JP2024046836A - Classification method and computer program - Google Patents

Abstract

The present invention provides a technology that can quantitatively classify embryos at the cleavage stage based on three-dimensional images.
[Solution] First, a three-dimensional image D2 of an embryo is obtained by optical coherence tomography. Next, in the three-dimensional image D2, blastomere regions occupied by individual blastomeres and fragmentation regions occupied by fragmentation are extracted. Next, the volumes of the individual blastomere regions and the volume of the fragmentation region are calculated. Next, a first index value P1 indicating the variation in the volume of the blastomere in the embryo is calculated based on the volume of the blastomere region. Also, a second index value P2 indicating the proportion of fragmentation in the embryo is calculated based on the volume of the fragmentation region. Then, the embryo is classified based on the first index value P1 and the second index value P2. In this way, embryos in the cleavage stage can be quantitatively classified based on the three-dimensional image D2.
[Selected figure] Figure 3

Description

The present invention relates to a classification method for classifying embryos at the cleavage stage, and a computer program for causing a computer to execute the classification method.

In assisted reproductive technology aimed at infertility treatment, embryos are fertilized outside the body, cultured for a certain period of time, and then returned to the body. In this type of assisted reproductive technology, it is important to accurately evaluate the condition of multiple embryos and return good embryos to the body in order to increase the success rate of pregnancy.

Traditionally, the Veeck classification has mainly been used to evaluate embryos in the cleavage stage. The Veeck classification classifies embryos into five grades, from grade 1 to grade 5, based on the uniformity of the size of the blastomeres produced by cleavage and the rate of fragmentation. In conventional assisted reproductive technology, embryologists would evaluate embryos according to the criteria of the Veeck classification while looking at microscopic images of the embryos.

However, traditionally, the microscopic images used by embryologists for evaluation are two-dimensional images of the embryo viewed from one direction, making it difficult to accurately grasp the three-dimensional structure of the embryo. For this reason, the size of each blastomere and the size of fragmentation have been left to the subjective judgment of the embryologist.

In recent years, it has been proposed to photograph embryos using optical coherence tomography (OCT) and use the resulting three-dimensional images to observe the embryos. Conventional technology for photographing embryos using optical coherence tomography is described in, for example, Patent Document 1.

JP 2019-133429 A

However, methods for quantitatively classifying the state of an embryo based on three-dimensional images obtained by optical coherence tomography have not yet been fully established. The subjective judgment by the embryologist mentioned above can lead to discrepancies in judgment depending on the skill of the embryologist. To further increase the success rate of assisted reproductive technology, there is a demand for technology that can quantitatively classify the state of an embryo based on three-dimensional images, without relying on the skill of the embryologist.

The present invention has been developed in consideration of these circumstances, and aims to provide a technology that can quantitatively classify embryos at the cleavage stage based on three-dimensional images.

In order to solve the above problems, the first invention of the present application is a classification method for classifying embryos in the cleavage stage, comprising the steps of: a) acquiring a three-dimensional image of the embryo by optical coherence tomography; b) extracting, in the three-dimensional image, blastomere regions occupied by individual blastomeres and fragmentation regions occupied by fragmentation; c) calculating the volumes of the individual blastomere regions and the fragmentation regions; d) calculating a first index value indicating the variation in the volume of the blastomere in the embryo based on the volume of the blastomere regions; e) calculating a second index value indicating the proportion of the fragmentation in the embryo based on the volume of the fragmentation region; and f) classifying the embryo based on the first index value and the second index value.

The second invention of the present application is the classification method of the first invention, in which in step f), the embryos are classified into five grades, grade 1 to grade 5, based on the first index value and the second index value.

The third invention of the present application is the classification method of the first invention, in which in step f), the embryos are classified into two types, good embryos and non-good embryos, based on the first index value and the second index value.

The fourth invention of the present application is the classification method of the first invention, further comprising a step in which a computer generates a trained model by learning the relationship between the first index value and the second index value and the classification result of the embryo based on a machine learning algorithm, and in step f), the computer inputs the first index value and the second index value to the trained model and classifies the embryo based on the classification result output from the trained model.

The fifth invention of the present application is the classification method of the first invention, in which in step f), the embryo is classified based on the first index value, the second index value, and a preset threshold value.

The sixth invention of the present application is the classification method of the first invention, in which in step d), if the number of blastomere regions is not a power of two, some blastomere regions with a large volume are divided into two blastomere regions to change the number of blastomere regions to a power of two, and the first index value is calculated for the multiple blastomere regions after the change.

The seventh invention of the present application is the classification method of the first invention, in which in step d), the volume of the blastomere region is standardized by dividing it by the average value or the median value, and the first index value is calculated based on the standardized volume.

The eighth invention of the present application is a computer program that causes a computer to execute the classification method of any one of the first to seventh inventions.

According to the first to eighth aspects of the present application, embryos in the cleavage stage can be quantitatively classified based on three-dimensional images.

In particular, according to the second invention of the present application, Veek classification, which was previously performed by an observer on the basis of a subjective judgment based on a two-dimensional microscope image, can now be performed objectively based on a three-dimensional image.

In particular, according to the third aspect of the present invention, good embryos can be easily selected based on three-dimensional images.

In particular, according to the fourth aspect of the present invention, by using machine learning, embryos can be classified without the user having to set a classification threshold.

In particular, according to the fifth aspect of the present invention, embryos can be classified based on a preset threshold value.

In particular, according to the sixth aspect of the present invention, even when an embryo contains a mixture of cleaved and immediately prior to cleavage, the first index value indicating the variation in blastomere volume can be appropriately calculated.

In particular, according to the seventh invention of the present application, it is possible to suppress the variation in the first index value due to the generation of cleavage.

FIG. 1 is a diagram showing the configuration of an embryo classification device. FIG. 2 is a control block diagram of the embryo classification device. FIG. 1 is a block diagram conceptually illustrating the functions of a computer for classifying embryos. FIG. 1 shows the structure of an embryo at the cleavage stage. 13 is a flowchart showing the flow of an embryo classification process. 11 is a flowchart showing a flow of a calculation process of a first index value. FIG. 13 is a diagram showing how a blastomere region is adjusted. 11A and 11B are diagrams for explaining a first technique for calculating a first index value. FIG. 13 is a diagram for explaining a second method for calculating a first index value. FIG. 11 is a diagram for explaining a second method for calculating a first index value.

The following describes an embodiment of the present invention with reference to the drawings.

1. Configuration of the observation device
1 is a diagram showing the configuration of an embryo classification device 1 according to an embodiment of the present invention. This embryo classification device 1 is a device that performs optical coherence tomography of an embryo 9 held in a sample container 90 and classifies the state of the embryo 9 based on the obtained three-dimensional image. The embryo 9 is an embryo in the division stage from when a fertilized egg starts to divide until it becomes a morula. The embryo 9 is, for example, a human embryo, but may also be an embryo of a non-human animal.

As shown in FIG. 1, the embryo classification device 1 includes a stage 10, an imaging unit 20, and a computer 30.

The stage 10 is a support base that supports the sample container 90. For example, a well plate is used as the sample container 90. The well plate has a number of wells (recesses) 901. Each well 901 has a U-shaped or V-shaped bottom. The embryo 9 is held near the bottom of each well 901 together with a culture solution. The material used for the sample container 90 is a transparent resin or glass that transmits light.

The stage 10 has an opening 11 that penetrates in the vertical direction. The sample container 90 is supported horizontally while being fitted into the opening 11 of the stage 10. Therefore, the bottom surface of the sample container 90 is not covered by the stage 10 and is exposed toward the imaging unit 20.

The imaging unit 20 is a unit that captures an image of the embryo 9 in the sample container 90. The imaging unit 20 is disposed below the sample container 90 supported by the stage 10. The imaging unit 20 is an optical coherence tomography (OCT) device that is capable of acquiring tomographic images and three-dimensional images of the embryo 9.

As shown in FIG. 1, the imaging unit 20 has a light source 21, an object optical system 22, a reference optical system 23, a detection unit 24, and an optical fiber coupler 25. The optical fiber coupler 25 is formed by connecting a first optical fiber 251 to a fourth optical fiber 254 at a connection portion 255. The light source 21, the object optical system 22, the reference optical system 23, and the detection unit 24 are connected to each other via an optical path formed by the optical fiber coupler 25.

The light source 21 has a light-emitting element such as an LED. The light source 21 emits low-coherence light containing a wide band of wavelength components. In order to allow the light to reach the inside of the embryo 9 without invading the embryo 9, it is desirable that the light emitted from the light source 21 be near-infrared. The light source 21 is connected to a first optical fiber 251. The light emitted from the light source 21 enters the first optical fiber 251, and is branched at the connection portion 255 into light that enters the second optical fiber 252 and light that enters the third optical fiber 253.

The second optical fiber 252 is connected to the object optical system 22. The light traveling from the connection part 255 to the second optical fiber 252 is incident on the object optical system 22. The object optical system 22 has a plurality of optical components including a collimator lens 221 and an objective lens 222. The light emitted from the second optical fiber 252 passes through the collimator lens 221 and the objective lens 222 and is irradiated onto the embryo 9 in the sample container 90. At this time, the objective lens 222 converges the light toward the embryo 9. Then, the light reflected by the embryo 9 (hereinafter referred to as "observation light") passes through the objective lens 222 and the collimator lens 221 and is incident on the second optical fiber 252 again.

As shown in FIG. 1, the object optical system 22 is connected to a scanning mechanism 223. The scanning mechanism 223 moves the object optical system 22 slightly in the vertical and horizontal directions in accordance with commands from the computer 30. This allows the incident position of the light on the embryo 9 to be moved slightly in the vertical and horizontal directions.

The imaging unit 20 can also be moved horizontally by a moving mechanism (not shown). This allows the field of view of the imaging unit 20 to be switched between multiple wells 901.

The third optical fiber 253 is connected to the reference optical system 23. Light traveling from the connection portion 255 to the third optical fiber 253 is incident on the reference optical system 23. The reference optical system 23 has a collimator lens 231 and a mirror 232. The light emitted from the third optical fiber 253 passes through the collimator lens 231 and is incident on the mirror 232. Then, the light reflected by the mirror 232 (hereinafter referred to as "reference light") passes through the collimator lens 231 and is again incident on the third optical fiber 253.

As shown in FIG. 1, the mirror 232 is connected to the advance/retract mechanism 233. The advance/retract mechanism 233 moves the mirror 232 slightly in the optical axis direction according to commands from the computer 30. This makes it possible to change the optical path length of the reference light.

The fourth optical fiber 254 is connected to the detection unit 24. The observation light incident on the second optical fiber 252 from the object optical system 22 and the reference light incident on the third optical fiber 253 from the reference optical system 23 are joined at the connection unit 255 and incident on the fourth optical fiber 254. The light emitted from the fourth optical fiber 254 is then incident on the detection unit 24. At this time, interference occurs between the observation light and the reference light due to a phase difference. The spectrum of this interference light differs depending on the height of the reflection position of the observation light.

The detection unit 24 has a spectroscope 241 and a photodetector 242. The interference light emitted from the fourth optical fiber 254 is split into wavelength components by the spectroscope 241 and enters the photodetector 242. The photodetector 242 detects the split interference light and outputs the detection signal to the computer 30.

The image acquisition unit 41 of the computer 30, which will be described later, performs a Fourier transform on the detection signal obtained from the photodetector 242 to obtain the vertical light intensity distribution of the observation light. In addition, the scanning mechanism 223 moves the object optical system 22 in the horizontal direction while repeating the calculation of the above light intensity distribution, thereby obtaining the light intensity distribution of the observation light at each coordinate in three-dimensional space. As a result, the computer 30 can obtain a tomographic image and a three-dimensional image of the embryo 9.

A tomographic image is data composed of multiple pixels arranged on two-dimensional coordinates, with a brightness value specified for each pixel. A three-dimensional image is data composed of multiple pixels (voxels) arranged on three-dimensional coordinates, with a brightness value specified for each pixel. In other words, both tomographic images and three-dimensional images are captured images in which brightness values are distributed on specified coordinates.

The computer 30 functions as a control unit that controls the operation of the imaging unit 20. The computer 30 also functions as a data processing unit that creates tomographic images and three-dimensional images based on the detection signals input from the imaging unit 20, and classifies the embryos 9 based on the obtained tomographic images and three-dimensional images.

Figure 2 is a control block diagram of the embryo classification device 1. As conceptually shown in Figure 2, the computer 30 has a processor 31 such as a CPU, a memory 32 such as a RAM, and a storage unit 33 such as a hard disk drive. Stored in the storage unit 33 are a control program CP1, which is a computer program for controlling the operation of each unit in the embryo classification device 1, and a data processing program CP2, which is a computer program for creating a tomographic image D1 and a three-dimensional image D2 to classify the embryo 9.

The control program CP1 and the data processing program CP2 are read from a storage medium readable by the computer 30, such as a CD or DVD, and stored in the storage unit 33. However, the control program CP1 and the data processing program CP2 may also be downloaded to the computer 30 via a network.

As shown in FIG. 2, the computer 30 is communicatively connected to the light source 21, the scanning mechanism 223, the advance/retract mechanism 233, and the photodetector 242. The computer 30 is also communicatively connected to a display unit 70 such as a liquid crystal display. The computer 30 controls the operation of each of the above units according to the control program CP1. This allows the process of photographing the embryo 9 held in the sample container 90 to proceed.

Figure 3 is a block diagram conceptually showing the functions of the computer 30 for classifying the embryos 9. As shown in Figure 3, the computer 30 has an image acquisition unit 41, a region extraction unit 42, a volume calculation unit 43, a first index value calculation unit 44, a second index value calculation unit 45, and a classification unit 46. The functions of the image acquisition unit 41, the region extraction unit 42, the first index value calculation unit 44, the second index value calculation unit 45, and the classification unit 46 are realized by the processor 31 of the computer 30 operating in accordance with the above-mentioned data processing program CP2. Details of the processes executed by the image acquisition unit 41, the region extraction unit 42, the first index value calculation unit 44, the second index value calculation unit 45, and the classification unit 46 will be described later.

<2. Classification of embryos>
Next, the classification process of the embryo 9 using the above-mentioned embryo classification device 1 will be described.

Figure 4 shows the structure of an embryo 9 in the cleavage stage. The embryo 9 is transparent or translucent. As shown in Figure 4, the embryo 9 in the cleavage stage has multiple cells, or blastomeres 92, contained within a spherical zona pellucida 91. As cleavage progresses, the number of blastomeres 92 increases to 2, 4, 8, and so on. When the culture of the embryo 9 is progressing well, the multiple blastomeres 92 are of approximately the same size.

Also, as shown in FIG. 4, the embryo 9 may have fragmentation 93 in the zona pellucida 91. The fragmentation 93 is a collection of small cell pieces formed by irregular cell division. When the culture of the embryo 9 is progressing well, the proportion of fragmentation 93 contained in the embryo 9 is small.

To select embryos 9 that are being cultured well, this embryo classification device 1 classifies the state of the embryos 9 based on the uniformity of the volume of the blastomeres 92 and the proportion of fragmentation 93 in the embryos 9.

Figure 5 is a flow chart showing the flow of the classification process of the embryo 9. When classifying the embryo 9 in the embryo classification device 1, first, a sample container 90 is set on the stage 10 (step S1, preparation process). The embryo 9 is held in the sample container 90 together with the culture fluid.

Next, the embryo classification device 1 photographs the embryo 9 using the imaging unit 20 (step S2, image acquisition process). The imaging unit 20 performs optical coherence tomography. Specifically, light is emitted from the light source 21, and the object optical system 22 is moved slightly by the scanning mechanism 223 while the interference light of the observation light and the reference light is detected by the photodetector 242 for each wavelength component. The image acquisition unit 41 of the computer 30 calculates the light intensity distribution at each coordinate position of the embryo 9 based on the detection signal output from the photodetector 242. This allows a tomographic image D1 and a three-dimensional image D2 of the embryo 9 to be obtained.

The embryo classification device 1 acquires multiple tomographic images D1 and one three-dimensional image D2 for one embryo 9. The embryo classification device 1 also acquires the multiple tomographic images D1 and three-dimensional images D2 of the embryos 9 by repeating the process of step S2 while changing the well 901 to be photographed. The acquired tomographic images D1 and three-dimensional images D2 are stored in the memory unit 33 of the computer 30. The computer 30 also displays the acquired tomographic images D1 and three-dimensional images D2 on the display unit 70.

Next, the region extraction unit 42 of the computer 30 extracts multiple regions in the three-dimensional image D2 of the embryo 9 (step S3, region extraction process). Specifically, the region extraction unit 42 extracts, in the three-dimensional image D2 of the embryo 9, the entire embryo region occupied by the entire embryo 9, the zona pellucida region occupied by the zona pellucida 91, the blastomere assembly region occupied by multiple blastomeres 92, and the fragmentation region occupied by the fragmentation 93. Each of these regions can be extracted using, for example, the well-known Local Thickness method.

The region extraction unit 42 also identifies the boundaries of each blastomere 92 in the blastomere assembly region described above. This divides the blastomere assembly region into blastomere regions, which are regions for each blastomere 92. The boundaries of the blastomere 92 can be identified, for example, using a known Watershed algorithm. When the blastomere regions corresponding to the individual blastomeres 92 are identified in the three-dimensional image D2, the computer 30 counts the number of blastomere regions (i.e., the number of blastomeres 92 in the embryo 9) and stores the number in the memory unit 33.

The computer 30 stores the entire embryo region, the zona pellucida region, the multiple blastomere regions, and the fragmentation region extracted by the region extraction unit 42 in the memory unit 33. The computer 30 may also display the zona pellucida region, the multiple blastomere regions, and the fragmentation region in the three-dimensional image D2 displayed on the display unit 70, for example by color-coding each region.

Then, the volume calculation unit 43 of the computer 30 calculates the volume of each of the extracted regions (step S4, volume calculation process). Specifically, the volume calculation unit 43 calculates the volume of the entire embryo region, the volume of the zona pellucida region, the volume of each blastomere region, and the volume of the fragmentation region. The volume of each region can be represented, for example, by the number of voxels that make up the region. However, the volume calculation unit 43 may calculate the volume of each region in real space by multiplying the volume per voxel in real space by the number of voxels.

Next, the first index value calculation unit 44 of the computer 30 calculates a first index value P1 indicating the variation in the volume of the blastomere 92 in the embryo 9 (step S5, first index value calculation process). Here, the first index value calculation unit 44 calculates the first index value P1 indicating the variation in the volume of the blastomere 92 based on the volume of each blastomere region calculated in the above-mentioned step S4.

Figure 6 is a flowchart showing the flow of the calculation process of the first index value P1. As shown in Figure 6, the first index value calculation unit 44 first determines whether or not the number of blastomere regions extracted in the above-mentioned step S3 is a power of 2 (step S51). For example, the number of blastomere regions is 2, 4, or 8 when the number is a power of 2. For example, the number of blastomere regions is 3, 5, 6, or 7 when the number is not a power of 2.

If the number of blastomere regions is not a power of 2 (step S51: No), it is considered that the embryo 9 contains a mixture of blastomeres 92 after cleavage and blastomeres 92 immediately before cleavage. In this case, the first index value calculation unit 44 adjusts the blastomere regions (step S52).

Figure 7 is a diagram showing how the blastomere regions are adjusted in the three-dimensional image D2. As shown by the dashed line in Figure 7, the first index value calculation unit 44 divides some of the extracted blastomere regions 92A that have a large volume in half. In other words, among the multiple blastomere regions 92, the blastomere 92 immediately before division is treated as having been divided. This changes the number of blastomere regions 92A to a power of two.

As shown in FIG. 7, when the number of blastomere regions 92A is three, the first index value calculation unit 44 changes the number of blastomere regions 92A to four by dividing the largest blastomere region 92A in half. When the number of blastomere regions is five, the first index value calculation unit 44 changes the number of blastomere regions to eight by dividing the three largest blastomere regions in half. When the number of blastomere regions is six, the first index value calculation unit 44 changes the number of blastomere regions to eight by dividing the two largest blastomere regions in half. When the number of blastomere regions is seven, the first index value calculation unit 44 changes the number of blastomere regions to eight by dividing the largest blastomere region in half.

In step S51, if the number of blastomere regions corresponds to a power of 2 (step S51: Yes), or if the number of blastomere regions is changed to a power of 2 by adjusting the blastomere regions in step S52, the first index value calculation unit 44 then calculates a first index value P1 based on the volumes of the multiple blastomere regions (step S53). Three methods for calculating the first index value P1 are described below.

First, a first method for calculating the first index value P1 will be described. Fig. 8 is a diagram for explaining the first method by taking an example in which the number of blastomere regions is four. In the first method, the first index value P1 is calculated based on the difference Vmax - Vmin between the volume Vmax of the largest blastomere region and the volume Vmin of the smallest blastomere region. Specifically, the average value of the volume of the blastomere regions in the embryo 9 is Vm, and the number of blastomere regions is N, and the first index value P1 is calculated by the following formula (1).
P1=(Vmax/Vm-Vmin/Vm)/N (1)

Normally, as the generation of the embryo 9 progresses through cleavage, the average volume of the blastomeres 92 becomes smaller. For this reason, in the above formula (1), the volumes Vmax and Vmin of the blastomere regions are divided by the average value Vm to standardize (normalize) the volumes. This suppresses the fluctuation of the first index value P1 due to the cleavage generation. Also, normally, the difference value between the maximum and minimum values of the volume of the blastomere regions fluctuates according to the total number of blastomeres 92. For this reason, in the above formula (1), the standardized difference value (Vmax/Vm-Vmin/Vm) is divided by the number N of blastomere regions to calculate the percentage of variation in the volume of the blastomere regions.

Next, a second method for calculating the first index value P1 will be described. Fig. 9 is a diagram for explaining the second method, taking an example in which the number of blastomere regions is four. In the second method, the first index value P1 is calculated based on the cumulative value of the volumes of the blastomere regions. Specifically, the volume of the nth blastomere region in the embryo 9 is Vn, the average volume of the blastomere regions is Vm, and the number of blastomere regions is N, and the first index value P1 is calculated by the following formula (2).
P1=(Σ(Vn/Vm)−Vmin/Vm)/N (2)

The difference from the first method is that the total volume of the blastomere regions in the embryo 9, ΣVn, is used instead of the volume Vmax of the largest blastomere region. In this way, it is possible to take into account not only the maximum and minimum values of the blastomere region volume, but also the volumes of the blastomere regions with sizes between them. Therefore, the first index value P1, which indicates the variation in the volume of the blastomere regions, can be calculated with greater accuracy.

Next, a third method for calculating the first index value P1 will be described. Fig. 10 is a diagram for explaining the third method, taking as an example a case where the number of blastomere regions is four. In the third method, the first index value P1 is calculated based on the difference |Vm-Vn| between the volume Vn of each blastomere region and the average value Vm. Specifically, the volume of the nth blastomere region in the embryo 9 is Vn, the average value of the volumes of the blastomere regions is Vm, and the number of blastomere regions is N, and the first index value P1 is calculated by the following formula (1).
P1=(Σ(|Vm-Vn|/Vm)/N ... (3)

According to this third method, the deviation amount |Vm-Vn| from the average volume Vm is calculated for each blastomere region, and the deviation amounts for all blastomere regions are summed up. This allows the deviation amount from the average volume of all cell regions to be taken into consideration. Therefore, the first index value P1, which indicates the variation in the volume of the blastomere regions, can be calculated with greater accuracy.

As described above, three methods for calculating the first index value P1 have been described, but the method for calculating the first index value P1 is not limited to the above-mentioned first to third methods. For example, the median, minimum, or maximum value may be used instead of the above-mentioned average value Vm. Also, the first index value P1 indicating the variation in the volume of the blastomere region may be calculated using standard deviation, variance, or the like.

The calculated first index value P1 is stored in the memory unit 33 of the computer 30. The computer 30 also displays the calculated first index value P1 on the display unit 70.

Next, the second index value calculation unit 45 of the computer 30 calculates a second index value P2 indicating the proportion of fragmentation 93 in the embryo 9 (step S6, second index value calculation process). Here, the second index value P2 is calculated based on the volume of the fragmentation region calculated in the above-mentioned step S4.

The second index value calculation unit 45 calculates the second index value P2, for example, by dividing the volume of the fragmentation region by the volume of the entire embryo region. However, the second index value calculation unit 45 may calculate the second index value P2 by dividing the volume of the fragmentation region by the total volume of all blastomere regions (the volume of the blastomere assembly region). The second index value calculation unit 45 may also calculate the second index value P2 by dividing the volume of the fragmentation region by the total volume of the blastomere assembly region and the volume of the fragmentation region.

Then, the classification unit 46 classifies the embryo 9 based on the first index value P1 and the second index value P2 (step S7, classification process). In this embodiment, the classification unit 46 performs so-called Veeck classification, classifying the state of the embryo 9 into five grades, grade 1 to grade 5, based on the first index value P1 and the second index value P2.

As shown in FIG. 3, the classification unit 46 has a trained model M that has been created in advance. Specifically, the trained model M is stored in the memory unit 33 of the computer 30. The trained model M is an inference model for estimating the classification result of the embryo 9 (the Veeck classification result of grades 1 to 5) based on the first index value P1 and the second index value P2. That is, the input variables of the trained model M are the first index value P1 and the second index value P2. The output variable of the trained model M is the classification result of the embryo 9.

Prior to the classification process of FIG. 5, the computer 30 performs a learning process to create a trained model M. Specifically, the computer 30 learns the relationship between the first index value P1 and the second index value P2 and the classification result of the embryo 9 based on a supervised machine learning algorithm. At that time, a set of the first index value P1 and the second index value P2 and the corresponding known classification result is used as training data. Then, a large amount of known training data is prepared, and supervised machine learning is performed. As a result, a trained model M is generated that can accurately output the classification result of the embryo 9 by inputting the first index value P1 and the second index value P2.

In step S7, the classification unit 46 inputs the first index value P1 and the second index value P2 to the trained model M. The classification result is then output from the trained model M. The classification unit 46 classifies the embryo 9 into one of grades 1 to 5 based on the classification result output from the trained model M. The computer 30 stores the classification result of the embryo 9 in the memory unit 33. The computer 30 also displays the classification result of the embryo 9 on the display unit 70.

In this manner, in this embodiment, the computer 30 calculates a first index value P1 indicating the variation in the volume of the blastomere 92 and a second index value P2 indicating the proportion of fragmentation 93 based on the three-dimensional image D2. The computer 30 then classifies the embryo 9 based on the first index value P1 and the second index value P2. For this reason, the Veeck classification of the embryo 9, which was conventionally performed by the embryologist based on a two-dimensional microscope image, can be quantitatively performed based on the three-dimensional image D2. This makes it possible to obtain a highly accurate Veeck classification result based on the three-dimensional image D2. Therefore, by using the classification method of this embodiment, it is possible to improve the results of blastocyst achievement rate, implantation rate, pregnancy rate, and the like in assisted reproductive technology. In addition, it is possible to reduce the burden on the embryologist and obtain an objective Veeck classification result that is not dependent on the skill of the embryologist.

In particular, in this embodiment, the computer 30 performs Veeck classification of the embryos 9 using a trained model M generated by machine learning. In this way, the embryos 9 can be classified without the user having to set a threshold for classification.

3. Modifications
Although one embodiment of the present invention has been described above, the present invention is not limited to the above embodiment.

<3-1. First modified example>
In the above embodiment, the first index value P1 and the second index value P2 are input to the trained model M to output the classification result. However, the classification unit 46 may classify the embryo 9 based on the first index value P1 and the second index value P2 and a preset threshold value. For example, a first threshold value that the first index value P1 should clear and a second threshold value that the second index value P2 should clear may be set for each grade of the classification result. In that case, when the first index value P1 clears the first threshold value and the second index value P2 clears the second threshold value, the classification unit 46 may classify the embryo 9 into the grade of the threshold value.

Alternatively, the classification unit 46 may determine a provisional grade of the embryo 9 based on the first index value P1, based on the first index value P1 and the first threshold value. The classification unit 46 may also determine a provisional grade of the embryo 9 based on the second index value P2, based on the second index value P2 and the second threshold value. The classification unit 46 may then determine a final grade of the embryo 9 based on the above two provisional grades.

<3-2. Second modified example>
In the above embodiment, the classification unit 46 performs Veeck classification of the embryos 9. However, the classification unit 46 may classify the embryos 9 in a classification manner other than the Veeck classification. For example, the classification unit 46 may classify the embryos 9 into only two types, a "good embryo" and a "non-good embryo", based on the first index value P1 and the second index value P2.

<3-3. Other Modifications>
In the above embodiment, the first index value P1 is calculated, and then the second index value P2 is calculated. However, the order of calculation of the first index value P1 and the second index value P2 may be reversed. Furthermore, the first index value P1 and the second index value P2 may be calculated in parallel at the same timing.

In the above embodiment, the sample container 90 was a well plate having multiple wells (recesses) 901. Each well 901 held one embryo 9. However, the sample container 90 that holds the embryo 9 is not limited to a well plate. For example, the sample container 90 may be a dish having only one recess.

Furthermore, the elements appearing in the above embodiments and variations may be combined as appropriate to the extent that no contradictions arise.

REFERENCE SIGNS LIST 1 Embryo classification device 9 Embryo 10 Stage 20 Imaging unit 21 Light source 22 Object optical system 23 Reference optical system 24 Detection unit 25 Optical fiber coupler 30 Computer 41 Image acquisition unit 42 Region extraction unit 43 Volume calculation unit 44 First index value calculation unit 45 Second index value calculation unit 46 Classification unit 70 Display unit 90 Sample container 91 Zona pellucida 92 Blastomere 93 Fragmentation CP1 Control program CP2 Data processing program D1 Tomographic image D2 Three-dimensional image M Trained model P1 First index value P2 Second index value

Claims (8)

A method for classifying cleavage stage embryos, comprising:
a) obtaining a three-dimensional image of the embryo by optical coherence tomography;
b) extracting blastomere regions occupied by individual blastomeres and fragmentation regions occupied by fragmentations in the three-dimensional image;
c) calculating the volume of each of the blastomere regions and the volume of the fragmentation region;
d) calculating a first index value indicating the variation in the volume of the blastomere in the embryo based on the volume of the blastomere region;
e) calculating a second index value indicating the proportion of fragmentation in the embryo based on the volume of the fragmentation region;
f) classifying the embryo based on the first index value and the second index value;
A classification method having the following features: 2. The classification method according to claim 1,
In the step f), the embryos are classified into five grades, from grade 1 to grade 5, based on the first index value and the second index value. 2. The classification method according to claim 1,
In the step f), the embryos are classified into two types, that is, good embryos and non-good embryos, based on the first index value and the second index value. 2. The classification method according to claim 1,
The method further comprises a step of generating a trained model by learning a relationship between the first index value and the second index value and a classification result of the embryo based on a machine learning algorithm,
In the step f), a computer inputs the first index value and the second index value into the trained model, and classifies the embryo based on the classification result output from the trained model. 2. The classification method according to claim 1,
In the step f), the embryo is classified based on the first index value, the second index value and a preset threshold value. 2. The classification method according to claim 1,
In the step d), if the number of blastomere regions is not a power of two, the number of blastomere regions is changed to a power of two by dividing some blastomere regions with a large volume into two blastomere regions, and the first index value is calculated for the multiple blastomere regions after the change. 2. The classification method according to claim 1,
A classification method, in which in the step d), the volume of the blastomere region is standardized by dividing it by an average value or a median value, and the first index value is calculated based on the standardized volume. A computer program for causing a computer to execute the classification method according to any one of claims 1 to 7.

Images