Genotype estimation device and genotype estimation program relating to eye disease

Abstract

To provide a device and a program for accurately estimating the genotype of a patient developing a hereditary eye disease in a short time.SOLUTION: A genotype estimation device includes an estimation unit that estimates a genotype related to an eye disease of a patient, and the estimation unit estimates the genotype of the patient from image information of an eye of the patient by using a machine learning model generated based on training data including genetic information of the patient and the image information of the eye of the patient.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a device and a program for predicting gene mutations related to hereditary eye diseases such as corneal dystrophy. [Background technology]

[0002] Corneal dystrophy, a type of hereditary eye disease, is a general term for inherited corneal disorders that gradually cause the cornea to become cloudy, and includes a variety of diseases. Because treatment methods vary depending on the type of corneal dystrophy, it is necessary to identify the type of corneal dystrophy in order to provide appropriate treatment.

[0003] Traditionally, doctors have diagnosed the type of corneal dystrophy based on information such as the condition of the affected area observed with a slit lamp microscope, the age at onset of the disease, and family medical history. However, even for specialists, the accuracy of the diagnosis can vary depending on the knowledge and skill of each doctor.

[0004] Therefore, the only way to reliably identify the type of corneal dystrophy is to analyze the patient's genes and identify the genetic mutation (which base is mutated and how) that is thought to be related to the onset of corneal dystrophy.Currently, a definitive diagnosis can only be made after a doctor's diagnosis and genetic analysis.

[0005] In order to analyze a patient's genes, blood must be taken from the patient, and it can take several months to obtain the results of the gene analysis, which can delay the time it takes to provide the patient with appropriate treatment. There is also the problem of the cost involved in gene analysis.

[0006] As a method for predicting ophthalmological diseases, including the above-mentioned corneal dystrophy, development is underway on a diagnostic system that uses machine learning to diagnose diseases from images of a patient's anterior eye, as disclosed in Patent Document 1, for example.

[0007] However, while conventional disease diagnostic systems such as that described in Patent Document 1 may be able to roughly estimate the type of disease, such as whether the disease is an infectious inflammatory disease, a non-infectious inflammatory disease including corneal dystrophy, or some other disease, no system has yet been developed that can accurately estimate the genotype of a patient with a hereditary eye disease. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] US2021 / 0244272A1 Summary of the Invention [Problem to be solved by the invention]

[0009] The present invention has been made in view of the above-mentioned problems, and aims to provide an apparatus and a program for accurately estimating the genotype of a patient with a hereditary eye disease in a short period of time. [Means for solving the problem]

[0010] In other words, the genotype estimation device of the present invention is characterized in that it includes an estimation unit that estimates a patient's genotype related to an eye disease, and the estimation unit estimates the patient's genotype from image information of the patient's eye using a machine learning model generated based on training data including the patient's genetic information and image information of the patient's eye.

[0011] In a specific embodiment of the present invention, the image information may be image data of the anterior segment of a patient's eye or parameters calculated from the image data, where the anterior segment refers to the front half of the eye, including the cornea, iris, and angle of the eye.

[0012] The parameters can include, for example, one or more of the following values ​​(i.e., values ​​obtained by quantifying or encoding these indicators): the area of ​​the opacity (e.g., the area on the image data when the eye is photographed from the front), the depth of the opacity (position information of the opacity in the thickness direction of the cornea), the density of the opacity in the peripheral part of the cornea, the shape of the opacity in the peripheral part of the cornea, the shape of the edge of the opacity, etc.

[0013] In conventional ophthalmologic diagnoses, it is common to use image information obtained by a slit lamp microscope using light focused through a slit to obtain information on the depth of ocular opacity, etc. Based on this common technical knowledge, the present inventors have conducted extensive research and have unexpectedly found that, in the diagnostic device according to the present invention, estimation accuracy can be further improved when image information obtained by an imaging device that uses light that is not focused through a slit or the like or diffused light is used as the image information. [Effects of the Invention]

[0014] According to the present invention, it is possible to provide an estimation device and an estimation program that can accurately estimate the genotype of a patient who has developed a hereditary eye disease in a short period of time. With such an estimation device and estimation program, it is possible to obtain an estimation result regarding the genotype related to an eye disease from image information of a patient's eye in just a few seconds. As a result, appropriate treatment can be provided to patients without having to wait several months for the results of genetic analysis, as has been the case in the past. [Brief explanation of the drawings]

[0015] [Figure 1] 1 is an overall schematic diagram of a genotype estimation device according to one embodiment of the present invention. [Figure 2] FIG. 1 is a schematic diagram showing a procedure for generating a machine learning model according to the present embodiment. [Figure 3] FIG. 2 is a schematic diagram showing the procedure for estimating a patient's genotype using the genotype estimation device according to the present embodiment. [Figure 4]FIG. 10 is a schematic diagram showing a machine learning device according to another embodiment of the present invention. [Figure 5] FIG. 1 is a diagram showing original image data used in an embodiment of the present invention. [Figure 6] 1 shows examples of genotypes included in training data used in one embodiment of the present invention. [Figure 7] FIG. 1 is a diagram showing the results of an experiment verifying the effectiveness of the genotype estimation method according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] An embodiment of the present invention will be described below with reference to the drawings.

[0017] <Configuration of the genotype estimation device according to this embodiment> The genotype estimation device 100 of this embodiment, as shown in FIG. 1, includes an image acquisition unit 1 that acquires original image data of a patient's eye, and an information processing unit 2 that receives the original image data output from the image acquisition unit 1 and estimates and outputs the patient's genotype based on the original image data.

[0018] The image acquisition unit 1 includes, for example, an imaging device 11 that captures an image of an affected area (anterior segment of the eye) of a patient. The original image data refers to, for example, unprocessed image data itself captured by the imaging device 11. The original image data is preferably a photographic image of the anterior segment of the patient's eye, and is preferably a photographic image captured by the imaging device 11 placed opposite (more preferably directly facing) the patient's eye. As the imaging device 11, for example, a slit lamp microscope, a commercially available digital camera, a camera attached to a smartphone, or the like can be used.

[0019] Any device capable of capturing an image of the patient's eye may be used as the imaging device 11, and for example, a commercially available digital camera may be used. The imaging device 11 may capture an image of the patient's eye using light narrowed by a slit or the like, as in a slit lamp microscope, or it may capture an image using light whose illumination width is not narrowed by a slit or the like, or diffused light.

[0020] The image acquisition unit 1 is configured to be able to transmit and receive data to and from the information processing unit 2 via wire or wirelessly.

[0021] The information processing unit 2 is, for example, a general-purpose computer having analog electrical circuits including buffers, amplifiers, etc., digital electrical circuits including a CPU, memory, DSP, etc., and an A / D converter, etc. interposed between them.

[0022] The information processing unit 2 is configured to function as a data receiving unit 21 that receives the above-mentioned original image data and an estimation unit 22 that estimates the patient's genotype based on the original image data received by the data receiving unit 21, by having the CPU and its peripheral devices work together in accordance with a predetermined program stored in memory.

[0023] The information processing unit 2 of this embodiment also functions as a memory unit 23 that stores and accumulates teacher data created based on data received by the data receiving unit 21, and a machine learning model generation unit 24 that generates a machine learning model by performing machine learning based on the teacher data accumulated in the memory unit 23. The estimation unit 22 is configured to estimate the genotype of the patient using the machine learning model generated by the machine learning model generation unit 24.

[0024] <Genotype estimation method using a gene estimation device according to this embodiment> The procedure for estimating a genotype by the genotype estimation device 100 configured in this manner is as follows: acquire training data for generating a machine learning model (P1, P2), create a machine learning model based on the training data (P4), and then estimate the patient's genotype from image information of the patient's actual eye based on this machine learning model (S1 to S3), as shown in Figures 2 and 3. Each of these steps will be described in detail below.

[0025] The training data used in this embodiment is a set of data including image data (previous diagnostic image data) created from the original image data of a patient's eye acquired by the image acquisition unit 1 and genetic information obtained from the genetic analysis results of the patient. The previous diagnostic image data and genetic information included in one data set are stored in the storage unit in a linked state. Note that the patient who provides the above-mentioned previous diagnostic image data and genetic information is a patient whose type of corneal dystrophy has been identified by a doctor's diagnosis based on genetic analysis.

[0026] In this embodiment, the past diagnostic image data is generated by performing predetermined processing on the original image data acquired by the image acquisition unit 1, and the past diagnostic image data can be generated, for example, by the data accepting unit 21. In this case, the data accepting unit 21 can be said to be a pre-processing unit.

[0027] In this embodiment, the past diagnostic image data used is the original image data cut into an arbitrary shape, such as a rectangle, so as to include as much of the area corresponding to the affected eye as possible and to include as little of the area other than the eye, such as the surface of the skin, as possible.

[0028] In this embodiment, the genetic information used is the patient's blood as a sample, and is a gene sequence that is believed to be related to the onset of the target disease whose genotype is to be estimated, in particular a DNA sequence that includes a mutation site in a specific gene and a site where a sequence difference has been observed.

[0029] Target diseases in this embodiment include, but are not limited to, corneal dystrophy and congenital achromatopsia, which are hereditary eye diseases that have been well studied. Examples of corneal dystrophies include epithelial and subepithelial corneal dystrophies, stroma corneal dystrophies, endothelial corneal dystrophies, TGFBI-related corneal dystrophies, etc. Examples of epithelial and subepithelial corneal dystrophies include epithelial basement membrane dystrophy (EBMD), epithelial recurrent erosive dystrophies (EREDs), subepithelial mucous corneal dystrophy (SMCD), Meesmann septum dystrophy (MECD), Lisch epithelial corneal dystrophy (LECD), gelatinous droplet corneal dystrophy (GDLD), etc. Examples of corneal dystrophies occurring in the stroma include Macular corneal dystrophy (MCD), Schnyder corneal dystrophy (SCD), congenital stromal corneal dystrophy (CSCD), Fleck corneal dystrophy (FCD), posterior amorphous corneal dystrophy (PACD), central Francois corneal dystrophy, and Predescemet's membrane corneal dystrophy (PDCD). Examples of corneal dystrophies occurring in the endothelium include Fuchs endothelial corneal dystrophy (FECD), posterior polymorphic corneal dystrophy (PPCD), congenital hereditary endothelial dystrophy (CHED), and X-linked endothelial corneal dystrophy (XECD). Examples of TGFBI-related corneal dystrophies include Reis-Bucklers corneal dystrophy, Thiel-Behnke corneal dystrophy (TBCD), lattice corneal dystrophy type 1, granular corneal dystrophy type 1 (GCD1), and granular corneal dystrophy type 2 (GCD2). In particular, with regard to TGFBI-related corneal dystrophy, genetic analysis has progressed in recent years and genotypes have been well organized, so the effects of the present invention are expected to be particularly pronounced.

[0030] The patient's genetic information may be input by the user to the information processing unit 2, or may be input directly to the information processing unit 2 from the genetic analysis device via wired or wireless communication, or via communication means such as the Internet.

[0031] Next, the data receiving unit 21 outputs a set of data including the past diagnostic image data and the genetic sequence of the patient from whom the past diagnostic image data was obtained to the memory unit 23 as training data, and the memory unit 23 stores and accumulates this.

[0032] When the training data as described above is accumulated in the memory unit 23, the machine learning model generation unit 24 generates a machine learning model regarding the correlation between past diagnostic image data and gene sequences based on the training data accumulated in the memory unit 23.

[0033] After the machine learning model is generated in this manner, the image acquisition unit 1 acquires raw image data of the eye of the patient whose genotype is to be actually estimated. The data accepting unit accepts the raw image data of the estimation target patient acquired in this manner and performs processing similar to that of the previously described past diagnostic image to generate estimation target image data (image information in this embodiment). When this estimation target image data is sent to the estimation unit 22, which estimates the genotype, the estimation unit 22 estimates a gene sequence based on the previously described machine learning model and the newly acquired estimation target image data, and outputs this as the genotype of the estimation target patient. In this embodiment, not only the most likely gene sequence but also the likelihood of this gene sequence (estimation accuracy) are output as the estimation result.

[0034] The likelihood (estimation accuracy) of a gene sequence can be calculated, for example, from the degree of agreement between the past diagnostic image data included in the training data and the estimation target image data.

[0035] <Effects of this embodiment> The genotype estimation device 100 according to the present embodiment configured as described above estimates a patient's genotype using training data including the patient's gene sequence and image data of the patient's eye, and therefore can help doctors provide appropriate treatment to patients with hereditary eye diseases such as corneal dystrophy, which require genetic analysis to provide appropriate treatment, without waiting for the results of genetic analysis. In this sense, the genotype estimation device 100 according to the present embodiment can also be said to be a diagnostic support device that assists doctors in making diagnoses. Furthermore, the estimation accuracy of the estimation result is also output as the estimation result, so that the estimation result can be easily used as information for doctors to decide on a treatment plan for the patient.

[0036] <Other embodiments of the present invention> The present invention is not limited to the above-described embodiment. For example, in the above embodiment, the image data is used as is as image information. However, parameters calculated from the image information may be used as image information. Examples of such parameters include the area of ​​the opacity (e.g., the area in the image data when the eye is photographed from the front), the depth of the opacity (position information of the opacity in the thickness direction of the cornea), the density of the opacity in the peripheral part of the cornea, the shape of the opacity in the peripheral part of the cornea, and the shape of the edge of the opacity, which are quantified or coded. Only one of these parameters may be used as the image information included in the training data and the image information of the patient whose genotype is to be estimated, or a combination of multiple parameters may be used. Furthermore, the parameters may be used in addition to the image data. The extraction of each parameter from the image data may be performed by the data accepting unit or manually by the user. Furthermore, the original image data may be used as image information without processing.

[0037] In addition to the genetic information listed here, other examples include amino acid sequences encoded by gene sequences believed to be related to the onset of the target disease, and information on amino acid substitutions in the amino acid sequences (substitution positions and types of substituted amino acids). The gene sequence may be a DNA sequence or an RNA sequence. Furthermore, the genetic information may be one type of information from the various types described above, or a combination of multiple types.

[0038] In this embodiment, the gene sequence is estimated as the genotype, but other genotypes may also be estimated, such as one or more of the amino acid sequence encoded by the gene sequence believed to be related to the onset of the target disease, amino acid substitution information in the amino acid sequence (substitution position and type of substituted amino acid), etc. Furthermore, information on the zygote of the allele (homozygote or heterozygote) may be output as the estimation result. The genetic information included in the training data and the genotype output as the estimation result may be the same type of information, or may be different types of information. The interpretation of the estimated genotype should be appropriate depending on the characteristics of the target disease and related genetic factors. It is also useful to consider the subject's kinship information when estimating and interpreting genotypes. Clinically, this is particularly useful in pedigree analysis and penetrance assessment. Pedigree analysis utilizes the genetic information and disease history of a subject's relatives to estimate the inheritance pattern of specific gene mutations. This can be useful in identifying diseases by considering whether it follows known inheritance patterns. Penetrance assessment analyzes patterns of disease occurrence within families and helps estimate the penetrance of specific genetic variants, providing a more precise understanding of genotype-phenotype relationships. The analysis method that takes into account the above-mentioned kinship information improves the accuracy of genotype estimation and enables more comprehensive genetic risk assessment. Clinically, this information can be used to comprehensively assess genetic risk and determine appropriate genetic counseling, preventive measures, and treatment plans for individual patients. Such genotype interpretation may be performed by a doctor who obtains the estimation result, or, for example, training data for generating a machine learning model may further include information on the patient's blood relationship (specifically, genetic information and disease history of relatives, disease onset patterns within the family, etc.). In this case, the data accepting unit may accept the patient's blood relationship information (specifically, genetic information and disease history of relatives, disease onset patterns within the family, etc.) in addition to the original image data as information about the estimation subject. Then, the estimation unit may estimate the genotype based on the machine learning model, the estimation subject image data, and the blood relationship information.

[0039] In the above embodiment, the data accepting unit functions as a preprocessing unit, but the image acquisition unit may also include an information processing unit that functions as a preprocessing unit and outputs preprocessed image information and training data including the image information and genetic information to the information processing unit. Also, the user may manually preprocess image information in advance and input it to the information processing unit, or the user may prepare a dataset of training data in advance and input it to the information processing unit.

[0040] A part of the information processing unit may be, for example, a machine learning device configured as an independent server device or the like that can communicate with multiple measuring instruments via the Internet, or may collect training data from image acquisition units such as smartphones used by an unspecified number of users, and distribute estimation results or machine learning models estimated using a machine learning model to each of multiple image acquisition units.

[0041] In this case, the machine learning device may include, for example, a device as shown in Figure 4, which includes a data receiving unit that receives training data, a memory unit, and a machine learning model generation unit, accumulates training data output from multiple image acquisition units, generates a machine learning model, and outputs the generated machine learning model for each image acquisition unit.

[0042] In addition, some or all of the above-described embodiments and modified embodiments may be combined as appropriate, and it goes without saying that various modifications are possible within the scope of the spirit thereof. [Example]

[0043] The effects of the gene inference device and program according to the present invention will be explained below using more specific examples, but the present invention is not limited to these.

[0044] In this example, an experiment was conducted to estimate the genotype of a patient from image information of the patient regarding corneal dystrophy as an example of a hereditary eye disease.

[0045] First, as training data, the gene sequences of patients who had been definitively diagnosed with corneal dystrophy by a doctor based on genetic analysis were linked to the patient's anterior ocular segment image data (previous diagnostic image data) and input into the data receiving unit. Similar training data was obtained for over 2,600 patients and input into the data receiving unit. As shown in Figure 5, multiple image data were prepared for the anterior ocular segment, including those captured using a slit lamp microscope and those captured using different types of light, such as a light source capable of emitting diffused light and a commercially available digital camera. The specific conditions when using a slit lamp microscope are as follows: A slit lamp (TOPCON SL-D7) was used, and the light intensity of the lamp (TOPCON FD21) was set between 50 watts and 100 watt-seconds. While comparable results can be obtained within the aforementioned range, in this example, the light intensity was set to 75 watt-seconds. The light source incident angle was set to 30° to 60°, and images were taken using a diffuser lens at the maximum lamp aperture. The magnification was set to 75x to 200x. There is a range of conditions, but equivalent images can be obtained within this range.

[0046] Approximately 20 types of genotypes related to corneal dystrophy are known, and in this example, training data containing 15 of these genotypes (gene sequences) and corresponding past diagnostic image data was used, as shown in Figure 6.

[0047] Using the training data described above, the machine learning model generation unit generated a machine learning model. Next, using this machine learning model, the image data of the anterior segment of the patient to be estimated was input to the data reception unit, and the estimation unit estimated the most likely genotype. The results are shown in Figure 7, and in both cases, high estimation accuracy was demonstrated.

[0048] The precision in FIG. 7 indicates the ratio of the actual positive results to the positive results predicted by the model, as shown in the following formula (1). Precision rate = true positive / (true positive + false positive)...(1) The recall in Figure 7 is calculated using the following formula (2), which indicates the proportion of cases that the model predicted to be positive out of those that were actually positive. Recall rate = true positive / (true positive + false negative)...(2) The average precision (also called accuracy rate) in Figure 7 is an index that combines the precision and recall rates mentioned above, as shown in the following formula (3). The higher this average precision rate, the better the model performance. Average precision = (true positive + true negative) / (true positive + true negative + false positive + false negative) (3)

[0049] Even when using a slit lamp microscope and a light source commonly used in conventional examinations, under the conditions described above (i.e., when using unfocused or diffused light), we were able to confirm that the average accuracy rate (correct answer rate) of genotyping could be improved to 87%, as shown in Figure 7. Similarly, similar results were obtained when using a smartphone or other device capable of capturing images of the patient's anterior segment (raw image data) using unfocused or diffused light. One possible reason for this result is that the photographer's intention in capturing the raw image data is minimized, and the genotype can be estimated from objective images that include as much of the affected eye area as possible, rather than the conventional visual judgment (clinical diagnosis). [Explanation of symbols]

[0050] 100 Genotype estimation device 1. Image acquisition unit 2. Information Processing Unit 21 Data reception section 22...Estimation part 23...Storage section 24 Machine learning model generation section

Claims

1. an estimation unit for estimating a genotype related to a hereditary eye disease of a patient; A genotype estimation device characterized in that the estimation unit estimates the genotype of a patient from image information of the patient's eye using a machine learning model generated based on training data including the patient's genetic information and image information of the patient's eye.

2. The genotype estimation device according to claim 1 , wherein the image information includes image data of the anterior segment of the patient's eye and / or one or more parameters calculated from the image data.

3. The genotype estimation device according to claim 2 , wherein the parameters are values ​​relating to the area of ​​the opacity, the depth of the opacity, the density of the opacity in the peripheral part of the cornea, the shape of the opacity in the peripheral part of the cornea, or the shape of the edge of the opacity.

4. The genotype estimation device according to claim 1 , further comprising an imaging device that captures images using unfocused light or diffused light.

5. A genotype prediction program for predicting a genotype related to an eye disease of a patient, comprising: A genotype estimation program that causes a computer to function as an estimation unit that estimates a patient's genotype from image information of the patient's eye using a machine learning model generated based on training data including the patient's genetic information and image information of the patient's eye.

6. A machine learning device used in a genotype estimation device that estimates a genotype related to an eye disease of a patient, comprising: a data receiving unit that acquires training data including genetic information of a patient and image information of the patient's eye; and a machine learning model generation unit that generates a machine learning model for estimating a genotype related to a patient's eye disease from image data of the patient's eye based on the training data acquired by the data receiving unit.

Images