noise component for the direction of light incident on the ocular tissue is a noise component obtained by Fourier transform of a product of a random variable following a mean normal distribution and a coefficient including the spatial spectral intensity distribution of a light source, and the non-optical noise component is a noise component obtained by multiplying a random variable following a mean normal distribution and a coefficient including no spatial spectral intensity distribution of a light source.

[0217] (Aspect 19) The ocular tissue evaluation method according to any one of Aspects 14 to 18, wherein in the comparing step, the signal intensity pattern is compared with the normal signal intensity pattern for each layer included in the retina or cornea.

[0218] (Aspect 20) The ocular tissue evaluation method according to Aspect 19, wherein the comparison step includes at least one of evaluating glaucoma using the signal intensity pattern of at least any one of the NFL, GCL, IPL, INL, OPL, and ONL of the retina, evaluating age-related macular degeneration using the signal intensity pattern of at least any one of the INL, OPL, ONL, photoreceptor-associated layer, retinal pigment epithelium layer, and choriocapillaris layer, and evaluating retinitis pigmentosa using the signal intensity pattern of at least any one of the INL, OPL, ONL, and photoreceptor-associated layer.

[0219] (Aspect 21) The ocular tissue evaluation method according to Aspect 19, wherein in the comparison step, the signal intensity pattern of at least one of the corneal epidermis, Bowman's layer, stroma, Descemet's membrane, and corneal endothelium is used to evaluate corneal degeneration or corneal opacity, or to evaluate corneal cell density.

[0220] (Aspect 22) The ocular tissue evaluation method according to Aspect 20 or 21, wherein in the comparison step, the signal intensity pattern is compared with the normal signal intensity pattern in each of the en face sectors into which the retina or the cornea is divided, or in a combined area thereof.

[0221] (Aspect 23) The ocular tissue evaluation method according to any one of Aspects 14 to 22, wherein the local signal intensity is the signal intensity of a region in which the length in each axis direction of a three-dimensional coordinate system, with the incident direction of the light as one axis direction, is 0.1 to 100 times the resolution in that axis direction.

[0222] (Aspect 24) An ocular tissue evaluation device according to one embodiment of the present embodiment includes an acquisition unit that acquires an interference signal between incident light and scattered light of the incident light when light in a wavelength range from infrared light to visible light is incident on a target ocular tissue, a generation unit that performs numerical processing on local signal intensities of the interference signal to generate a local signal intensity pattern, and a comparison unit that compares the signal intensity pattern of the target with a normal signal intensity pattern that is obtained by numerically processing local signal intensities of the interference signal obtained by incidenting the light on normal ocular tissue.

[0223] (Aspect 25) A program according to one embodiment of this embodiment is a program that causes a computer to execute an acquisition process of acquiring an interference signal between incident light and scattered light of the incident light when light in a wavelength range from infrared light to visible light is incident on a target ocular tissue; a generation process of generating a local signal intensity pattern by numerically processing the local signal intensity of the interference signal; and a comparison process of comparing the signal intensity pattern of the target with a normal signal intensity pattern obtained by numerically processing the local signal intensity of the interference signal obtained by incidenting the light on normal ocular tissue.

[0224] 1...OCT system, 10...light source, 20...beam splitter, 30a, 30b, 50a, 50b...collimator, 40...reference mirror, 60a, 60b...galvanometer mirror, 70...spectroscope, 100...measurement signal processing device, 110...controller, 112...measurement system control unit, 114...measurement signal acquisition unit, 116...learning data generation unit, 118...model learning unit, 120...characteristic estimation unit, 122, 216, 314, 414...output processing unit, 130, 220, 320, 420...storage unit, 140, 230, 330, 430...input / output unit, 150 , 240, 340, 440...display unit, 160, 250, 350, 450...operation unit, 200...optical interference signal generating device, 210...control unit, 211...characteristic value setting unit, 212...signal generating unit, 213...noise generating unit, 214...superimposing unit, 215...image generating unit, 300...learning device, 310...control unit, 311...characteristic value setting unit, 312...superimposed signal generating unit, 313...learning unit, 360, 4131...machine model, 410...ocular tissue evaluation device, 410, 410A...control unit, 411...acquisition unit, 412...generation unit, 413, 413A...comparison unit

Claims

1. A characteristic value setting step of setting characteristic values of a pseudo-scattering body, a signal generation step of generating a pseudo-optical interference signal from the pseudo-scattering body having the set characteristic values, a noise generation step of generating pseudo-noise using a noise model representing the noise of the pseudo-optical interference signal, and a superimposing step of superimposing the pseudo-optical interference signal and the pseudo-noise. A method for generating an optical interference signal.

2. The method for generating an optical interference signal according to claim 1, wherein the noise model includes a shot noise component, a relative intensity noise component, and a non-optical noise component in a local region.

3. The shot noise component in the direction of the light incident on the pseudo-scattering body is a noise component obtained by Fourier-transforming a product of a coefficient including the square root of the spatial spectral intensity distribution of the light source and a random variable following a normal distribution with a mean of zero. The method for generating an optical interference signal according to claim 2.

4. The relative intensity noise component in the direction of the light incident on the pseudo-scattering body is a noise component obtained by Fourier-transforming a product of a coefficient including the spatial spectral intensity distribution of the light source and a random variable following a normal distribution with a mean of zero. The method for generating an optical interference signal according to claim 2.

5. The non-optical noise component is a noise component obtained by multiplying a random variable following a normal distribution with a mean of zero by a coefficient not including the spatial spectral intensity distribution of the light source. The method for generating an optical interference signal according to claim 2.

6. The method for generating an optical interference signal according to any one of claims 1 to 5, further including an image generation step of generating an image of a pseudo-speckle pattern from a signal obtained by superimposing the pseudo-optical interference signal and the pseudo-noise.

7. The method for generating an optical interference signal according to any one of claims 1 to 5, wherein the light incident on the pseudo-scattering body is wavelength-swept light or broadband wavelength light.

8. In the characteristic value setting step, at least one of the density of the pseudo-scattering body, the signal-to-noise ratio, the spatial resolution, the speckle contrast, and the effective number of scattering bodies is set as the characteristic value. The method for generating an optical interference signal according to any one of claims 1 to 5.

9. An optical interference signal generation device comprising: a characteristic value setting unit that sets characteristic values of a pseudo-scattering body; a signal generation unit that generates a pseudo-optical interference signal from the pseudo-scattering body having the set characteristic values; a noise generation unit that generates pseudo-noise using a noise model representing the noise of the pseudo-optical interference signal; and a superimposing unit that superimposes the pseudo-optical interference signal and the pseudo-noise.

10. A method for learning a machine model, including: a characteristic value setting step of setting characteristic values of a pseudo-scattering body; a superimposed signal generation step of generating a superimposed optical interference signal by superimposing the pseudo-optical interference signal from the pseudo-scattering body having the set characteristic values and the pseudo-noise generated using a noise model representing the noise of the pseudo-optical interference signal; and a learning step of learning a machine model that takes, as an input, an optical interference signal actually measured for an object and outputs at least one of the characteristic values of the object, using the set characteristic values and the generated superimposed optical interference signal.

11. A machine model learning device comprising: a characteristic value setting unit that sets characteristic values of a pseudo-scattering body; a superimposed signal generation unit that generates a superimposed optical interference signal by superimposing the pseudo-optical interference signal from the pseudo-scattering body having the set characteristic values and the pseudo-noise generated using a noise model representing the noise of the pseudo-optical interference signal; and a learning unit that learns a machine model that takes, as an input, an optical interference signal actually measured for an object and outputs at least one of the characteristic values of the object, using the set characteristic values and the generated superimposed optical interference signal.

12. A learned machine model that is learned using an optical interference signal obtained by superimposing set characteristic values, a pseudo-optical interference signal from an object having the characteristic values, and pseudo-noise generated using a noise model representing the noise included in the pseudo-optical interference signal, and that takes, as an input, an optical interference signal actually measured for a measurement object and outputs at least one of the characteristic values of the measurement object.

13. A program for causing a computer to execute: a characteristic value setting process of setting characteristic values of a pseudo-scattering body; a signal generation process of generating a pseudo-optical interference signal from the pseudo-scattering body having the set characteristic values; a noise generation process of generating pseudo-noise using a noise model representing the noise of the pseudo-optical interference signal; and a superimposing process of superimposing the pseudo-optical interference signal and the pseudo-noise.

14. An acquisition step of acquiring an interference signal between incident light and scattered light of the incident light when light in a wavelength region from infrared light to visible light is incident on a target eye tissue; a generation step of numerically processing local signal intensities of the interference signal to generate a local signal intensity pattern; and a comparison step of comparing the signal intensity pattern of the target with a normal signal intensity pattern obtained by numerically processing local signal intensities of the interference signal obtained by irradiating a normal eye tissue with the light. An eye tissue evaluation method comprising the steps.

15. The eye tissue evaluation method according to claim 14, wherein the signal intensity pattern is a scatterer density of the eye tissue obtained by inputting the local signal intensity or image data obtained by imaging the signal intensity into a trained machine model.

16. The eye tissue evaluation method according to claim 14, wherein the signal intensity pattern is an evaluation quantity obtained by statistically processing the local signal intensity.

17. The eye tissue evaluation method according to claim 15, wherein the machine model is a machine model trained using a noise model including a shot noise component, a relative intensity noise component, and a non-optical noise component.

18. The shot noise component in the direction of the light incident on the eye tissue is a noise component obtained by Fourier-transforming a product of a coefficient including the square root of the spatial spectral intensity distribution of the light source and a random variable following a normal distribution with a mean of zero. The relative intensity noise component in the direction of the light incident on the eye tissue is a noise component obtained by Fourier-transforming a product of a coefficient including the spatial spectral intensity distribution of the light source and a random variable following a normal distribution with a mean of zero. The non-optical noise component is a noise component obtained by multiplying a random variable following a normal distribution with a mean of zero by a coefficient not including the spatial spectral intensity distribution of the light source. The eye tissue evaluation method according to claim 17.

19. The eye tissue evaluation method according to any one of claims 14 to 18, wherein in the comparison step, the signal intensity pattern is compared with the normal signal intensity pattern for each layer included in the retina or the cornea.

20. In the comparison step, performing at least any one of: evaluating glaucoma using the signal intensity pattern of at least any one of NFL, GCL, IPL, INL, OPL, and ONL of the retina; evaluating age-related macular degeneration using the signal intensity pattern of at least any one of INL, OPL, ONL, the photoreceptor cell-related layer, the retinal pigment epithelium layer, and the choroidal capillary lamina; and evaluating retinitis pigmentosa using the signal intensity pattern of at least any one of INL, OPL, ONL, and the photoreceptor cell-related layer. The method for evaluating eye tissue according to claim 19.

21. In the comparison step, performing at least any one of: evaluating corneal degeneration or corneal opacity using the signal intensity pattern of at least any one of the corneal epithelium, Bowman's layer, stroma, Descemet's membrane, and corneal endothelium; or evaluating the cell density of the cornea. The method for evaluating eye tissue according to claim 19.

22. In the comparison step, comparing the signal intensity pattern and the normal signal intensity pattern in each en face sector obtained by dividing the region of the retina or the cornea, or in a composite region thereof. The method for evaluating eye tissue according to claim 20.

23. The local signal intensity is the signal intensity of a region in which the length of each axis direction of a three-dimensional coordinate having the incident direction of the light as one axis direction has a length from 0.1 times to 100 times the resolution of the axis direction. The method for evaluating eye tissue according to any one of claims 14 to 18.

24. An eye tissue evaluation apparatus comprising: an acquisition unit that acquires an interference signal between incident light and scattered light of the incident light when light in a wavelength region from infrared light to visible light is incident on a target eye tissue; a generation unit that numerically processes the local signal intensity of the interference signal to generate a local signal intensity pattern; and a comparison unit that compares the signal intensity pattern of the target with a normal signal intensity pattern obtained by numerically processing the local signal intensity of the interference signal obtained by incident the light on a normal eye tissue.

25. A program that causes a computer to execute an acquisition process of acquiring an interference signal between incident light and scattered light of the incident light when light in a wavelength region from infrared light to visible light is incident on a target eye tissue, a generation process of numerically processing a local signal intensity of the interference signal to generate a local signal intensity pattern, and a comparison process of comparing the signal intensity pattern of the target with a normal signal intensity pattern obtained by numerically processing a local signal intensity of the interference signal obtained by irradiating the normal eye tissue with the light.

Images