JP6900647B2 - Ophthalmic device and IOL power determination program - Google Patents

Description

The present disclosure relates to an ophthalmic device for determining the power of an IOL to be inserted into an eye to be inspected, and an IOL power determination program.

In cataract surgery, an ophthalmic device that determines the power (hereinafter, also referred to as power) of an intraocular lens (IOL) inserted into the eye to be inspected after removal of the crystalline lens nucleus is known (Patent Document 1). reference). In such a device, in order to determine the intraocular lens power, for example, a regression formula or a theoretical formula is used to calculate the IOL power.

Japanese Unexamined Patent Publication No. 2011-206368

However, it cannot always be said that an appropriate IOL frequency can be obtained by the conventional calculation formula. For example, in an eye having a non-average shape, the error of IOL power may be large.

The technical subject of the present disclosure is to provide an ophthalmic apparatus for calculating an appropriate intraocular lens power and an IOL power determination program in view of the conventional problems.

In order to solve the above problems, the present disclosure is characterized by having the following configurations.

(1) An ophthalmological device for determining the power of an IOL to be inserted into an eye to be inspected, which includes a parameter acquisition means for acquiring a plurality of eye shape parameters of the eye to be inspected and an arithmetic control means for calculating the IOL power. By inputting the plurality of eye shape parameters into a mathematical model trained by a machine learning algorithm, the arithmetic control means obtains selection information for selecting an appropriate formula from a plurality of IOL frequency calculation formulas. An ophthalmic device characterized by outputting to a model.
(2) An IOL power determination program executed in an ophthalmic apparatus for determining the power of an IOL to be inserted into an eye to be inspected, and is executed by a processor of the ophthalmic apparatus to perform a plurality of eye shape parameters of the eye to be inspected. By inputting the plurality of eye shape parameters into the mathematical model trained by the machine learning algorithm and the parameter acquisition step of acquiring the above, the selection information for selecting an appropriate formula from the plurality of IOL frequency calculation formulas is provided. It is characterized in that the ophthalmic apparatus executes an arithmetic step to be output to a mathematical model.

It is a schematic block diagram explaining the structure of the ophthalmologic imaging apparatus which concerns on this Example. It is a figure which shows an example of a mathematical model. It is a figure which shows the flowchart of the control operation. It is a figure which shows the front eye part observation screen which displayed the image | image of the anterior eye part. It is a figure which shows an example of the cross-sectional image of the anterior segment. It is a figure for demonstrating the calculation of the expected ELP. It is a figure for demonstrating the calculation of the expected ELP.

<Embodiment>
The ophthalmic apparatus of this embodiment will be described. The ophthalmic apparatus (eg, ophthalmic apparatus 10) determines the power of the intraocular lens (also referred to as IOL) to be inserted into the eye to be inspected. This apparatus mainly includes, for example, a parameter acquisition unit (for example, an OCT device 5 or a control unit 80 or an operation unit 84) and an arithmetic control unit (for example, a control unit 80). The parameter acquisition unit acquires, for example, a plurality of eye shape parameters of the eye to be inspected. The ocular shape parameters may be, for example, corneal shape parameters such as anterior corneal curvature, posterior corneal curvature, corneal thickness, corneal width, and corneal height, and anterior crystalline lens curvature, posterior crystalline lens curvature, crystalline lens thickness, and crystalline lens diameter. It may be a crystalline lens shape parameter such as anterior atrioventricular depth, an overall shape parameter such as axial length, corneal angle, corneal distance, and pupil diameter, or a retinal shape parameter such as retinal thickness. You may. The curvature of the crystalline lens may be either contracted or relaxed, or both curvatures may be used. The parameter acquisition unit acquires eye shape parameters based on, for example, a tomographic image taken by the OCT optical system. The parameter acquisition unit may acquire the eye shape parameter of the subject from a server or the like outside the device via the communication network, or may acquire the eye shape parameter by input from the operation unit.

The arithmetic control unit calculates IOL-related information using a machine learning algorithm. The machine learning algorithm is, for example, a machine learning algorithm is, for example, a neural network, a random forest, boosting, or the like.

Neural networks are a technique that mimics the behavior of biological nerve cell networks. Neural networks include, for example, feed-forward neural networks, RBF networks (radial basis functions), spiking neural networks, convolutional neural networks, recursive neural networks (recurrent neural networks, feedback neural networks, etc.), and probabilistic neural networks. Neural nets (Boltzmann machines, Basian networks, etc.).

Boosting is a method of generating a strong classifier by combining a plurality of weak classifiers. A strong classifier is constructed by sequentially learning simple and weak classifiers.

Random forest is a method of learning to generate a large number of decision trees based on randomly sampled training data. When using a random forest, the branches of a plurality of decision trees learned in advance as a discriminator are traced, and the results obtained from each decision tree are averaged (or majority voted).

For example, the arithmetic control unit acquires IOL-related information using a mathematical model trained by a machine learning algorithm (for example, mathematical model 800). A mathematical model refers, for example, to a data structure for predicting the relationship between input data and output data. Mathematical models are constructed by training with training datasets. The training data set is a set of training data for input and training data for output. The training data for input is sample data input to the mathematical model. For example, previously measured eye shape parameters are used for the input training data. The training data for output is sample data of values predicted by a mathematical model. For example, the postoperative anterior chamber depth or IOL frequency is used as the output training data. Postoperative anterior chamber depth is the distance from the cornea to the IOL. The mathematical model is trained to output the corresponding output training data when certain input training data is input. For example, training updates the correlation data (eg, weights) between each input and output. The arithmetic control unit calculates IOL-related information based on the correlation data.

For example, the arithmetic control unit may acquire the output of the predicted postoperative anterior chamber depth as IOL-related information from the mathematical model. In this case, the trained mathematical model is a mathematical model that outputs the expected postoperative anterior chamber depth by inputting eye shape parameters including at least one of lens anterior curvature, lens posterior curvature, lens thickness, and anterior chamber depth. May be good. For example, two lenses, the anterior lens curvature and the posterior lens curvature, may be used as inputs to the mathematical model that outputs the expected postoperative anterior atrioventricular depth, or the anterior lens curvature, the posterior lens surface curvature, and the anterior lens depth may be used. It may be used, or four lenses, the front lens curvature, the rear lens surface curvature, the anterior atrioventricular depth, and the lens thickness, may be used. In addition, at least two of the anterior segment anterior shape parameter (eg, anterior curvature) and the anterior segment posterior surface shape parameter (eg, posterior curvature) acquired by the anterior segment cross-sectional image device are used as inputs to the mathematical model. Then, as the output of the mathematical model, the offset distance from the front surface of the crystalline lens to the contact point between the zonule of Zinn and the crystalline lens may be acquired as IOL-related information. The expected postoperative anterior chamber depth may be acquired by adding the acquired offset distance to the anterior chamber depth. In this case, the lens thickness may be added as an input to the mathematical model.

For example, the arithmetic control unit may calculate the IOL frequency based on the predicted postoperative anterior chamber depth output by the mathematical model. For example, the arithmetic control unit obtains the IOL power by substituting the predicted postoperative anterior chamber depth and other eye shape parameters into the IOL power calculation formula. For example, the arithmetic control unit is a prediction technique calculated by a mathematical model at the relevant part of an existing IOL frequency calculation formula such as SRK / T formula, Binkhorst formula, Holladay formula, Holladay2 formula, HofferQ formula, Olsen formula, Haigis formula, etc. The IOL frequency may be calculated by substituting the depth of the posterior chamber. By using machine learning in this way, it is possible to calculate an appropriate IOL power for an eye to be inspected of various shapes.

The arithmetic control unit may output the IOL frequency as IOL-related information from the mathematical model by inputting a plurality of eye shape parameters into the mathematical model. In this case, the trained mathematical model may be a mathematical model that outputs the IOL frequency by inputting the corneal curvature, axial length, expected postoperative anterior chamber depth, and the like. For example, the corneal curvature and the axial length may be used as inputs to the mathematical model that outputs the IOL frequency, the corneal curvature and the expected postoperative anterior chamber depth may be used, or the eye. Two may be used: axial length and expected postoperative anterior chamber depth, or three may be used: corneal curvature, axial length and expected postoperative anterior chamber depth. Here, the corneal curvature may be the anterior corneal curvature, the posterior corneal curvature, or both.

For example, the arithmetic control unit calculates the predicted postoperative anterior chamber depth based on a plurality of eye shape parameters, and inputs the eye shape parameters and the predicted postoperative anterior chamber depth into the mathematical model to obtain the IOL frequency from the mathematical model. It may be output. In this way, even if the arithmetic control unit outputs the IOL frequency from the mathematical model by inputting the primary parameter which is the measured value and the secondary parameter calculated based on the measured value into the mathematical model. Good.

The arithmetic control unit may use two trained mathematical models. For example, the first mathematical model may be a mathematical model that outputs the predicted postoperative anterior chamber depth by inputting a plurality of eye shape parameters such as the lens anterior curvature, the lens posterior curvature, the lens thickness, and the anterior chamber depth. Further, the second mathematical model may be a mathematical model that outputs the IOL frequency by inputting eye shape parameters such as the predicted postoperative anterior chamber depth, corneal curvature, and axial length. In this case, the arithmetic control unit inputs the plurality of eye shape parameters into the first mathematical model, and the predicted postoperative anterior chamber depth output from the first mathematical model and the eye shape parameters are referred to as the first mathematical model. May output the IOL frequency from the second mathematical model by inputting to a different second mathematical model. This makes it possible to accurately predict the predicted postoperative anterior chamber depth and IOL power.

The arithmetic control unit may input feature parameters into the mathematical model in addition to the eye shape parameters. In this case, the trained mathematical model may be a mathematical model that outputs the predicted postoperative anterior chamber depth, IOL frequency, etc. by inputting at least one of the eye shape parameter and the feature parameter. By including the feature parameter in the input of the mathematical model, it is possible to output the ELP or IOL power in consideration of the influence other than the eye shape.

The trait parameter may be, for example, a parameter relating to at least one of the characteristics of the patient (subject)'s race, gender, age, and operator (examiner). Further, the characteristic parameter may be a refractive index such as a patient's corneal refractive index, crystalline lens refractive index, vitreous refractive index, and aqueous humor refractive index. In addition, the feature parameter may be a desired parameter (or required parameter) such as the postoperative refractive power, visual acuity, working distance, and IOL focus number desired by the patient or the operator. Further, the feature parameter may include a lens constant used in each IOL power calculation formula. The lens constant is, for example, the feature parameter is the A constant in the SRK / T equation, the Sajan factor (SF) in the Holladay equation, the IOL constant in the Haigis equation, and the like.

Further, the feature parameter may be a parameter related to the feature of the IOL model. For example, the feature parameters may be the total length of the IOL, the diameter of the support portion, the shape, the material, and the like. Further, the feature parameter may be, for example, a parameter obtained by a function of the equatorial dimension (for example, diameter) of the capsular bag. For example, the function of the diameter of the capsular bag may be a function of the amount of anteroposterior displacement of the IOL optical portion with respect to the length of the loop (support portion) of the IOL at the time of intraocular fixation.

For example, when using an ELP-related feature parameter such as an IOL model or sac diameter as a feature parameter, the mathematical model inputs the IOL model or sac diameter-related feature parameter, corneal curvature, anterior chamber depth, and axial length. This may be a mathematical model that outputs an ELP.

Instead of inputting the feature parameters related to the IOL model or the like into the mathematical model, a mathematical model that outputs ELP by inputting the eye shape parameters may be prepared for each IOL model.

The feature parameter may be used not only for machine learning but also as a correction coefficient for correcting the predicted postoperative anterior chamber depth or IOL frequency obtained by machine learning. For example, the arithmetic control unit obtains the predicted position of the ELP (for example, the distance from the cornea to the contact point between the capsular bag and the IOL loop) that is assumed to be independent of the IOL model by machine learning, and obtains the predicted position of the IOL model or the sac diameter. By correcting the tentative predicted ELP according to the feature parameters related to ELP such as, an effective ELP suitable for each IOL model may be obtained. As a result, it is possible to reduce the complexity of acquiring the ELP predicted value by using machine learning for each IOL model.

The arithmetic control unit selects an appropriate IOL frequency calculation formula from the IOL frequency calculation formulas such as SRK / T formula, Binkhorst formula, Holladay formula, Holladay2 formula, HofferQ formula, Olsen formula, and Haigis formula. The information may be output to a mathematical model. In this case, the mathematical model may be a mathematical model that outputs selection information of an appropriate IOL calculation formula by inputting a feature amount required for selection of the calculation formula. Here, the selection information may be, for example, information of an appropriate formula among a plurality of IOL frequency calculation formulas, or the recommended degree of each formula IOL frequency calculation formula. For example, the arithmetic control unit inputs the feature quantity related to any of the axial length, the IOL model, the surgeon, the race, the gender, and the age into the mathematical model, so that the selection information of the IOL frequency calculation formula is input to the mathematical model. May be output.

The apparatus may further include a presenting unit (for example, a monitor 70) that presents the selection information output by the mathematical model to the user. In addition, the device may further include a warning unit (eg, monitor 70) that warns when the user attempts to use an inappropriate expression in the selection information output by the mathematical model.

The arithmetic control unit may optimize the constants or coefficients used in the IOL frequency calculation formula by the above mathematical model. This makes it possible to calculate the IOL frequency by using a value more appropriate than the empirically obtained value as a constant or a coefficient.

The arithmetic control unit may execute an IOL frequency determination program stored in a storage unit (for example, a memory 85) or the like. The IOL frequency determination program includes, for example, a parameter acquisition step and a calculation step. The parameter acquisition step is a step of acquiring a plurality of eye shape parameters of the eye to be inspected. The calculation step is a step of outputting IOL-related information from the mathematical model by inputting a plurality of eye shape parameters into the mathematical model trained by the machine learning algorithm.

<Example>
Hereinafter, the ophthalmologic imaging apparatus 10 according to the present disclosure will be described with reference to the drawings. FIG. 1 is a schematic configuration diagram showing a measurement unit 200 of the ophthalmologic imaging apparatus 10 according to the present embodiment. The following optical system is built in a housing (not shown). Further, the housing is three-dimensionally moved with respect to the eye E to be inspected via an operating member (for example, a touch panel, a joystick, etc.) by driving a well-known alignment moving mechanism. In the following description, the optical axis direction of the eye to be inspected (eye E) will be the Z direction, the horizontal direction will be the X direction, and the vertical direction will be the Y direction. The surface direction of the fundus may be considered as the XY direction.

In the following description, an ophthalmologic imaging device 10 including an optical coherence tomography device (OCT device) 5 and a corneal shape measuring device 300 will be described as an example. The OCT device 5 is used as an anterior segment imaging device for capturing a cross-sectional image of the eye E to be inspected. Further, the OCT device 5 is used as an axial length measuring device for measuring the axial length of the eye E. The corneal shape measuring device 300 is used for measuring the corneal shape.

The OCT device 5 includes an interference optical system (OCT optical system) 100. The OCT optical system 100 irradiates the eye E with the measurement light. The OCT optical system 100 detects an interference state between the measurement light reflected from each part of the eye to be inspected (for example, the cornea, the crystalline lens, the fundus, etc.) and the reference light by the light receiving element (detector 120). The OCT optical system 100 includes a scanning unit (for example, an optical scanner) 108 that scans the measurement light on the eye to be inspected in order to change the imaging position on the eye to be inspected. The control unit 80 controls the operation of the scanning unit 108 based on the set imaging position information, and acquires a cross-sectional image based on the received signal from the detector 120.

The OCT optical system 100 has a device configuration of a so-called optical coherence tomography (OCT) for ophthalmology. The OCT optical system 100 divides the light emitted from the measurement light source 102 into the measurement light (sample light) and the reference light by the coupler (optical divider) 104. Then, the OCT optical system 100 guides the measurement light to the eye to be inspected by the measurement optical system 106, and guides the reference light to the reference optical system 110. After that, the detector (light receiving element) 120 receives the interference light obtained by combining the measurement light reflected by each part of the eye to be inspected with the reference light.

The light emitted from the light source 102 is divided into a measured luminous flux and a reference luminous flux by the coupler 104. Then, the measured luminous flux is emitted into the air after passing through the optical fiber. The luminous flux is focused on the eye to be inspected via the scanning unit 108 and other optical members of the measurement optical system 106. Then, the light reflected by each part of the eye to be inspected is returned to the optical fiber through the same optical path.

The scanning unit 108 scans the measurement light on the eye E in the XY direction (transverse direction). The scanning unit 108 is, for example, two galvanometer mirrors, and the reflection angle thereof is arbitrarily adjusted by the drive mechanism 109.

As a result, the light flux emitted from the light source 102 changes its reflection (traveling) direction and is scanned in an arbitrary direction on the eye E. As a result, the imaging position on the eye to be inspected is changed. The scanning unit 108 may be configured to deflect light. For example, in addition to a reflection mirror (galvano mirror, polygon mirror, resonant scanner), an acoustic optical element (AOM) that changes the traveling (deflection) direction of light is used.

The reference optical system 110 generates a reference light that is combined with the reflected light acquired by the reflection of the measurement light by the eye E. The reference optical system 110 may be a Michaelson type or a Machzenda type. The reference optical system 110 is formed by, for example, a reflective optical system (for example, a reference mirror), and the light from the coupler 104 is reflected by the reflective optical system to be returned to the coupler 104 again and guided to the detector 120. As another example, the reference optical system 110 is formed by a transmitted optical system (for example, an optical fiber) and guides the light from the coupler 104 to the detector 120 by transmitting it without returning it.

The reference optical system 110 has a configuration in which the optical path length difference between the measurement light and the reference light is changed by moving the optical member in the reference optical path. For example, the reference mirror is moved in the optical axis direction. The configuration for changing the optical path length difference may be arranged in the measurement optical path of the measurement optical system 106.

The detector 120 detects an interference state between the measurement light and the reference light. In the case of Fourier domain OCT, the spectral intensity of the interfering light is detected by the detector 120, and the depth profile (A scan signal) in a predetermined range is acquired by the Fourier transform on the spectral intensity data. Here, the control unit 80 can acquire a cross-sectional image by scanning the measurement light with each unit of the eye to be inspected in a predetermined transverse direction by the scanning unit 108. For example, a cross-sectional image of the anterior segment of the eye to be inspected is taken. For example, by scanning in the X direction or the Y direction, a cross-sectional image of the XZ plane or the YZ plane of the eye to be inspected can be obtained (in the present embodiment, the measurement light is one-dimensional with respect to the anterior eye portion in this way. B scan is a method of scanning to obtain a tomographic image). The acquired cross-sectional image is stored in the memory 85 connected to the control unit 80. Further, it is also possible to acquire a three-dimensional image of the eye to be inspected by scanning the measurement light two-dimensionally in the XY directions.

For example, the Fourier domain OCT includes Spectral-domain OCT (SD-OCT) and Swept-source OCT (SS-OCT). Further, it may be Time-domain OCT (TD-OCT).

In the case of SD-OCT, a low coherent light source (broadband light source) is used as the light source 102, and the detector 120 is provided with a spectroscopic optical system (spectrum meter) that disperses the interference light into each frequency component (each wavelength component). .. The spectrum meter includes, for example, a diffraction grating and a line sensor.

In the case of SS-OCT, a wavelength scanning light source (wavelength variable light source) that changes the emission wavelength at high speed in time is used as the light source 102, and for example, a single light receiving element is provided as the detector 120. The light source 102 is composed of, for example, a light source, a fiber ring resonator, and a wavelength selection filter. Then, as a wavelength selection filter, for example, a combination of a diffraction grating and a polygon mirror, and a filter using Fabry-Perot Etalon can be mentioned.

The corneal shape measuring device 300 is roughly classified into a kerato projection optical system 50, an alignment projection optical system 40, and an anterior segment frontal imaging optical system 30.

The kerato projection optical system 50 has a ring-shaped light source 51 arranged around the measurement optical axis L1 and projects a ring index onto the cornea to be inspected to measure the corneal shape (curvature, astigmatic axis angle, etc.). Used for As the light source 51, for example, an LED that emits infrared light or visible light is used. The projection optical system 50 may be an intermittent ring light source as long as at least three or more point light sources are arranged on the same circumference centered on the optical axis L1. Further, it may be a purachido index projection optical system that projects a plurality of ring indexes.

The alignment projection optical system 40 is arranged inside the light source 51, has a projection light source 41 (for example, λ = 970 nm) that emits infrared light, and is used for projecting an alignment index on the cornea Ec to be inspected. Then, the alignment index projected on the cornea Ec is used for alignment with respect to the eye to be inspected (for example, automatic alignment, alignment detection, manual alignment, etc.). In the present embodiment, the projection optical system 50 is an optical system that projects a ring index onto the corneal Ec to be inspected, and the ring index also serves as a Meyer ring. Further, the light source 41 of the projection optical system 40 also serves as anterior segment illumination that illuminates the anterior segment with infrared light from an oblique direction. In the projection optical system 40, an optical system that projects parallel light onto the cornea Ec may be further provided, and front-back alignment may be performed by combining with the finite light by the projection optical system 40.

The observation optical system 30 is used to image (acquire) a frontal image of the anterior segment of the eye. The observation optical system 30 includes a dichroic mirror 33, an objective lens 47, a dichroic mirror 62, a filter 34, an image pickup lens 37, and a two-dimensional image pickup element 35, and is used for capturing an anterior segment front image of the eye to be inspected. The two-dimensional image sensor 35 is arranged at a position substantially conjugate with the anterior segment of the eye to be inspected.

The light reflected from the anterior segment of the eye by the projection optical system 40 and the projection optical system 50 is imaged on the two-dimensional image sensor 35 via the dichroic mirror 33, the objective lens 47, the dichroic mirror 62, the filter 34, and the image pickup lens 37. The lens.

The light source 1 is a fixed vision lamp. Further, for example, a part of the anterior segment reflected light acquired by the reflection of the light emitted from the light source 1 in the anterior segment is reflected by the dichroic mirror 33 and imaged by the anterior segment front imaging optical system 30. Will be done.

Next, the control system will be described. The control unit 80 controls the entire device and calculates the measurement result. The control unit 80 is connected to each member of the OCT device 5, each member of the corneal shape measuring device 300, a monitor 70, an operation unit 84, a memory 85, and the like. In addition to various control programs, the memory 85 stores an IOL frequency calculation program and the like, which will be described later.

Subsequently, the IOL frequency calculation program used in this embodiment will be described. The IOL frequency calculation program of this embodiment utilizes a mathematical model trained by a machine learning algorithm. In the following description, a case where a feedforward neural network is used will be described as an example. Mathematical models in this neural network generally consist of an input layer for inputting data, an output layer for generating the data you want to predict, and one or more hidden layers between the input layer and the output layer. It is configured, and multiple nodes (also called units) are arranged in each layer. The node receives multiple inputs and computes one output. For example, the data input to each node of each layer is output to each node of the adjacent layer. At this time, different weights are added for each route. For example, the output value transmitted from one node to the next is enhanced or attenuated by the weight of each path. When the weighted data is input to the node, a function such as an activation function is applied and output to each node in the next layer. This input / output is repeated between the adjacent layers, and finally the prediction data is output from the output layer.

For example, if the node of the first layer is represented by i = 1, ..., I and the node of the second layer is represented by j = 1, ..., J, the total input u _j received by the node of the second layer is expressed by the following equation ( as in 1), a multiplied by different weights w _ji each input x _i of the first layer by adding all, the plus one single value b _i that this called bias.

Further, the output z _i of the nodes of the second layer, as the following equation (2), the output of the function f, such as activation function for the total input u _i. Examples of the activation function include functions such as a logistic jigmoid function, a hyperbolic tangent function, a rectified linear function, and a max-out function.

Mathematical models in neural networks as described above can be trained with training datasets to make predictions about new data. The training data set is, for example, a set of input training data and output training data so that when the input training data is input to the input layer, a value close to the output training data is output from the output layer. The weight and bias of each node in each layer are adjusted. Multiple training datasets are available, and by adjusting the iterative weights and biases, it is possible to obtain versatile weights and biases for various data, and it is possible to predict even unknown data. Become. Training of the mathematical model is continued, for example, until the error between the output of the input training data and the corresponding output training data is within an acceptable range. Backpropagation (backpropagation method) or the like is used to adjust the weight.

In a trained mathematical model, the adjusted weights represent the correlation between inputs and outputs. Based on this relationship, the mathematical model outputs a predicted value for the input of new data different from the training data. For example, each weight is applied as a coefficient to the output of each node. This reflects the correlation obtained by training in the output of the mathematical model.

In the mathematical model 800 of this embodiment, as shown in FIG. 2, four nodes N11 to N14 are arranged in the input layer M1, three nodes N21 to N23 are arranged in the intermediate layer M2, and one node N31 is arranged in the output layer M3. .. This mathematical model 800 uses a plurality of sets of training data sets including a subject's anterior lens curvature, lens posterior curvature, lens thickness and anterior chamber depth, and postoperative anterior chamber depth (ELP). Have been trained. That is, in the mathematical model 800 of this embodiment, when four eye shape parameters such as the lens anterior curvature, the lens posterior curvature, the crystalline lens thickness, and the anterior chamber depth are input to the input layer M1, the output layer M3 predicts before the operation. It is generated so that the tuft depth (predicted value of the postoperative anterior chamber depth) is output. The mathematical model 800 is implemented using an arbitrary computer language or the like so as to be executed on the processor of the control unit 80.

<Control operation>
The control operation of this device will be described below. FIG. 3 is a diagram showing a simple flow chart of a control operation when determining the intraocular lens power using this device. First, in step S1, the control unit 80 aligns the measurement unit 200 with respect to the eye to be inspected. At the time of alignment, the control unit 80 lights the light source 1, the light source 41, and the light source 51. At this time, as shown in FIG. 4, the monitor 70 displays a front image of the eye to be inspected taken by the two-dimensional image sensor 35. In the front image, for example, an electronically displayed reticle LT and a ring index Q1 and a ring index Q2 formed by the light source 41 and the light source 51 are shown.

When the subject is made to stare at the fixation lamp, the examiner operates an operation unit such as a touch panel so that the reticle LT and the ring index Q1 are concentric. The control unit 80 moves the measurement unit up / down / left / right according to the operation received by the operation unit. As a result, the device is aligned in the XY direction so that the optical axis L1 of the apparatus passes through the apex of the cornea of the eye to be inspected. In addition, the examiner operates the operation unit so that the ring index Q1 is in focus. The control unit 80 moves the measurement unit in the front-rear direction according to the operation received by the operation unit.

When the alignment with respect to the anterior segment is completed, the control unit 80 takes a front image of the anterior segment of the eye to be inspected by the observation optical system 30 (step S2). Further, the control unit 80 captures a cross-sectional image 500 of the eye to be inspected as shown in FIG. 5 by the OCT optical system 100 based on a preset scanning pattern (step 3). The acquired anterior segment image and cross-sectional image are stored in a memory 85 or the like.

Subsequently, in step 4, the control unit 80 acquires the eye shape parameter. For example, the control unit 80 calculates the corneal shape of the eye to be inspected based on the ring index images Q1 and Q2 in the anterior eye portion image 400 stored in the memory 85, respectively. The corneal shape is, for example, the radius of curvature of the cornea on the anterior surface of the cornea in the strong main meridian direction and the weak main meridian direction, the astigmatic axis angle of the cornea, and the like. Further, the control unit 80 analyzes a cross-sectional image taken by using the OCT device 5. For example, the control unit 80 detects the positions of the cornea, the crystalline lens, and the like by detecting the edge of the cross-sectional image, and measures the corneal film thickness, the anterior chamber depth, and the crystalline lens thickness based on the positions. Further, the control unit 80 makes a circular approximation (or an elliptical approximation, a conic curve approximation, etc.) of the detected anterior or posterior surface of the lens and the crystalline lens, and based on this approximate curve, the radius of curvature of the posterior surface of the corneum, the anterior curvature of the crystalline lens, and the posterior surface of the crystalline lens. Measure the curvature, etc. Furthermore, if the retina can be imaged by the OCT optical system, the axial length is measured. The acquired eye shape parameter is stored in, for example, a memory 85 or the like.

In step S5, the control unit 80 reads out the front lens curvature, the rear lens curvature, the lens thickness, and the anterior chamber depth acquired by the measurement unit 200 from the memory 85 and inputs them to each node of the input layer M1. Then, the control unit 80 performs the calculation according to the rules of the mathematical model 800, and outputs the value of the predicted postoperative anterior chamber depth from the output layer M3 (step S6). The control unit 80 stores the output expected postoperative anterior chamber depth in the memory 85.

When the acquisition of each parameter is completed, the control unit 80 calculates the intraocular lens power by partially diverting the known SRK / T formula, Binkhrst formula, and the like. For example, in step S7, the control unit 80 acquires the IOL frequency by substituting the above parameters into the SRK / T equation, the Binkhrst equation, and the like (step S8). When the SRK / T equation (the following equation (3)) is used, the intraocular lens power is calculated using the corneal radius of curvature, the axial length, the expected postoperative anterior chamber depth, and the like.

ここで、R:角膜曲率半径[mm](R＝(n_k−1.000)×1000/K)、n_k:検者によって選択された屈折率、LO:AL＋RT[mm]、RT:網膜の厚み[mm](RT＝0.65696−0.02029×AL)、AL:眼軸長[mm]、AD’:予想術後前房深度の補正値[mm](AD’＝H＋OF,OF＝AD−3.336)、AD:予想術後前房深度[mm](AD＝0.62467×A−68.747)、A:A定数、H:角膜高さ[mm](H=R−(R×R−((Cw×Cw)/4))^1/2)（ただし、(R×R−((Cw×Cw)/4))＜0の場合、H＝R）、Cw:角膜幅[mm]、Cw＝−5.41＋0.58412×LC＋0.098×K、LC:眼軸長の補正値[mm]（AL≦24.2の場合LC＝AL、AL＞24.2の場合LC＝−3.446＋1.716×AL−0.0237×AL²）、DR:術後希望する矯正用レンズの屈折力[D]、LP:移植するIOLの度数[D]、V:頂点距離、n_a:房水および硝子体の屈折率(＝1.336)、n_c:角膜の屈折率(＝1.333)、n_cml: n_c−1(＝0.333)である。

Here, R: corneal radius of curvature [mm] (R = (n _k −1.000) × 1000 / K), n _k : refractive index selected by the examiner, LO: AL + RT [mm], RT: thickness of retina [mm] (RT = 0.65696-0.02029 x AL), AL: Axial length [mm], AD': Predicted postoperative anterior chamber depth correction value [mm] (AD'= H + OF, OF = AD-3.336), AD: Expected postoperative anterior chamber depth [mm] (AD = 0.62467 × A-68.747), A: A constant, H: Corneal height [mm] (H = R− (R × R− ((Cw × Cw)) / 4)) ^1/2 ) (However, (R × R− ((Cw × Cw) / 4)) <0, H = R), Cw: corneal width [mm], Cw = −5.41 + 0. 58412 x LC + 0.098 x K, LC: Correction value of axial length [mm] (LC = AL when AL ≤ 24.2, LC = -3.446 + 1.716 x AL -0.0237 x AL ² when AL> 24.2), DR: Refractive force of the desired correction lens after surgery [D], LP: _{Degree of IOL to be transplanted [D], V: Appointment distance, n a} : Refractory coefficient of aqueous humor and vitreous body (= 1.336), n _c : Corneal refractive index (= 1.333), n _c ml: n _c −1 (= 0.333).

Conventionally, a regression equation or a theoretical equation has been used to predict ELP, and a certain degree of prediction accuracy has been obtained for an eyeball that meets certain rules. However, it could not cope with an eyeball having a characteristic shape (for example, an eye having an axial length but a shallow anterior chamber depth, an eye having a short axial axis but a deep anterior chamber depth, etc.), causing a sudden frequency shift. As described above, by using a mathematical model trained by machine learning to predict the postoperative anterior chamber depth considering the correlation of multiple eye shape parameters, it is characteristic that the shape deviates from the average shape. Even with the eyeball, the appropriate IOL frequency can be calculated with a higher probability.

In the above embodiment, the predicted postoperative anterior chamber depth was output by a mathematical model, but the present invention is not limited to this. For example, the IOL frequency may be output. In this case, a mathematical model is trained using a plurality of eye shape parameters such as corneal curvature, axial length, and anterior chamber depth as input training data, and using IOL power as output training data. As a result, the control unit 80 generates a mathematical model that outputs the IOL power when the eye shape parameter measured by the measurement unit 200 is input. By using this mathematical model, the IOL power can be calculated directly from the measured eye shape parameters.

The control unit 80 may acquire the toric IOL frequency by a mathematical model. In this case, a mathematical model is trained using a plurality of eye shape parameters as input training data and a toric IOL frequency as output training data. When determining the toric IOL frequency, the ocular shape parameters include, for example, anterior corneal astigmatism, anterior corneal astigmatism axis, posterior corneal astigmatism, posterior corneal astigmatism axis, corneal thickness (CT), anterior atrioventricular depth ACD, and postoperative anterior atrioventricular depth. (ELP), postoperative evoked astigmatism (SIA), pupil diameter (PS), anterior corneal astigmatism, anterior corneal astigmatism, posterior corneal astigmatism, posterior corneal astigmatism, incision position, incision width, incision angle, auxiliary port position , The number of auxiliary ports, and at least one of the auxiliary port widths. The incision position, incision width, and incision angle are the positions, widths, and angles (incision directions) of the incisions made on the cornea, corneal margin, or sclera of the eye to be incised when the IOL is inserted. The auxiliary port is, for example, an incision (or wound) for inserting an auxiliary tool into the eye when inserting an IOL. The astigmatism power of the toric IOL to be inserted into the eye to be inspected is calculated based on, for example, information on anterior lens astigmatism and posterior lens astigmatism. In addition to the toric IOL frequency, toric IOL-related information such as the toric angle and the postoperative rotation potential of the IOL may be acquired. When acquiring the rotatability of the IOL, the diameter of the capsular bag, the minor and major diameters of the capsular bag, the total length of the IOL, and the like may be used as input data to the mathematical model.

In addition, the arithmetic control unit operates from the mathematical model by inputting the characteristic parameters related to at least one of the incision position, the incision width, the incision angle, the auxiliary port position, the number of auxiliary ports, the auxiliary port width, and the operator into the mathematical model. The posterior evoked dysfunction may be output. Postoperative astigmatism is astigmatism caused by an incision during cataract surgery. Therefore, postoperative astigmatism may be affected by the surgical habits of the surgeon. Therefore, in order to output postoperative astigmatism, characteristic parameters related to at least one of the incision position, the incision width, the incision angle, the auxiliary port position, the number of auxiliary ports, the auxiliary port width, and the operator are used for the input of the mathematical model. ..

In the above embodiment, the IOL frequency was obtained by substituting the predicted postoperative anterior chamber depth calculated by the mathematical model into the substitution location of the IOL frequency calculation formula, but the IOL frequency was calculated again by another mathematical model. You may. For example, even if the IOL power is obtained using a mathematical model trained to output the IOL power by inputting the expected postoperative anterior chamber depth calculated by the mathematical model and other eye shape parameters. Good.

In the above embodiment, the output of the neural network is one, but a plurality of nodes may be provided in the output layer to output a plurality of values. For example, by inputting eye shape parameters such as corneal curvature, axial length, anterior chamber depth, anterior lens curvature, and posterior lens curvature, two parameters, expected postoperative anterior chamber depth and IOL frequency, may be calculated. In this case, a mathematical model may be trained using a plurality of eye shape parameters as input training data and postoperative anterior chamber depth and IOL frequency as output training data. Of course, the number of nodes may be any number not only for the output layer but also for the input layer and the hidden layer.

In the above embodiment, the neural network having one hidden layer is treated, but a deep neural network having two or more hidden layers may be used. In this case as well, a plurality of eye shape parameters are input to each node of the input layer, and the expected postoperative anterior chamber depth or IOL power is finally output from the output layer through the plurality of hidden layers. By using a deep neural network, IOL power information that more finely reflects the correlation of each eye shape parameter can be acquired.

In the above embodiment, the predicted postoperative anterior chamber depth was calculated by regression using a neutral network, but IOL-related information may be acquired by classification using machine learning. For example, the control unit 80 may select an appropriate IOL frequency calculation formula from a plurality of IOL frequency calculation formulas by using a mathematical model. For example, the control unit 80 obtains an appropriate IOL frequency calculation formula by inputting an eye shape parameter into a mathematical model. Thereby, the IOL power can be calculated by the IOL power calculation formula suitable for the shape of the eye to be inspected. When selecting the IOL frequency calculation formula, in addition to the eye shape parameters, characteristic parameters related to the IOL model, operator, race, gender, age, etc. may be input to the mathematical model.

In the above embodiment, the IOL frequency is calculated by the neural network, but the present invention is not limited to this. For example, other machine learning algorithms such as random forest and boosting may be used. For example, when a random forest is used, the IOL frequency is obtained from each of several decision trees, and the IOL frequency is obtained by averaging the values of the IOL frequencies obtained from each decision tree. In addition, the discriminator obtained by boosting may be used to classify which IOL power calculation formula is suitable for the eye to be inspected. An accurate IOL frequency can be calculated by using a machine learning algorithm, not limited to a neural network.

In the above embodiment, the eye shape parameter to be input to the mathematical model is measured by the measuring unit 200, but the eye shape parameter may be acquired from a server or the like. For example, a plurality of measurement results taken by a large number of models may be stored in a server through a network, and each device may be able to acquire data measured by another device from the server. For example, the control unit 80 may acquire eye shape parameters from an electronic medical record system in which registration information of a subject, measurement image data, and the like are managed.

In the above embodiment, the IOL frequency was obtained by applying the predicted postoperative anterior chamber depth calculated by the trained mathematical model to the IOL frequency calculation formula, but the predicted postoperative anterior chamber calculated by the theoretical formula was obtained. The IOL power may be output by inputting the depth and other eye shape parameters into the mathematical model. The method for calculating the expected postoperative anterior chamber depth will be described below.

The calculation of the predicted postoperative anterior chamber depth by the theoretical formula will be described with reference to FIG. The expected postoperative anterior chamber depth is calculated by adding the offset amount X and the correction amount α (for example, a function of the A constant) to the corneal height (height from the anterior surface of the cornea to the anterior surface of the crystalline lens). .. The corneal height is calculated by adding the corneal thickness CCT and the anterior chamber depth ACD.

The offset amount X indicates the distance from the position on the front surface of the crystalline lens to the position of the equator (the maximum diameter portion of the crystalline lens). The offset amount X is excluded because the amount of movement of the IOL to the posterior lens capsule side (correction amount α shown above) caused by the IOL receiving pressure from the capsular bag is not taken into consideration. Will be described later). The position of the tip of the IOL support (loop) is approximately the same as the equator position. The equator position may be a position where the front surface of the crystalline lens and the rear surface of the crystalline lens intersect.

Hereinafter, a method of calculating the offset amount X will be described. For example, the control unit 80 calculates the equator position using the lens front surface curvature radius R3, the lens rear surface curvature radius R4, and the lens thickness LT, and sets this equator position as the position of the postoperative loop tip. Since the position of the loop tip after the operation is treated as approximately the same as the position of the optical portion of the IOL, the offset amount X is calculated with the position of the loop tip as the position of the optical portion.

Further, in FIG. 7, the distance h indicates the distance from the optical axis L1 to the intersection of the approximate circle on the front surface of the crystalline lens and the approximate circle on the rear surface of the crystalline lens. The distance X1 indicates the distance from the front surface curvature center O4 of the crystalline lens on the optical axis L1 to the rear surface of the crystalline lens. The distance X1'indicates the distance from the intersection of the approximate circle on the front surface of the crystalline lens and the approximate circle on the posterior surface of the crystalline lens to the posterior surface of the crystalline lens. The distance X2 indicates the distance from the center of curvature of the rear surface of the crystalline lens O3 on the optical axis L1 to the front surface of the crystalline lens. The distance X indicates the distance from the intersection of the approximate circle on the front surface of the crystalline lens and the approximate circle on the rear surface of the crystalline lens to the front surface of the crystalline lens, and is an offset amount. According to Pythagorean theorem, the following equation holds.

そして、上記の式（４）において、距離ｈが同様であるため、これらの式をオフセット量Ｘについて解くと以下の式が成り立つ。

Then, in the above equation (4), since the distance h is the same, the following equation holds when these equations are solved for the offset amount X.

The control unit 80 calculates the expected postoperative anterior chamber depth based on the offset amount X calculated by the equation (5), the anterior chamber depth ACD, the corneal thickness CCT, and the correction amount α. The correction amount α is a parameter for correcting the amount of movement of the IOL when the IOL is pushed down toward the posterior capsule by the pressure of the lens capsule after the IOL is inserted. The correction amount α differs depending on the IOL model, and is calculated using, for example, an A constant set for each model based on clinical data. The expected postoperative anterior chamber depth ELP can be calculated by the following formula (6).

予想術後前房深度としては、角膜裏面から定義される場合もあるが、ここでは、角膜前面からＩＯＬ前面までの距離とした。

The expected postoperative anterior chamber depth may be defined from the back surface of the cornea, but here it is the distance from the front surface of the cornea to the front surface of the IOL.

The control unit 80 may directly calculate the IOL power by inputting the predicted postoperative anterior chamber depth calculated by the equation (6) and each eye shape parameter acquired by the measurement unit 200 into the mathematical model. .. In this way, the eye shape parameter, which is the measured value (primary parameter), and the predicted postoperative anterior chamber depth, which is the calculated value (secondary parameter) calculated based on the primary parameter, are used as inputs for the mathematical model. May be good. As a result, the control unit 80 can calculate the IOL frequency that reflects the correlation between the primary parameter and the secondary parameter.

It should be noted that a mathematical model trained by machine learning may be used to predict eye diseases and the like. For example, the control unit 80 inputs an eye shape parameter into a mathematical model to perform intermediate translucent opacity of the cornea or lens, keratoconus, LASIK postoperative corneal ectaticism, Perusid corneal marginal degeneration, etc. The degree of progression of corneal degeneration, the incidence of postoperative illness, etc. may be output. When obtaining the degree of progression of corneal degeneration, corneal curvature, corneal thickness, etc. are used as inputs to the mathematical model. When obtaining the progress of opacity of the intermediate translucent body, the brightness value of the intermediate translucent body reflected in the cross-sectional image of the anterior segment of the eye is used as an input of a mathematical model. The mathematical model may have a plurality of nodes in the output layer and output the above-mentioned disease information in addition to the IOL-related information. As a result, the IOL power can be obtained and information on the state of the eye to be inspected can be obtained.

In the above examples, the IOL fixed in the sac has been described, but the present invention is not limited to this. For example, when inserting an IOL for a crystalline lens, performing a piggyback (inserting a second IOL into an IOL-inserted eye), or exchanging an IOL, a mathematical model trained by machine learning can be used. IOL-related information can be obtained using this. In the case of piggyback or IOL exchange, the information about the IOL already inserted in the eye to be examined may be used for inputting the mathematical model as one of the feature parameters.

The present invention also applies to the case where a three-dimensional shape image is acquired by acquiring an anterior segment tomography image at a plurality of scanning positions in an optical coherence tomography device for photographing an anterior segment tomography (cross-sectional image). Applicable. The OCT device 5 is an anterior segment imaging device that acquires a three-dimensional cross-sectional image (three-dimensional anterior segment data) of the anterior segment, and the control unit 80 is a three-dimensional cross section acquired by the anterior segment imaging device. Based on the image, the offset distance from the front surface of the crystalline lens to the contact point between the zonule of Zinn and the crystalline lens is three-dimensionally obtained. In this case, the average of the front lens curvature and the rear lens curvature of the crystalline lens for each meridian direction in the three-dimensional anterior segment data is calculated, and the ELP is calculated based on this. Then, by acquiring the measured value from the three-dimensional shape image, the accuracy of the acquired measured value is improved.

In the present embodiment, as an anterior segment imaging device for capturing a cross-sectional image of the anterior segment, an optical coherence tomography device for photographing an anterior segment tomography (cross-sectional image) is mentioned as an example, but the present invention is not limited to this. A light projection optical system that projects light emitted from a light source toward the anterior segment of the eye to be inspected to form a light cut surface on the anterior segment of the eye, and before acquisition by scattering of the optical cut surface in the anterior segment of the eye. The configuration may be such that it has a light receiving optical system having a detector that receives light including scattered light in the eye, and forms a cross-sectional image of the anterior segment of the eye based on a detection signal from the detector. That is, it can also be applied to a device or the like that projects slit light onto the anterior segment of the optometry and obtains a cross-sectional image of the anterior segment with a Shine pluque camera.

Further, it can be applied to a device that acquires a three-dimensional shape image of the anterior segment by rotating the Shine Prook camera or moving it in the horizontal or vertical direction. In this case, by performing the deviation correction for each predetermined rotation angle, it is possible to accurately acquire the three-dimensional shape image of the anterior segment of the eye, and the accuracy of the measured value acquired from the three-dimensional shape image is improved. In this case, the displacement in the direction perpendicular to the imaging surface (slit cross section) is detected, and the displacement correction process is performed based on the detection result.

Further, in the above configuration, a cross-sectional image of the anterior segment is optically obtained, but the present invention is not limited to this. For example, a cross-sectional image of the anterior segment may be acquired by detecting reflection information from the anterior segment using an ultrasonic probe for B scan.

In this embodiment, as a method for calculating the IOL frequency, a known regression equation such as the SRK / T equation or the Binkhors equation or an IOL frequency calculation formula which is a theoretical formula is used, but the method is not limited thereto. For example, the IOL frequency can be calculated by a ray tracing method that simulates the behavior of light by geometrically following the state of reflection and refraction of light using light rays. In this case, the IOL frequency is calculated by the ray tracing method using the predicted postoperative anterior chamber depth ELP, corneal thickness CCT, axial length measurement result AL, corneal radius of curvature on the anterior surface of the cornea, and corneal radius of curvature on the posterior surface of the cornea. Since the ray tracing method calculates the IOL power by simulating the reflection and refraction of light, the IOL power can be calculated more accurately than the theoretical IOL power calculation formula.

In the present embodiment, the corneal shape measuring device 300 is used to calculate the corneal radius of curvature on the anterior surface of the cornea, and the OCT device 5 is used to calculate the radius of curvature of the cornea on the posterior surface of the cornea. Not limited. The OCT device 5 may be used to calculate the radius of curvature of the cornea on the anterior-posterior surface of the cornea. Further, the radius of curvature of the cornea on the anterior-posterior surface of the cornea may be treated with the same measured value. That is, the radius of curvature of the cornea on the anterior surface of the cornea calculated by the corneal shape measuring device 300 is used as the radius of curvature of the cornea on the anterior and posterior surfaces of the cornea.

In this embodiment, corneal topography can also be used as the corneal shape measuring device 300. In this case, when calculating the radius of curvature of the anterior surface of the cornea, the radius of curvature of the anterior surface of the cornea is calculated from the overall shape of the cornea, so that the radius of curvature of the anterior surface of the cornea is calculated with high accuracy. Therefore, when calculating the IOL frequency, the accuracy of the IOL frequency calculation is improved.

The present invention is not limited to the apparatus described in the present embodiment. For example, an IOL frequency calculation program that performs the functions of the above embodiment is supplied to a system or an apparatus via a network or various storage media. Then, a computer of the system or device (for example, a CPU or the like) can read and execute the program.

5 Optical coherence tomography device 30 Frontal imaging optical system 40 Alignment projection optical system 50 Kerat projection optical system 70 Monitor 80 Control unit 85 Memory 84 Operation unit

Claims (5)

An ophthalmic device for determining the frequency of the IOL to be inserted into the eye to be inspected.
Parameter acquisition means for acquiring multiple eye shape parameters of the eye to be inspected,
It is equipped with an arithmetic control means for calculating the IOL frequency.
The arithmetic control means is for inputting the plurality of eye shape parameters into a mathematical model trained by a machine learning algorithm to select an expression suitable for the shape of the eye to be inspected from a plurality of IOL frequency calculation formulas. An ophthalmic apparatus characterized in that selection information is output to the mathematical model. The ophthalmic apparatus according to claim 1, wherein the selection information is a recommended degree of each IOL frequency calculation formula. The arithmetic control means is characterized in that, in addition to the eye shape parameters, characteristic parameters relating to at least one of the characteristics of the patient's race, gender, age, IOL model, and operator are input to the mathematical model. Item 1 or 2 ophthalmic apparatus. The ophthalmic apparatus according to claim 3, wherein the feature parameter includes information about an IOL that has already been inserted into the eye to be inspected. It is an IOL power determination program executed in an ophthalmic apparatus for determining the power of an IOL to be inserted into an eye to be inspected, and is executed by a processor of the ophthalmic apparatus.
A parameter acquisition step for acquiring multiple eye shape parameters of the eye to be inspected,
By inputting the plurality of eye shape parameters into the mathematical model trained by the machine learning algorithm, the arithmetic step of outputting the selection information for selecting an appropriate formula from the plurality of IOL frequency calculation formulas to the mathematical model. , Is executed by the ophthalmic apparatus.

Images