Transfer system and transfer method of learning model for cell image analysis

Abstract

Provided is a learning model transfer system and method for cell image analysis. The learning model transfer system is a system for transferring a learning model from a first learning device to a second learning device. The second learning device includes an algorithm matching determination unit that determines whether a match between a first algorithm and a second algorithm is achieved based on second algorithm determination information and first algorithm determination information, and a learning model parameter setting unit that sets a first parameter used with the second algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to a learning model transfer system for cell image analysis and a learning model transfer method for cell image analysis. BACKGROUND

[0002] In the past, a technology of generating a learning model for analyzing a cell image has been disclosed. Such a technology of generating a learning model for analyzing a cell image is disclosed, for example, in Japanese Patent Application Publication No. 2021-64115.

[0003] In Japanese Patent Application Publication No. 2021-64115, a cell analysis device is disclosed in which a learned model is generated by machine learning using learning data in which a cell image is an input image and a staining image in which a cell skeleton is stained is a correct answer image.

[0004] Here, although not disclosed in Japanese Patent Application Publication No. 2021-64115, it is sometimes desired to move a base, and the like, that is, to transfer a learning model for cell image analysis to a learning device (second learning device) that is arranged at a base different from a base in which a cell image analysis device (first learning device) in which a learned model is generated is arranged. In this case, by copying data of an algorithm of a learning model used in analysis of a cell image in the first learning device as a transfer source, a parameter (first parameter) used in the analysis, and the like, to the second learning device as a transfer destination, it is possible to transfer the learning model from the first learning device to the second learning device.

[0005] However, from the perspective of permission, and the like, it is sometimes not possible to directly copy the algorithm of the learning model. In this case, it is possible to transfer the learning model from the first learning device to the second learning device by storing the same algorithm as the algorithm of the first learning device in the second learning device as the transfer destination and transferring only the first parameter. However, in the case of transferring only the first parameter used in the analysis to the second learning device in which the same algorithm as the algorithm of the first learning device is stored, it is sometimes not possible to apply the first parameter to the second learning device due to a difference in the version of the algorithm, and the like. In this case, it is necessary to perform learning using the stored algorithm in the second learning device, and it is difficult to easily perform the transfer of the learning model. Therefore, a technology capable of easily transferring a learning model is desired. SUMMARY

[0006] The present application was made to solve the above-described problems, and an object of the present application is to provide a learning model transfer system for cell image analysis and a learning model transfer method for cell image analysis capable of easily transferring a learning model.

[0007] To achieve the above object, the transfer system of a learning model for cell image analysis according to a first aspect of the present application is a transfer system of a learning model for cell image analysis that transfers a learning model used in analysis of a cell image from a first learning device to a second learning device, wherein the second learning device includes: a transfer information input reception unit that receives input of learning model transfer information including first algorithm determination information for determining a first algorithm of a learning model used in analysis of a cell image and a first parameter that is a learning parameter output when the learning model is generated and is used when the cell image is analyzed by the learning model; a second learning device storage unit that stores a second algorithm of a learning model used in analysis of a cell image; an algorithm matching determination unit that determines whether matching of an estimated result of the first algorithm and the second algorithm using the first parameter is achieved based on the first algorithm determination information and second algorithm determination information for determining the second algorithm stored in the second learning device storage unit; a notification unit that notifies whether matching of the first algorithm and the second algorithm is achieved; and a learning model parameter setting unit that sets the first parameter used with the second algorithm in a case where matching of the first algorithm and the second algorithm is achieved.

[0008] The transfer method of a learning model for cell image analysis according to a second aspect of the present application is a transfer method of a learning model for cell image analysis for transferring a learning model used in analysis of a cell image from a first computer to a second computer, the transfer method of a learning model for cell image analysis including: a step in which the second computer receives input of learning model transfer information including first algorithm determination information for determining a first algorithm of a learning model used in analysis of a cell image and a first parameter that is a learning parameter output when the learning model is generated and is used when the cell image is analyzed by the learning model; a step in which the second computer determines whether matching of an estimated result of the first algorithm and the second algorithm using the first parameter is achieved based on a second algorithm stored in a storage unit and the first algorithm determination information included in the learning model transfer information; a step in which the second computer notifies whether matching of the first algorithm and the second algorithm is achieved; and a step in which the second computer sets the first parameter used with the second algorithm in a case where matching of the first algorithm and the second algorithm is achieved.

[0009] In the transfer system of the learning model for cell image analysis in the first aspect described above and the transfer method of the learning model for cell image analysis in the second aspect described above, it is determined whether or not the matching of the first algorithm of the first learning device and the second algorithm of the second learning device using the estimated result at the time when the first parameter for analyzing the cell image is achieved. In addition, in the case where the above-described matching of the first algorithm and the second algorithm is achieved, the first parameter is set in the second learning device. Thus, it is possible to set the first parameter in the second learning device without performing the same learning as the first learning device in the second learning device. As a result, it is possible to easily transfer the learning model. In addition, in the case where the matching of the first algorithm and the second algorithm is not achieved, it is notified that the matching of the first algorithm and the second algorithm is not achieved. Thus, the user can grasp the above-described matching of the first algorithm of the first learning device as the transfer source and the second algorithm of the second learning device as the transfer destination is not achieved, and can grasp the reason why the learning model cannot be transferred. As a result, it is possible to improve the convenience of the user. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 is a block diagram illustrating a transfer system of a learning model for cell image analysis of the present embodiment.

[0011] Figure 2 is a block diagram for explaining the structure of the first learning device.

[0012] Figure 3 is a diagram for explaining the learning model transfer information.

[0013] Figure 4 is a diagram for explaining the additional information.

[0014] Figure 5 is a block diagram for explaining the structure of the second learning device.

[0015] Figure 6 is a functional block diagram for explaining the learning model transfer information derivation function of the first learning device.

[0016] Figure 7 is a functional block diagram for explaining the learning model transfer information import function of the second learning device.

[0017] Figure 8 is a block diagram for explaining the determination processing of the algorithm matching determination section in the case where the version of the first algorithm and the version of the second algorithm are different.

[0018] Figure 9 is a diagram illustrating a screen example at the time of deriving the learning model transfer information.

[0019] Figure 10 is a diagram illustrating a screen example at the time of importing the learning model transfer information.

[0020] Figure 11 This is an example of a screen showing a successful import of learning model transfer information.

[0021] Figure 12 This is an example of a screen displaying a scenario where the import of learning model transfer information fails.

[0022] Figure 13 This is a flowchart illustrating the export process of learning model transfer information for the first learning device.

[0023] Figure 14 This is a flowchart illustrating the import process of learning model transfer information for the second learning device.

[0024] Figure 15 This is a block diagram illustrating the transfer method of the learning model used for cell image analysis of the modified example. Detailed Implementation

[0025] The embodiments embodied in the present invention will now be described with reference to the accompanying drawings.

[0026] Reference Figures 1-12 The structure of the cell image analysis learning model transfer system 200 and the cell image analysis learning model transfer method of this embodiment will be explained.

[0027] <Transfer System for Learning Models in Cell Image Analysis>

[0028] Figure 1 The cell image analysis learning model transfer system 200 shown is a system that transfers a learning model used in the analysis of cell image 90 from a first learning device 10 to a second learning device 20. The first learning device 10 is located at a first location 1. The second learning device 20 is located at a second location 2, different from the first location 1. The learning model is used to analyze and process cell images 90 captured by a user performing cell culture, etc.

[0029] The cell image analysis learning model transfer system 200 includes a first learning device 10 and a second learning device 20. Additionally, the cell image analysis learning model transfer system 200 includes a computer 110, a computer 111, a camera device 150, and a camera device 151.

[0030] exist Figure 1In the present embodiment, an example of transferring the learning model in a client-server system in which the first learning device 10 functions as a server and the computer 110 functions as a client terminal, and a client-server system in which the second learning device 20 functions as a server and the computer 111 functions as a client terminal is shown. The first learning device 10, the computer 110, and the imaging device 150 are connected in a manner capable of communicating with each other via the network 120. In addition, the second learning device 20, the computer 111, and the imaging device 151 are connected in a manner capable of communicating with each other via the network 121.

[0031] The first learning device 10 performs various information processing according to a request (processing request) from the computer 110 operated by the user. The first learning device 10 performs learning of the learning model for analyzing the cell image 90 according to the request. In the present embodiment, the first learning device 10 causes the learning model to learn classification of the cell photographed in the cell image 90. For example, the first learning device 10 causes the learning model to learn classification of whether the cell photographed in the cell image 90 is a normal cell, or classification of whether the cell photographed in the cell image 90 has aged.

[0032] In addition, the first learning device 10 performs output processing (export processing) of the learning model transfer information 30 according to the request. The operation of the first learning device 10, and display of the analysis result and the image after analysis obtained by the analysis using the first learning device 10 are accepted on the GUI (graphical user interface) displayed on the display section 130 of the computer 110.

[0033] The second learning device 20 performs various information processing according to a request (processing request) from the computer 111 operated by the user. The second learning device 20 performs learning of the learning model for analyzing the cell image 90 according to the request. In the present embodiment, the second learning device 20 causes the learning model to learn classification of the cell photographed in the cell image 90. For example, the second learning device 20 causes the learning model to learn classification of whether the cell photographed in the cell image 90 is a normal cell, or classification of whether the cell photographed in the cell image 90 has aged.

[0034] In addition, the second learning device 20 performs input processing (import processing) of the learning model transfer information 30 according to the request. The operation of the second learning device 20, and display of the analysis result and the image after analysis obtained by the analysis using the second learning device 20 are accepted on the GUI displayed on the display section 131 of the computer 111.

[0035] The network 120 connects the first learning device 10, the computer 110, and the imaging device 150 in a manner that enables them to communicate with each other. The network 120 can be, for example, a LAN (Local Area Network) constructed inside the first site 1. Thus, the second learning device 20 cannot be directly accessed from the computer 110.

[0036] The network 121 connects the second learning device 20, the computer 111, and the imaging device 151 in a manner that enables them to communicate with each other. The network 121 can be, for example, a LAN constructed inside the second site 2 that is different from the first site 1 in which the first learning device 10 is provided. Thus, the first learning device 10 cannot be directly accessed from the computer 111.

[0037] The computer 110 and the computer 111 are so-called personal computers that have a processor and a storage. The computer 110 is connected to the display 130 and the input 140. The computer 111 is connected to the display 131 and the input 141.

[0038] The display 130 and the display 131 are, for example, liquid crystal display devices. The display 130 and the display 131 can also be electroluminescent display devices, projectors, or head-mounted displays. In the present embodiment, the display 131 is configured to notify whether the matching of the first algorithm 40 (refer to Figure 5 ) and the second algorithm 50 (refer to Figure 7 ) is achieved. Specifically, the display 131 is configured to make the notification by displaying the notification content 91 (refer to ) described later. Note that the display 131 is an example of the "notification section" of the present application.

[0039] The input 140 and the input 141 are, for example, input devices including a mouse and a keyboard. The input 140 and the input 141 can also be touch panels. One or more computers 110 are provided in the learning model transfer system 200 for cell image analysis. One or more computers 111 are provided in the learning model transfer system 200 for cell image analysis.

[0040] The imaging device 150 and the imaging device 151 generate a cell image 90 obtained by photographing a cell. The imaging device 150 can transmit the generated cell image 90 to the computer 110 and / or the first learning device 10 via the network 120. The imaging device 150 photographs a microscope image of a cell. The imaging device 150 is image-processed by bright field observation, dark field observation, phase contrast observation, differential interference observation, or the like. One or more imaging devices 150 can be used depending on the imaging method. One or more imaging devices 150 can be provided in the learning model transfer system 200 for cell image analysis.

[0041] The camera device 151 is capable of sending the generated cell image 90 to the computer 111 and / or the second learning device 20 via the network 121. The camera device 151 captures microscopic images of the cells. The camera device 151 performs image processing using imaging methods such as bright-field observation, dark-field observation, phase-difference observation, and differential interferometry. One or more camera devices 151 can be used depending on the imaging method. One or more camera devices 151 can be provided in the cell image analysis learning model transfer system 200.

[0042] <First Learning Device>

[0043] like Figure 2 As shown, the first learning device 10 includes a first processor 11 and a first learning device storage unit 12.

[0044] The first processor 11 performs computational processing as a first learning device 10 by executing a predetermined program. The first processor 11 includes CPUs (Central Processing Units), GPUs (Graphics Processing Units), FPGAs (Field-Programmable Gate Arrays), and ASICs (Application Specific Integrated Circuits). In this embodiment, the first processor 11 is configured to learn a learning model using a first algorithm 40. Furthermore, the first processor 11 is configured to perform derivation processing of learning model transfer information 30. The structure of the first processor 11 performing the derivation processing of learning model transfer information 30 will be described in detail later.

[0045] The first learning device storage unit 12 includes a non-volatile storage device. The non-volatile storage device is, for example, a hard disk drive or a solid-state drive. The first learning device storage unit 12 is configured to store learning model transfer information 30. Additionally, the first learning device storage unit 12 stores a first algorithm 40 and a first preprocessing program 41. Furthermore, the first learning device storage unit 12 stores a cell image 90.

[0046] The first algorithm 40 is an algorithm for a learning model used in the analysis of the cell image 90. The first algorithm 40 is an analysis method used when analyzing the cell image 90. The first algorithm 40 includes, for example, U-NET. Furthermore, the first algorithm 40 is used when analyzing the cell image 90 in the first learning device 10.

[0047] The first preprocessing program 41 is a program of preprocessing of the cell image 90 at the time of learning the learning model. In the present embodiment, the first preprocessing program 41 includes, for example, a program for correcting unevenness of brightness generated in the background of the cell image 90. In addition, the first preprocessing program 41 is used in the preprocessing of the cell image 90 at the time of analyzing the cell image 90 in the first learning device 10.

[0048] 〈Learning model transfer information〉

[0049] Next, details of the learning model transfer information 30 will be described with reference to Figure 3 The learning model transfer information 30 includes first algorithm determination information 30a, first parameters 30b, first preprocessing program determination information 30c, second parameters 30d, and additional information 30e. Further, the first preprocessing program determination information 30c is an example of the "preprocessing program determination information" of the present application.

[0050] The first algorithm determination information 30a is information for determining the first algorithm 40. In the present embodiment, the first algorithm determination information 30a includes first identification information 31 corresponding to the first algorithm 40. In the present embodiment, the first identification information 31 includes a first hash value 32 generated based on the first algorithm 40.

[0051] The first hash value 32 is acquired by the first processor 11 and stored in the first learning device storage 12. The first processor 11 acquires the first hash value 32 from the first algorithm 40, for example, using a hash function. The first processor 11 acquires the first hash value 32 based on data that cannot be changed by the user, rather than data that can be changed by the user. That is, the first processor 11 acquires the first hash value 32 based on the source code or the like of the first algorithm 40, rather than the file name or the like of the first algorithm 40. Even in the case where the source code or the like of the first algorithm 40 has only one character different, the first hash value 32 becomes a value different from each other.

[0052] The first parameters 30b are learning parameters at the time of analyzing the cell image 90 by the learning model, which are output at the time of generating the learning model. Specifically, the first parameters 30b are weight parameters used at the time of analyzing the cell image 90. The first parameters 30b are acquired by the first processor 11 by learning of the learning model using the first algorithm 40.

[0053] The first preprocessing program determination information 30c is information for determining the first preprocessing program 41. In the present embodiment, the first preprocessing program determination information 30c includes first identification information 31 based on the first preprocessing program 41 (refer to FIG. 2). In the present embodiment, the first identification information 31 includes a first hash value 32 generated based on the first preprocessing program 41. Figure 2The first preprocessor hash value 33 is obtained, for example, by the first processor 11 using a hash function from the first preprocessor 41. The first processor 11 obtains the first preprocessor hash value 33 based on data that the user cannot change, rather than data that the user can change. That is, the first processor 11 obtains the first preprocessor hash value 33 based on the source code of the first preprocessor 41, rather than the filename of the first preprocessor 41. Even if only one character differs in the source code of the first preprocessor 41, the first preprocessor hash value 33 will be a different value.

[0054] The second parameter 30d is the parameter corresponding to the first preprocessing procedure 41. Specifically, the second parameter 30d is the parameter used by the first processor 11 to process the cell image 90 (refer to...) using the first preprocessing procedure 41. Figure 1 The value is set by the user during preprocessing.

[0055] Additional information 30e is information obtained when the first processor 11 performs the first algorithm 40. Additional information 30e includes information on the learning conditions set by the user when learning the learning model in the first learning device 10, as well as the verification results of the learning, etc.

[0056] <Additional Information>

[0057] like Figure 4 As shown, the additional information 30e includes at least one of the following: the name 34 of the dataset used in the learning model; the number 35 of the cell images 90 used in the learning model; the ratio 36 of training data, validation data, and test data within the data used in the learning model; the size 37 of the cell images 90 used in the learning model; and the metric value 38 when the learning model is being learned. In this embodiment, the additional information 30e includes all of the following: the name 34 of the dataset; the number 35 of the cell images 90; the ratio 36 of training data, validation data, and test data; the size 37 of the cell images 90; and the metric value 38 when the learning model is being learned.

[0058] <Second Learning Device>

[0059] like Figure 5 As shown, the second learning device 20 includes a second processor 21 and a second learning device storage unit 22.

[0060] The second processor 21 performs the operation processing as the second learning device 20 by executing a prescribed program. The second processor 21 includes a CPU, a GPU, an FPGA, an ASIC, or the like. In the present embodiment, the second processor 21 uses the second algorithm 50 to cause the learning model to learn the analysis of the cell image 90. In addition, the second processor 21 is configured to perform the import processing of the learning model transfer information 30. The structure of the second processor 21 performing the import processing of the learning model transfer information 30 is described in detail later.

[0061] The second learning device storage 22 includes a volatile storage device. The non-volatile storage device is, for example, a hard disk drive, a solid state drive, or the like. The second learning device storage 22 is configured to store the second algorithm 50 of the learning model used in the analysis of the cell image 90. In addition, the second learning device storage 22 is configured to store the second preprocessing program 51, which is a program for preprocessing the cell image 90. In addition, the second learning device storage 22 is configured to store the cell image 90.

[0062] The second algorithm 50 is an algorithm of the learning model used in the analysis of the cell image 90. The second algorithm 50 is the same algorithm as the first algorithm 40. In the present embodiment, the second algorithm 50 includes, for example, a U-NET. Further, the second algorithm 50 is acquired independently of the first algorithm 40 and is stored in the second learning device storage 22 in advance.

[0063] The second preprocessing program 51 is a program for preprocessing the cell image 90 when the learning model is learned. The second preprocessing program 51 is a program that performs the same preprocessing as the first preprocessing program 41. In the present embodiment, the second preprocessing program 51 includes, for example, a program for correcting brightness unevenness generated in the background of the cell image 90. Further, the second preprocessing program 51 is acquired independently of the first preprocessing program 41 and is stored in the second learning device storage 22 in advance.

[0064] 〈Export processing of learning model transfer information〉

[0065] Next, the structure of the first learning device 10 exporting the learning model transfer information 30 will be described with reference to Figure 6

[0066] As shown in Figure 6 , the first processor 11 has a transfer information generation section 11a, a transfer information encryption section 11b, and a transfer information output section 11c as functional blocks. In other words, the first processor 11 functions as the transfer information generation section 11a, the transfer information encryption section 11b, and the transfer information output section 11c by executing the program stored in the first learning device storage 12. ​

[0067] The transfer information generation section 11a generates the learning model transfer information 30 based on various information stored in the first learning device storage section 12. Specifically, the transfer information generation section 11a acquires the first algorithm determination information 30a, the first parameter 30b, the first preprocessing program determination information 30c, the second parameter 30d, and the additional information 30e from the first learning device storage section 12. In addition, the transfer information generation section 11a generates the learning model transfer information 30 by associating the acquired first algorithm determination information 30a, the first parameter 30b, the first preprocessing program determination information 30c, the second parameter 30d, and the additional information 30e.

[0068] The transfer information encryption section 11b is configured to encrypt the learning model transfer information 30 stored in the first learning device storage section 12. Specifically, the transfer information encryption section 11b acquires the learning model transfer information 30 generated by the transfer information generation section 11a and encrypts the learning model transfer information 30, thereby acquiring an encrypted file in a unique form that contains the encrypted learning model transfer information 39. In addition, the transfer information encryption section 11b outputs the encrypted learning model transfer information 39 to the transfer information output section 11c.

[0069] The transfer information output section 11c is configured to output the learning model transfer information 30 stored in the first learning device storage section 12 (refer to Figure 1 ). Specifically, the transfer information output section 11c is configured to output the learning model transfer information 39 encrypted by the transfer information encryption section 11b. In the present embodiment, the transfer information output section 11c acquires the encrypted learning model transfer information 39 from the transfer information encryption section 11b. In addition, the transfer information output section 11c outputs the acquired encrypted learning model transfer information 39 to the portable storage medium 60. The portable storage medium 60 includes, for example, a USB (Universal Serial Bus) memory.

[0070] The user stores the encrypted learning model transfer information 39 in the second learning device storage section 22 of the second learning device 20 using the portable storage medium 60 in which the encrypted learning model transfer information 39 is stored, thereby transferring the learning model transfer information 30 from the first learning device 10 to the second learning device 20.

[0071] 〈Importing Process of Learning Model Transfer Information〉

[0072] Next, the structure in which the second learning device 20 imports the learning model transfer information 30 will be described with reference to Figure 7

[0073] As described above, the second learning device 20 imports the learning model transfer information 30 from the first learning device 10 by acquiring the encrypted learning model transfer information 39 from the portable storage medium 60. Figure 7 ​As shown, the second processor 21 has a transfer information input accepting section 21a, a hash value acquiring section 21b, an algorithm matching determination section 21c, a preprocessing program matching determination section 21d, a notification content acquiring section 21e, a learning model parameter setting section 21f, and a preprocessing parameter setting section 21g as functional blocks. In other words, the second processor 21 functions as the transfer information input accepting section 21a, the hash value acquiring section 21b, the algorithm matching determination section 21c, the preprocessing program matching determination section 21d, the notification content acquiring section 21e, the learning model parameter setting section 21f, and the preprocessing parameter setting section 21g by executing the program stored in the second learning device storage section 22.

[0074] The transfer information input accepting section 21a accepts input of the learning model transfer information 30 (refer to FIG. 2) including the first algorithm determination information 30a and the first parameter 30b. In the present embodiment, the transfer information input accepting section 21a accepts input of the encrypted learning model transfer information 39 from the portable storage medium 60. In addition, the transfer information input accepting section 21a is configured to decrypt the learning model transfer information 39 encrypted by the transfer information encryption section 1 lb. That is, the transfer information input accepting section 21a acquires the learning model transfer information 30 by decrypting the encrypted learning model transfer information 39 input from the portable storage medium 60. Further, the transfer information input accepting section 21a is configured to be able to decrypt the file encrypted in a unique form by the transfer information encryption section 1 lb by executing the decryption program stored in the second learning device storage section 22. Figure 1

[0075] The transfer information input accepting section 21a outputs the first hash value 32 in the learning model transfer information 30 to the algorithm matching determination section 21c. In addition, the transfer information input accepting section 21a outputs the first preprocessing program hash value 33 in the learning model transfer information 30 to the preprocessing program matching determination section 21d.

[0076] The hash value acquiring section 21b is configured to acquire the second algorithm determination information 70a for determining the second algorithm 50. The second algorithm determination information 70a includes the second identification information 71 corresponding to the second algorithm 50. In the present embodiment, the second identification information 71 includes the second hash value 72 generated based on the second algorithm 50.

[0077] In addition, the hash value acquiring section 21b is configured to acquire the second preprocessing program determination information 70b for determining the second preprocessing program 51. In the present embodiment, the second preprocessing program determination information 70b includes the second preprocessing program hash value 73.

[0078] ​The hash value acquisition section 21b acquires the second algorithm 50 and the second preprocessing program 51 from the second learning device storage section 22. The hash value acquisition section 21b acquires the second hash value 72 from the second algorithm 50 using a hash function, for example. In addition, the hash value acquisition section 21b acquires the second preprocessing program hash value 73 from the second preprocessing program 51 using a hash function, for example. The hash value acquisition section 21b acquires the second hash value 72 and the second preprocessing program hash value 73 based on data that cannot be changed by the user, rather than data that can be changed by the user. That is, the hash value acquisition section 21b acquires the second hash value 72 based on the source code and the like of the second algorithm 50, rather than the file name and the like of the second algorithm 50. Even in a case where the source code and the like of the second algorithm 50 differ by only one character, the second hash value 72 becomes a value that is different from each other. In addition, the hash value acquisition section 21b acquires the second preprocessing program hash value 73 based on the source code and the like of the second preprocessing program 51, rather than the file name of the second preprocessing program 51. Even in a case where the source code and the like of the second preprocessing program 51 differ by only one character, the second preprocessing program hash value 73 becomes a value that is different from each other.

[0079] The hash value acquisition section 21b outputs the acquired second hash value 72 to the algorithm match determination section 21c. In addition, the hash value acquisition section 21b outputs the acquired second preprocessing program hash value 73 to the preprocessing program match determination section 21d.

[0080] The algorithm match determination section 21c is configured to determine whether or not the match of the estimated result of the first algorithm 40 and the second algorithm 50 when the first parameter 30b is used is obtained, based on the first algorithm determination information 30a and the second algorithm determination information 70a for determining the second algorithm 50 stored in the second learning device storage section 22.

[0081] In the present embodiment, the algorithm match determination section 21c is configured to determine the match of the first algorithm 40 and the second algorithm 50 by determining whether or not the first identification information 31 (refer to Figure 3 ) and the second identification information 71 coincide. Specifically, the algorithm match determination section 21c is configured to determine whether or not the first hash value 32 and the second hash value 72 coincide.

[0082] In addition, the algorithm match determination section 21c outputs the algorithm match determination result 90a to the notification content acquisition section 21e. Furthermore, in a case where the match of the first algorithm 40 (refer to Figure 2) With the matching of the second algorithm 50, the algorithm matching determination section 21c outputs "1" to the notification content acquisition section 21e, the learning model parameter setting section 21f, and the preprocessing parameter setting section 21g as the algorithm matching determination result 90a, for example. In addition, in a case where the matching of the first algorithm 40 and the second algorithm 50 is not obtained, the algorithm matching determination section 21c outputs "0 (zero)" to the notification content acquisition section 21e, the learning model parameter setting section 21f, and the preprocessing parameter setting section 21g as the algorithm matching determination result 90a, for example.

[0083] The preprocessing program matching determination section 21d is configured to determine whether or not the matching of the first preprocessing program 41 (refer to Figure 3 ) and the second preprocessing program 51 stored in the second learning device storage section 22 is obtained, based on the first preprocessing program determination information 30c (refer to Figure 2 ) and the second preprocessing program 51. Specifically, the preprocessing program matching determination section 21d determines whether or not the matching of the first preprocessing program 41 and the second preprocessing program 51 is obtained by determining whether or not the first preprocessing program hash value 33 and the second preprocessing program hash value 73 coincide.

[0084] In addition, the preprocessing program matching determination section 21d outputs the preprocessing program matching determination result 90b to the notification content acquisition section 21e. Further, in a case where the matching of the first preprocessing program 41 (refer to Figure 2 ) and the second preprocessing program 51 is obtained, the preprocessing program matching determination section 21d outputs "1" to the notification content acquisition section 21e, the learning model parameter setting section 21f, and the preprocessing parameter setting section 21g as the preprocessing program matching determination result 90b, for example. In addition, in a case where the matching of the first preprocessing program 41 and the second preprocessing program 51 is not obtained, the preprocessing program matching determination section 21d outputs "0 (zero)" to the notification content acquisition section 21e, the learning model parameter setting section 21f, and the preprocessing parameter setting section 21g as the preprocessing program matching determination result 90b, for example.

[0085] The learning model parameter setting section 21f is configured to set the learning model parameter 71 in the second learning device storage section 22 in a case where the matching of the first algorithm 40 (refer to Figure 2 ) and the second algorithm 50 (refer to Figure 5In the case of a match, a first parameter 30b is set for use with the second algorithm 50. Specifically, when "1" is input from the algorithm matching determination unit 21c to the learning model parameter setting unit 21f as an algorithm matching determination result 90a, the learning model parameter setting unit 21f outputs the first parameter 30b to the second learning device storage unit 22. More specifically, the learning model parameter setting unit 21f stores the first parameter 30b in a manner associated with the second algorithm 50 stored in the second learning device storage unit 22.

[0086] Furthermore, although a match was achieved between the first algorithm 40 and the second algorithm 50, the first preprocessing procedure 41 (see reference) was not obtained. Figure 2 ) and the second preprocessing procedure 51 (refer to Figure 5 In the case of a match between the first algorithm 40 and the second algorithm 50, and a match between the first preprocessor 41 and the second preprocessor 51, the learning model parameter setting unit 21f does not set the first parameter 30b. That is, the learning model parameter setting unit 21f sets the first parameter 30b only when both a match between the first algorithm 40 and the second algorithm 50 and a match between the first preprocessor 41 and the second preprocessor 51 are achieved.

[0087] The preprocessing parameter setting unit 21g is configured such that, after obtaining the first preprocessing procedure 41 (refer to...) Figure 2 ) and the second preprocessing procedure 51 (refer to Figure 5 If the preprocessing result matches the preprocessing result 90b, the second parameter 30d is set for the second preprocessing program 51. Specifically, if a "1" is input from the preprocessing program matching determination unit 21d to the preprocessing parameter setting unit 21g as the preprocessing program matching determination result 90b, the preprocessing parameter setting unit 21g outputs the second parameter 30d to the second learning device storage unit 22. More specifically, the preprocessing parameter setting unit 21g stores the second parameter 30d in a manner associated with the second preprocessing program 51 stored in the second learning device storage unit 22.

[0088] Furthermore, although a match was achieved between the first preprocessor 41 and the second preprocessor 51, the first algorithm 40 (see reference) was not obtained. Figure 2 ) and the second algorithm 50 (refer to) Figure 5 In the case of a match between the first preprocessing program 41 and the second preprocessing program 51, and a match between the first algorithm 40 and the second algorithm 50, the preprocessing parameter setting unit 21g does not set the second parameter 30d. That is, when a match is achieved between the first preprocessing program 41 and the second preprocessing program 51, and a match is achieved between the first algorithm 40 and the second algorithm 50, the preprocessing parameter setting unit 21g sets the second parameter 30d.

[0089] The notification content acquisition unit 21e is configured to acquire the notification content 91. Specifically, the notification content acquisition unit 21e acquires the notification content 91 based on the algorithm matchability determination result 90a input from the algorithm matchability determination unit 21c and the preprocessing program matchability determination result 90b input from the preprocessing program matchability determination unit 21d. In this embodiment, the notification content acquisition unit 21e acquires the notification content 91 based on the algorithm matchability determination result 90a and the preprocessing program matchability determination result 90b, and also based on whether the setting of the first parameter 30b and the second parameter 30d is successful.

[0090] The notification content 91 includes a message 91a indicating that the import of the learning model is successful (see FIG. 6A) and a message 91b indicating that the import of the learning model is unsuccessful (see FIG. 6B). In a case where the results of both the algorithm matchability determination result 90a and the preprocessing program matchability determination result 90b are that the match is obtained and the setting of the first parameter 30b and the second parameter 30d is successful, the notification content acquisition unit 21e acquires the message 91a indicating that the import of the learning model is successful as the notification content 91. That is, in a case where both the algorithm matchability determination result 90a and the preprocessing program matchability determination result 90b are “1” and the setting of the first parameter 30b and the second parameter 30d is successful, the notification content acquisition unit 21e acquires the message 91a indicating that the import of the learning model is successful as the notification content 91. Figure 11 ) and a message 91b indicating that the import of the learning model is unsuccessful (see Figure 12 ). In a case where the results of both the algorithm matchability determination result 90a and the preprocessing program matchability determination result 90b are that the match is obtained and the setting of the first parameter 30b and the second parameter 30d is successful, the notification content acquisition unit 21e acquires the message 91a indicating that the import of the learning model is successful as the notification content 91. That is, in a case where both the algorithm matchability determination result 90a and the preprocessing program matchability determination result 90b are “1” and the setting of the first parameter 30b and the second parameter 30d is successful, the notification content acquisition unit 21e acquires the message 91a indicating that the import of the learning model is successful as the notification content 91.

[0091] In a case where the result of either the algorithm matchability determination result 90a or the preprocessing program matchability determination result 90b is that the match is not obtained, or in a case where the setting of either the first parameter 30b or the second parameter 30d is unsuccessful, the notification content acquisition unit 21e acquires the message 91b indicating that the import of the learning model is unsuccessful as the notification content 91. That is, in a case where at least either the algorithm matchability determination result 90a or the preprocessing program matchability determination result 90b is “0 (zero)”, or in a case where the setting of either the first parameter 30b or the second parameter 30d is unsuccessful, the notification content acquisition unit 21e acquires the message 91b indicating that the import of the learning model is unsuccessful as the notification content 91. The notification content acquisition unit 21e outputs the acquired notification content 91 to the display unit 131.

[0092] The display unit 131 displays the notification content 91 acquired from the notification content acquisition unit 21e.

[0093] Compatibility of Algorithms

[0094] Next, the compatibility of algorithms will be described with reference to FIGS. 6A and 6B. Figure 8The structure in which the algorithm matching determination unit 21c determines whether or not the matching of the first algorithm 40 and the second algorithm 50 is achieved when at least either of the first algorithm 40 and the second algorithm 50 is upgraded will be described.

[0095] As shown in Figure 8 , regarding the first algorithm determination information 30a, in a case where the first algorithm 40 is upgraded, the first algorithm determination information 30a of the version having compatibility is included in the first algorithm determination information 30a after the upgrade. Figure 8 In the example shown in , as the first algorithm determination information 30a, three hash values of the first hash value 32a, the first hash value 32b, and the first hash value 32c are included.

[0096] Figure 8 In addition, as shown in Figure 8 , regarding the second algorithm determination information 70a, in a case where the second algorithm 50 is upgraded, the second algorithm determination information 70a of the version having compatibility is included in the second algorithm determination information 70a after the upgrade.

[0097] In the present embodiment, the algorithm matching determination unit 21c is configured to determine that the matching of the first algorithm 40 and the second algorithm 50 is achieved when the versions of the first algorithm 40 and the second algorithm 50 have compatibility with each other even in a case where the versions of the first algorithm 40 and the second algorithm 50 are different.

[0098] As shown in Figure 8 , even in a case where the first hash value 32a and the second hash value 72a are different, the algorithm matching determination unit 21c determines that the matching of the first algorithm 40 and the second algorithm 50 is achieved when the first hash value 32a and the second hash value 72b are equal to each other. That is, in a case where there are a plurality of the first hash values 32 and the second hash values 72, the algorithm matching determination unit 21c determines the matching of the first algorithm 40 and the second algorithm 50 depending on whether or not there is a hash value that is consistent with the second hash value 72 among the plurality of the first hash values 32 or whether or not there is a hash value that is consistent with the first hash value 32 among the plurality of the second hash values 72.

[0099] 〈Learning Model Transfer Information Export Screen〉

[0100] Next, a screen example at the time of exporting the learning model transfer information 30 will be described with reference to Figure 9

[0101] Figure 9 A screen in which the learning model transfer information 30 is displayed in the display unit 130 (refer to Figure 1 ​This is an example of a learning model transfer information export screen 130a. The learning model transfer information export screen 130a displays the learning model name 130b, dataset name 130c, and preprocessing information 130d. Additionally, the learning model transfer information export screen 130a displays a download button 130e and an OK button 130f. The download button 130e and the OK button 130f are buttons on the GUI.

[0102] By pressing the download button 130e, the first processor 11 executes... Figure 6 The learning model transfer information 30 shown (refer to) Figure 1 Export processing of ).

[0103] Additionally, by pressing the OK button 130f, the first processor 11 closes the learning model transfer information export screen 130a.

[0104] Next, refer to Figures 10-12 Here is a screenshot illustrating the process of importing and transferring information from the learning model (30).

[0105] Figure 10 The image shown is displayed on display unit 131 (see reference). Figure 1 The learning model import screen 131a displays a learning name input field 131b. Additionally, the learning model import screen 131a displays a file selection button 131c, a registration button 131d, and a cancel button 131e. These buttons are GUI buttons. Furthermore, the learning name is a name stored for learning processing during the analysis of cell images 90 in the second learning device 20.

[0106] By pressing the file selection button 131c, the second processor 21 is displayed in the execution... Figure 8 and Figure 9 The screen showing the import processing of the learning model transfer information 30 and the selection screen for the encrypted file of the encrypted learning model transfer information 39 when setting the first parameter 30b and the second parameter 30d is not shown.

[0107] By pressing the registration button 131d, the second processor 21 executes... Figure 8 The process of importing the learning model transfer information 30 shown, as well as the process of setting the first parameter 30b and the second parameter 30d.

[0108] Figure 11 The information is displayed on display unit 131 when the learning model transfer information 30 is successfully imported (see reference). Figure 1) is successfully imported. The message 91a indicating that the learning model is successfully imported is displayed in the import success screen 131f. The import success screen 131f is displayed as a pop-up screen on the learning model import screen 131a.

[0109] Figure 12 An example of an import failure screen 131g displayed in the display section 131 (refer to Figure 1 ) when the learning model transfer information 30 is not successfully imported. The message 91b indicating that the learning model is not successfully imported is displayed in the import failure screen 131g. The import failure screen 131g is displayed as a pop-up screen on the learning model import screen 131a.

[0110] Next, the process in which the first learning device 10 (refer to Figure 13 ) exports the learning model transfer information 30 (refer to Figure 1 ) will be described with reference to Figure 1 . In addition, the export process of the learning model transfer information 30 performed by the first learning device 10 is performed before the second learning device 20 (the second processor 21) receives the input of the learning model transfer information 30.

[0111] In step 101, the first learning device 10 (the first processor 11) generates the learning model transfer information 30. In addition, the first processor 11 stores the generated learning model transfer information 30 in the first learning device storage section 12.

[0112] In step 102, the transfer information encryption section 11b encrypts the learning model transfer information 30. In the present embodiment, the transfer information encryption section 11b acquires the encrypted learning model transfer information 39 by encrypting the learning model transfer information 30 stored in the first learning device storage section 12.

[0113] In step 103, the first learning device 10 (the transfer information output section 11c) outputs the learning model transfer information 30 stored in the first learning device storage section 12. Specifically, the transfer information output section 11c outputs the learning model transfer information 39 encrypted by the transfer information encryption section 11b. After that, the process ends.

[0114] Next, the process in which the second learning device 20 (refer to Figure 1 ) imports the learning model transfer information 30 (refer to Figure 1 ) will be described with reference to Figure 14 . The process in which the second learning device 20 imports the learning model transfer information 30 is performed after the learning model transfer information 30 is exported in the first learning device 10. In addition, the process in which the second learning device 20 imports the learning model transfer information 30 is performed by the user selecting the import button 131b on the learning model import screen 131a (refer toFigure 10 ) and presses the registration button 131d to start.

[0115] In step 201, the transfer information input accepting section 21a (refer to Figure 7 ) accepts input of the learning model transfer information 30 including the first algorithm determination information 30a (refer to Figure 3 ) for determining the first algorithm 40 (refer to Figure 3 ) of the learning model used in the analysis of the cell image 90 (refer to Figure 1 ) and the first parameter 30b (refer to Figure 2 ) that is a learning parameter output when the learning model is generated and used to analyze the cell image 90. In the present embodiment, the transfer information input accepting section 21a accepts input of the encrypted learning model transfer information 39 (refer to Figure 6 ).

[0116] In step 202, the transfer information input accepting section 21a (refer to Figure 7 ) acquires the learning model transfer information 30 (refer to Figure 1 ) by decrypting the encrypted learning model transfer information 39.

[0117] In step 203, the hash value acquiring section 21b (refer to Figure 7 ) acquires the second hash value 72 (refer to Figure 7 ).

[0118] In step 204, the algorithm matching determination section 21c (refer to Figure 7 ) determines whether or not the matching of the estimated results of the first algorithm 40 and the second algorithm 50 using the first parameter 30b is obtained, based on the second algorithm 50 (refer to Figure 5 ) stored in the second learning device storage section 22 (refer to Figure 5 ) and the first algorithm determination information 30a (refer to Figure 3 ) included in the learning model transfer information 30. In the present embodiment, the second processor 21 determines whether or not the matching of the first algorithm 40 and the second algorithm 50 is obtained by determining whether or not the first hash value 32 (refer to Figure 3 ) and the second hash value 72 (refer to Figure 7 ) coincide. In the case where the matching of the first algorithm 40 and the second algorithm 50 is obtained, the process proceeds to step 205. In the case where the matching of the first algorithm 40 and the second algorithm 50 is not obtained, the process proceeds to step 210.

[0119] In step 205, the hash value acquiring section 21b (refer to Figure 7) the second pre-processing program hash value 73 (refer to Figure 7 ).

[0120] In step 206, the pre-processing program matching determination section 21d (refer to Figure 7 ) determines whether matching of the first pre-processing program 41 (refer to Figure 2 ) and the second pre-processing program 51 (refer to Figure 5 ) is achieved. In the present embodiment, the pre-processing program matching determination section 21d determines whether the first pre-processing program 41 and the second pre-processing program 51 are identical by determining whether the first pre-processing program hash value 33 (refer to Figure 3 ) and the second pre-processing program hash value 73 (refer to Figure 7 ) are identical. In the case where the first pre-processing program 41 and the second pre-processing program 51 are identical, the processing proceeds to step 207. In the case where the first pre-processing program 41 and the second pre-processing program 51 are not identical, the processing proceeds to step 211.

[0121] In step 207, in the case where matching of the first algorithm 40 (refer to Figure 2 ) and the second algorithm 50 (refer to Figure 5 ) is achieved, the learning model parameter setting section 21f (refer to Figure 8 ) sets the first parameter 30b (refer to Figure 3 ) used with the second algorithm 50. In the present embodiment, the learning model parameter setting section 21f stores the second algorithm 50 and the first parameter 30b in association in the second learning device storage section 22 (refer to Figure 5 ), thereby setting the first parameter 30b.

[0122] In step 208, the pre-processing parameter setting section 21g (refer to Figure 8 ) sets the second parameter 30d in the second learning device 20. In the present embodiment, the pre-processing parameter setting section 21g stores the second pre-processing program 51 and the second parameter 30d in association in the second learning device storage section 22 (refer to Figure 5 ), thereby setting the second parameter 30d.

[0123] In step 209, it is determined whether the setting of the parameters is normally completed. In the case where the setting of the parameters is normally completed, the processing proceeds to step 210. In the case where the setting of the parameters is not normally completed, the processing proceeds to step 211.

[0124] In step 210, the display section 131 (refer to Figure 1 ) displays a message 91a (refer to Figure 11 ) indicating that the import was successful. Thereafter, the processing ends.

[0125] In the case where the process proceeds from step 204 or step 206 to step 211, in step 211, the display section 131 displays a message 91b indicating that the importation failed (refer to Figure 12 ). Thereafter, the process ends.

[0126] As shown in step 210 and step 211, the display section 131 (refer to Figure 1 ) notifies whether or not the matching of the first algorithm 40 (refer to Figure 2 ) and the second algorithm 50 (refer to Figure 5 ) is obtained. That is, in the case where the matching of the first algorithm 40 and the second algorithm 50 is obtained, the display section 131 displays a message 91a indicating that the importation succeeded. In addition, in the case where the matching of the first algorithm 40 and the second algorithm 50 is not obtained, the display section 131 displays a message 91b indicating that the importation failed.

[0127] (Effects of the Present Embodiment)

[0128] In the present embodiment, the following effects can be obtained.

[0129] In the present embodiment, as described above, the transfer system 200 of the learning model for cell image analysis is a transfer system of a learning model for cell image analysis that transfers the learning model used in the analysis of the cell image 90 from the first learning device 10 to the second learning device 20, wherein the second learning device 20 includes: a transfer information input reception section 21a that receives the input of the learning model transfer information 30 including the first algorithm determination information 30a for determining the first algorithm 40 of the learning model used in the analysis of the cell image 90 and the first parameter 30b that is output when the learning model is generated and is a learning parameter at the time of analysis of the cell image 90 by the learning model; a second learning device storage section 22 that stores the second algorithm 50 of the learning model used in the analysis of the cell image 90; an algorithm matching determination section 21c that determines whether or not the matching of the estimated result of the first algorithm 40 and the second algorithm 50 using the first parameter 30b is obtained, based on the first algorithm determination information 30a and the second algorithm determination information 70a for determining the second algorithm 50 stored in the second learning device storage section 22; a display section 131 that notifies whether or not the matching of the first algorithm 40 and the second algorithm 50 is obtained; and a learning model parameter setting section 21f that sets the first parameter 30b used together with the second algorithm 50 in the case where the matching of the first algorithm 40 and the second algorithm 50 is obtained.

[0130] Thus, since the algorithm matching propriety determination section 21c and the learning model parameter setting section 21f are provided, it is determined whether or not the matching of the estimation results of the first algorithm 40 of the first learning device 10 and the second algorithm 50 of the second learning device 20 using the first parameters 30b for analyzing the cell images 90 is achieved. In addition, in the case where the above matching of the first algorithm 40 and the second algorithm 50 is achieved, the first parameters 30b are set in the second learning device 20. Thus, the first parameters 30b can be set in the second learning device 20 without performing the same learning as the first learning device 10 in the second learning device 20. As a result, the transfer of the learning model can be easily performed. In addition, since the algorithm matching propriety determination section 21c and the display section 131 are provided, in the case where the matching of the first algorithm 40 and the second algorithm 50 is not achieved, it is notified that the matching of the first algorithm 40 and the second algorithm 50 is not achieved. Thus, the user can grasp the above matching of the first algorithm 40 of the first learning device 10 as the source of the transfer and the second algorithm 50 of the second learning device 20 as the destination of the transfer, and can grasp the reason why the learning model cannot be transferred. As a result, the convenience of the user can be improved.

[0131] In addition, in the present embodiment, as described above, the transfer method of the learning model for cell image analysis is a transfer method of a learning model for cell image analysis for transferring the learning model used in the analysis of the cell images 90 from the first computer (the first learning device 10) to the second computer (the second learning device 20), the transfer method of the learning model for cell image analysis including the steps of: the second computer (the second learning device 20) accepting the input of the learning model transfer information 30 including the first algorithm determination information 30a for determining the first algorithm 40 of the learning model used in the analysis of the cell images 90 and the first parameters 30b output when the learning model is generated and used in the analysis of the cell images 90; the second computer (the second learning device 20) determining whether or not the matching of the estimation results of the first algorithm 40 and the second algorithm 50 using the first parameters 30b is achieved based on the second algorithm 50 stored in the second learning device storage section 22 and the first algorithm determination information 30a included in the learning model transfer information 30; the second computer (the second learning device 20) notifying whether or not the matching of the first algorithm 40 and the second algorithm 50 is achieved; and in the case where the matching of the first algorithm 40 and the second algorithm 50 is achieved, the second computer (the second learning device 20) setting the first parameters 30b used together with the second algorithm 50.

[0132] Thus, the transfer method of the learning model for cell image analysis can be provided, which can easily transfer the learning model as with the transfer system 200 of the learning model for cell image analysis.

[0133] Further, in the above-described embodiment, by configuring as follows, the following effects can be further obtained.

[0134] That is, in the present embodiment, as described above, the first learning device 10 includes the first learning device storage section 12 that stores the learning model transfer information 30, and the transfer information output section 11c that outputs the learning model transfer information 30 stored in the first learning device storage section 12. Thereby, for example, by outputting the learning model transfer information 30 to the portable storage medium 60 or the like, the learning model transfer information 30 can be easily transferred from the first learning device 10 to the second learning device 20.

[0135] Further, in the present embodiment, as described above, the learning model transfer information 30 further includes preprocessing program determination information (first preprocessing program determination information 30c) for determining the first preprocessing program 41, which is a program for preprocessing of the cell image 90 at the time of learning the learning model, the second learning device storage section 22 is configured to further store a second preprocessing program 51, which is a program for preprocessing of the cell image 90, and the second learning device 20 further includes the preprocessing program matching determination section 21d that determines whether or not the matching of the results of preprocessing of the cell image 90 by the first preprocessing program 41 and the second preprocessing program 51 is obtained, based on the first preprocessing program determination information 30c and the second preprocessing program 51 stored in the second learning device storage section 22. Thereby, the user can easily grasp whether or not the above-described matching of the first preprocessing program 41 executed in the first learning device 10 and the second preprocessing program 51 in the second learning device 20 is obtained. As a result, the user can easily grasp whether or not the second parameters 30d for executing the preprocessing program can be transferred at the time of transferring the learning model from the first learning device 10 to the second learning device 20.

[0136] Further, in the present embodiment, as described above, the learning model transfer information 30 further includes the second parameter 30d corresponding to the first preprocessing program 41, and further includes the preprocessing parameter setting section 21g that sets the second parameter 30d for the second preprocessing program 51 in a case where matching of the preprocessed result of the first preprocessing program 41 and the preprocessed result of the second preprocessing program 51 is obtained. Thereby, the same preprocessing as the preprocessing performed by the first preprocessing program 41 can be performed by the second preprocessing program 51 without the user setting the second parameter 30d. As a result, compared to a structure in which the user sets the second parameter 30d, the user's burden can be reduced. Further, even in a case where it is difficult to copy the first preprocessing program 41 to the second learning device 20, by pre-storing the second preprocessing program 51 that is the same as the first preprocessing program 41 in the second learning device 20 and setting the second parameter 30d for the second learning device 20, the transfer of the preprocessing program can be easily performed.

[0137] Further, in the present embodiment, as described above, the learning model transfer information 30 further includes at least any one of the following: the name 34 of the data set used in the learning of the learning model; the number of images 35 of the cell images 90 used in the learning of the learning model; the proportions 36 of the training data, the validation data, and the test data within the data used in the learning of the learning model; the size 37 of the cell images 90 used in the learning of the learning model; and the index value 38 at the time of the learning of the learning model. Thereby, the user can confirm the name 34 of the data set, the number of images 35 of the cell images 90, the proportions 36 of the training data, the validation data, and the test data, the size 37 of the cell images 90, and the index value 38 in the second learning device 20. As a result, since the user can grasp how the learning was performed in the first learning device 10 in the second learning device 20, the convenience of the user can be improved.

[0138] Further, in the present embodiment, as described above, the first learning device 10 further includes the transfer information encryption section 1 lb that encrypts the learning model transfer information 30 stored in the first learning device storage section 12, the transfer information output section 1 lc is configured to output the learning model transfer information 39 encrypted by the transfer information encryption section 1 lb, and the transfer information input reception section 21a is configured to decrypt the learning model transfer information 39 encrypted by the transfer information encryption section 1 lb. Thereby, it is possible to transfer the learning model transfer information 30 from the first learning device 10 to the second learning device 20 in an encrypted state. As a result, it is possible to suppress leakage of the learning model transfer information 30, and thus, even in a case where the learning model transfer information 30 includes information with high confidentiality, it is possible to transfer the learning model transfer information 30 from the first learning device 10 to the second learning device 20.

[0139] Further, in the present embodiment, as described above, with respect to the first algorithm determination information 30a, in a case where the first algorithm 40 is subjected to version upgrade, the first algorithm determination information 30a of a version with compatibility is included in the first algorithm determination information 30a after the version upgrade, with respect to the second algorithm determination information 70a, in a case where the second algorithm 50 is subjected to version upgrade, the second algorithm determination information 70a of a version with compatibility is included in the second algorithm determination information 70a after the version upgrade, and the algorithm matching determination section 21c is configured to determine that the first algorithm 40 and the second algorithm 50 are matched even in a case where the version of the first algorithm 40 and the version of the second algorithm 50 are different, when the version of the first algorithm 40 and the version of the second algorithm 50 have compatibility with each other. Thereby, even in a case where the version of the first algorithm 40 and the version of the second algorithm 50 are different, it is possible to transfer the learning model from the first learning device 10 to the second learning device 20 in a case where the version of the first algorithm 40 and the version of the second algorithm 50 have compatibility. As a result, it is possible to improve the degree of freedom of the versions of the first algorithm 40 and the second algorithm 50.

[0140] Further, in the present embodiment, as described above, the first algorithm determination information 30a includes the first identification information 31 corresponding to the first algorithm 40, the second algorithm determination information 70a includes the second identification information 71 corresponding to the second algorithm 50, and the algorithm matching determination section 21c is configured to determine the matching of the first algorithm 40 and the second algorithm 50 by determining whether the first identification information 31 and the second identification information 71 are consistent. Thereby, since it is possible to determine the matching of the first algorithm 40 and the second algorithm 50 depending on whether the first identification information 31 and the second identification information 71 are consistent, by acquiring the first identification information 31 and the second identification information 71, it is possible to easily determine the matching of the first algorithm 40 and the second algorithm 50.

[0141] In addition, in the present embodiment, as described above, the first identification information 31 includes the first hash value 32 generated based on the first algorithm 40, the second identification information 71 includes the second hash value 72 generated based on the second algorithm 50, and the algorithm matching determination unit 21c is configured to determine whether the first hash value 32 and the second hash value 72 are identical. Here, the hash value is a unique value obtained based on the content of the algorithm. Thus, for example, even if the file name of the algorithm is changed, the hash value is not changed in the case where the content of the algorithm is not changed. Thus, by being configured as described above, for example, unlike a structure in which the matching of the first algorithm 40 and the second algorithm 50 is determined based on the file name of the first algorithm 40 and the file name of the second algorithm 50, even if the file names are different from each other, the matching of the first algorithm 40 and the second algorithm 50 can be determined to be identical in the case where the first hash value 32 and the second hash value 72 are identical. As a result, the determination accuracy of the matching of the first algorithm 40 and the second algorithm 50 can be improved.

[0142] In addition, in the present embodiment, as described above, the following steps are included: the first computer (the first learning device 10) generates and stores the learning model transfer information 30 before the second computer (the second learning device 20) receives input of the learning model transfer information 30; and the first computer (the first learning device 10) outputs the stored learning model transfer information 30 before the second computer (the second learning device 20) receives input of the learning model transfer information 30. Thus, by outputting the learning model transfer information 30 to the portable storage medium 60 or the like, the learning model transfer information 30 can be easily transferred from the first computer (the first learning device 10) to the second computer (the second learning device 20).

[0143] [Modified Example]

[0144] Furthermore, it should be understood that the embodiments disclosed herein are illustrative in all respects, not limiting. The scope of the application should be construed based on the claims, and all modifications (modified examples) within the meaning and range equivalent to the claims are included.

[0145] For example, in the above-described embodiments, an example in which the first learning device 10 and the second learning device 20 are constructed by a client-server model is shown, but the present application is not limited thereto. In the present application, for example, as shown in Figure 15 the first learning device 10 and the second learning device 20 can be configured by independent computers. In this case, for example, the first learning device 10 and the second learning device 20 can be configured by a plurality of computers connected to each other via a network. Figure 15In this example, the first learning device 10 comprises a first computer 300 having a first processor 310 and a first storage unit 320. The first computer 300 is connected to a display unit 330 and an input unit 340. The first computer 300 is connected to a camera device 150 in a communicative manner. The first processor 310 of the first computer 300 includes the embodiments described above (see [reference]). Figure 6 The transfer information encryption unit 11b and transfer information output unit 11c shown in the figure are used as functional blocks.

[0146] In addition, Figure 15 In this example, the second learning device 20 is comprised of a second computer 400 equipped with a second processor 410 and a second storage unit 420. The second computer 400 is connected to a display unit 430 and an input unit 440. The second computer 400 is connected to the camera device 151 in a communicative manner. The second processor 410 of the second computer 400 includes the embodiments described above (see [reference]). Figure 7 and Figure 8 The transfer information input receiving unit 21a, hash value acquisition unit 21b, algorithm matching determination unit 21c, preprocessor matching determination unit 21d, notification content acquisition unit 21e, learning model parameter setting unit 21f, and preprocessing parameter setting unit 21g shown in the figure are used as function blocks.

[0147] In addition, in the above-described implementation methods and Figure 15 In the illustrated variation, an example is shown where a single first processor 11 (310) is used to execute the processes as the transfer information generation unit 11a, the transfer information encryption unit 11b, and the transfer information output unit 11c, but the present invention is not limited thereto. The processes for deriving the learning model transfer information 30 can also be executed by multiple processors in a distributed manner. Alternatively, each process can be executed by a separate processor. Multiple processors can be configured on separate computers. That is, the first learning device 10 can also be composed of multiple computers.

[0148] In addition, in the above-described implementation methods and Figure 15 In the illustrated variation, an example is shown where a single second processor 21 (410) is used to execute each of the following processes: the transfer information input receiving unit 21a, the hash value acquisition unit 21b, the algorithm matching determination unit 21c, the preprocessor matching determination unit 21d, the notification content acquisition unit 21e, the learning model parameter setting unit 21f, and the preprocessing parameter setting unit 21g. However, the present invention is not limited to this. Each process of importing the learning model transfer information 30 can also be executed by multiple processors in a distributed manner. Alternatively, each process can be executed by a separate processor. Multiple processors can be configured on separate computers. That is, the second learning device 20 can also be composed of multiple computers.

[0149] Also, in the above-described embodiments, an example of a structure in which the first learning device 10 performs the derivation processing of the learning model transfer information 30, and the second learning device 20 performs the import processing of the learning model transfer information 30 is shown, but the present application is not limited thereto. For example, it can also be configured so that both the first learning device 10 and the second learning device 20 perform the derivation processing and the import processing of the learning model transfer information 30. That is, it can also be that the first learning device 10 (the first processor 11) and the second learning device 20 (the second processor 21) each have the transfer information generation section 11a, the transfer information encryption section 11b, the transfer information output section 11c, the transfer information input reception section 21a, the hash value acquisition section 21b, the algorithm matching determination section 21c, the preprocessing program matching determination section 21d, the notification content acquisition section 21e, the learning model parameter setting section 21f, and the preprocessing parameter setting section 21g.

[0150] Also, in the above-described embodiments, an example of a structure in which the algorithm matching determination section 21c determines the matching of the first algorithm 40 and the second algorithm 50 based on whether the first hash value 32 and the second hash value 72 are identical is shown, but the present application is not limited thereto. For example, the algorithm matching determination section 21c can also be configured to determine the matching of the first algorithm 40 and the second algorithm 50 based on the file name of the first algorithm 40 and the file name of the second algorithm 50.

[0151] Also, in the above-described embodiments, an example of a structure in which the second learning device 20 has the preprocessing program matching determination section 21d is shown, but the present application is not limited thereto. For example, it can also be that the second learning device 20 does not have the preprocessing program matching determination section 21d. However, in the case where the second learning device 20 does not have the preprocessing program matching determination section 21d, the user needs to confirm the matching of the first preprocessing program 41 and the second preprocessing program 51, and thus the burden on the user increases. Thus, it is preferable that the second learning device 20 have the preprocessing program matching determination section 21d.

[0152] Also, in the above-described embodiments, an example of a structure in which the second learning device 20 has the preprocessing parameter setting section 21g is shown, but the present application is not limited thereto. For example, it can also be that the second learning device 20 does not have the preprocessing parameter setting section 21g. However, in the case where the second learning device 20 does not have the preprocessing parameter setting section 21g, the user needs to set the second parameter 30d, and thus the burden on the user increases. Thus, it is preferable that the second learning device 20 have the preprocessing parameter setting section 21g.

[0153] Also, in the above-described embodiments, an example in which the learning model transfer information 30 includes the additional information 30e is shown, but the present application is not limited thereto. For example, it can also be that the learning model transfer information 30 does not include the additional information 30e.

[0154] In addition, in the above-described embodiment, an example in which the first learning device 10 has the structure of including the transfer information encryption unit 11b is shown, but the present application is not limited to this. For example, it can be that the first learning device 10 does not include the transfer information encryption unit 11b. However, in the case where the first learning device 10 does not include the transfer information encryption unit 11b, the learning model transfer information 30 can leak. Thus, it is preferable that the first learning device 10 include the transfer information encryption unit 11b.

[0155] In addition, in the above-described embodiment, an example in which the algorithm matching determination unit 21c determines that the matching of the first algorithm 40 and the second algorithm 50 is obtained even in the case where the version of the first algorithm 40 and the version of the second algorithm 50 are different, when the versions of the first algorithm 40 and the second algorithm 50 have compatibility with each other is shown, but the present application is not limited to this. For example, it can be that the algorithm matching determination unit 21c is configured to determine that the matching of the first algorithm 40 and the second algorithm 50 is not obtained in the case where the version of the first algorithm 40 and the version of the second algorithm 50 are different. However, in the case where the algorithm matching determination unit 21c is configured to determine that the matching of the first algorithm 40 and the second algorithm 50 is not obtained when the versions of the first algorithm 40 and the second algorithm 50 are different, the first algorithm 40 and the second algorithm 50 whose versions are different but have compatibility with each other cannot be transferred. Thus, it is preferable that the algorithm matching determination unit 21c be configured to determine that the matching of the first algorithm 40 and the second algorithm 50 is obtained even in the case where the version of the first algorithm 40 and the version of the second algorithm 50 are different, when the versions of the first algorithm 40 and the second algorithm 50 have compatibility with each other.

[0156] In addition, in the above-described embodiment, an example in which the transfer system 200 of the learning model for cell image analysis is configured to transfer the learning model for analyzing the cell image 90 from the first learning device 10 to the second learning device 20 is shown, but the present application is not limited to this. For example, the present application can also be applied to a case where a learning model that has learned to analyze measurement data acquired by a liquid chromatograph or the like is transferred from the first learning device 10 to the second learning device 20.

[0157] [Mode]

[0158] The above-described exemplary embodiments are concrete examples of the following modes, which can be understood by those skilled in the art.

[0159] (Project 1)

[0160] A transfer system of a learning model for cell image analysis that transfers a learning model used in the analysis of a cell image from a first learning device to a second learning device, in the transfer system of the learning model for cell image analysis,

[0161] The second learning device includes:

[0162] a transfer information input receiving unit that receives input of learning model transfer information, the learning model transfer information including first algorithm determination information for determining a first algorithm of a learning model used in analysis of the cell image and first parameters that are learning parameters output when the learning model is generated and are used when the cell image is analyzed by the learning model;

[0163] a second learning device storage unit that stores a second algorithm of the learning model used in analysis of the cell image;

[0164] an algorithm matching determination unit that determines whether or not matching of an estimated result of the first algorithm and the second algorithm when the first parameters are used is obtained, based on the first algorithm determination information and second algorithm determination information for determining the second algorithm stored in the second learning device storage unit;

[0165] a notification unit that notifies whether or not the matching of the first algorithm and the second algorithm is obtained; and

[0166] a learning model parameter setting unit that sets the first parameters used with the second algorithm, in a case where the matching of the first algorithm and the second algorithm is obtained.

[0167] (Item 2)

[0168] The transfer system of the learning model for cell image analysis according to Item 1, in which

[0169] The first learning device includes:

[0170] a first learning device storage unit that stores the learning model transfer information; and

[0171] a transfer information output unit that outputs the learning model transfer information stored in the first learning device storage unit.

[0172] (Item 3)

[0173] The transfer system of the learning model for cell image analysis according to Item 2, in which

[0174] the learning model transfer information further includes preprocessing program determination information for determining a first preprocessing program that is a program of preprocessing of the cell image when the learning model is learned,

[0175] The second learning device storage section is configured to further store a second preprocessing program, the second preprocessing program being a program for preprocessing of the cell image,

[0176] The second learning device further includes a preprocessing program matching determination section that determines whether matching of results of preprocessing of the cell image by the first preprocessing program and the second preprocessing program is achieved based on the preprocessing program determination information and the second preprocessing program stored in the second learning device storage section.

[0177] (Item 4)

[0178] The transfer system of the learning model for cell image analysis according to any one of items 3 to 5, in which

[0179] The learning model transfer information further includes a second parameter corresponding to the first preprocessing program,

[0180] The transfer system of the learning model for cell image analysis further includes a preprocessing parameter setting section that sets the second parameter with respect to the second preprocessing program in a case where matching of the results of preprocessing by the first preprocessing program and the second preprocessing program is achieved.

[0181] (Item 5)

[0182] The transfer system of the learning model for cell image analysis according to any one of items 2 to 4, in which

[0183] The learning model transfer information further includes at least any one of the following: a name of a data set used in learning of the learning model; a number of images of the cell image used in learning of the learning model; proportions of training data, validation data, and test data within data used in learning of the learning model; a size of the cell image used in learning of the learning model; and an index value at the time of learning of the learning model.

[0184] (Item 6)

[0185] The transfer system of the learning model for cell image analysis according to any one of items 2 to 4, in which

[0186] The first learning device further includes a transfer information encryption section that encrypts the learning model transfer information stored in the first learning device storage section,

[0187] The transfer information output section is configured to output the learning model transfer information encrypted by the transfer information encryption section,

[0188] The transfer information input receiving section is configured to decrypt the learning model transfer information encrypted by the transfer information encryption section.

[0189] (Item 7)

[0190] The learning model transfer system for cell image analysis according to any one of Items 1 to 5, wherein

[0191] Regarding the first algorithm determination information, in a case where the first algorithm is subjected to version upgrade, the first algorithm determination information of a version having compatibility is included in the first algorithm determination information after the version upgrade,

[0192] Regarding the second algorithm determination information, in a case where the second algorithm is subjected to version upgrade, the second algorithm determination information of a version having compatibility is included in the second algorithm determination information after the version upgrade,

[0193] The algorithm matching determination section is configured to determine that the matching of the first algorithm and the second algorithm is obtained even in a case where the version of the first algorithm and the version of the second algorithm are different, when the version of the first algorithm and the version of the second algorithm have compatibility with each other.

[0194] (Item 8)

[0195] The learning model transfer system for cell image analysis according to any one of Items 1 to 6, wherein

[0196] The first algorithm determination information includes first identification information corresponding to the first algorithm,

[0197] The second algorithm determination information includes second identification information corresponding to the second algorithm,

[0198] The algorithm matching determination section is configured to determine the matching of the first algorithm and the second algorithm by determining whether the first identification information and the second identification information are identical.

[0199] (Item 9)

[0200] The learning model transfer system for cell image analysis according to Item 8, wherein

[0201] The first identification information includes a first hash value generated based on the first algorithm,

[0202] The second identification information includes a second hash value generated based on the second algorithm,

[0203] The algorithm matching determination unit determines whether the first hash value and the second hash value are identical.

[0204] (Item 10)

[0205] A transfer method of a learning model for cell image analysis for transferring a learning model used in analysis of a cell image from a first computer to a second computer, the transfer method of the learning model for cell image analysis including the steps of:

[0206] The second computer receives input of learning model transfer information including first algorithm determination information for determining a first algorithm of the learning model used in the analysis of the cell image and first parameters output when the learning model is generated and used in the analysis of the cell image by the learning model;

[0207] The second computer determines whether matching of an estimated result of the first algorithm and the second algorithm using the first parameters is obtained based on the second algorithm stored in the storage and the first algorithm determination information included in the learning model transfer information;

[0208] The second computer notifies whether the matching of the first algorithm and the second algorithm is obtained; and

[0209] In a case where the matching of the first algorithm and the second algorithm is obtained, the second computer sets the first parameters used with the second algorithm.

[0210] (Item 11)

[0211] The transfer method of the learning model for cell image analysis according to Item 10, wherein the steps include:

[0212] The first computer generates and stores the learning model transfer information before the second computer receives input of the learning model transfer information; and

[0213] The first computer outputs the stored learning model transfer information before the second computer receives input of the learning model transfer information.

Claims

1. A transfer system of a learning model for cell image analysis that transfers a learning model used in analysis of a cell image from a first learning device to a second learning device, in the transfer system of the learning model for cell image analysis, the second learning device includes: a transfer information input reception section that receives input of learning model transfer information including first algorithm determination information for determining a first algorithm of the learning model used in analysis of the cell image and a first parameter that is a learning parameter output when the learning model is generated and is used in analysis of the cell image by the learning model; a second learning device storage section that stores a second algorithm of the learning model used in analysis of the cell image; an algorithm matching determination section that determines whether matching of an estimated result of the first algorithm and the second algorithm using the first parameter is obtained by determining whether the first algorithm determination information coincides with second algorithm determination information for determining the second algorithm stored in the second learning device storage section; a notification section that notifies whether the matching of the first algorithm and the second algorithm is obtained; and a learning model parameter setting section that sets the first parameter used with the second algorithm in a case where the matching of the first algorithm and the second algorithm is obtained.

2. The transfer system of the learning model for cell image analysis according to claim 1, wherein the first learning device includes: a first learning device storage section that stores the learning model transfer information; and a transfer information output section that outputs the learning model transfer information stored in the first learning device storage section.

3. The transfer system of the learning model for cell image analysis according to claim 2, wherein the learning model transfer information further includes preprocessing program determination information for determining a first preprocessing program that is a program of preprocessing of the cell image when the learning model is learned, the second learning device storage section is configured to further store a second preprocessing program that is a program of preprocessing of the cell image, the second learning device further includes a preprocessing program matching determination section that determines whether matching of a result of preprocessing of the cell image by the first preprocessing program and the second preprocessing program is obtained by determining whether the preprocessing program determination information coincides with second preprocessing program determination information for determining the second preprocessing program stored in the second learning device storage section.

4. The transfer system of the learning model for cell image analysis according to claim 3, wherein the learning model transfer information further includes a second parameter corresponding to the first preprocessing program, ​ The cell image analysis learning model transfer system further includes a pre-processing parameter setting unit configured to set the second parameter for the second pre-processing program when the pre-processed results of the first pre-processing program and the second pre-processing program match. 5.The cell image analysis learning model transfer system according to claim 2, wherein The learning model transfer information further includes at least one of: a name of a data set used in learning of the learning model; a number of images of the cell images used in learning of the learning model; a ratio of training data, validation data, and test data within the data used in learning of the learning model; a size of the cell images used in learning of the learning model; and an index value at the time of learning of the learning model. 6.The cell image analysis learning model transfer system according to claim 2, wherein The first learning device further includes a transfer information encryption unit configured to encrypt the learning model transfer information stored in the first learning device storage unit, The transfer information output unit is configured to output the learning model transfer information encrypted by the transfer information encryption unit, The transfer information input reception unit is configured to decrypt the learning model transfer information encrypted by the transfer information encryption unit. 7.The cell image analysis learning model transfer system according to claim 1, wherein In the case where the first algorithm is version-upgraded, the first algorithm determination information after the version-upgrade includes the first algorithm determination information of a version having compatibility, In the case where the second algorithm is version-upgraded, the second algorithm determination information after the version-upgrade includes the second algorithm determination information of a version having compatibility, The algorithm compatibility determination unit is configured to determine that the first algorithm and the second algorithm match each other even in the case where the versions of the first algorithm and the second algorithm are different from each other, when the versions of the first algorithm and the second algorithm have compatibility with each other. 8.The cell image analysis learning model transfer system according to claim 1, wherein The first algorithm determination information includes first identification information corresponding to the first algorithm, The second algorithm determination information includes second identification information corresponding to the second algorithm, The algorithm compatibility determination unit is configured to determine the compatibility of the first algorithm and the second algorithm by determining whether the first identification information and the second identification information match each other. 9.The cell image analysis learning model transfer system according to claim 8, wherein The first identification information includes a first hash value generated based on the first algorithm, The second identification information includes a second hash value generated based on the second algorithm, The algorithm matching determination unit determines whether the first hash value and the second hash value are identical.

10. A transfer method of a learning model for cell image analysis, for transferring a learning model used in analysis of a cell image from a first computer to a second computer, the transfer method of the learning model for cell image analysis comprising the steps of: the second computer receiving input of learning model transfer information, the learning model transfer information including first algorithm determination information for determining a first algorithm of the learning model used in the analysis of the cell image and first parameters output when the learning model is generated and used in the analysis of the cell image by the learning model; the second computer determining whether the first algorithm and a second algorithm of the second computer match in estimated results when the first parameters are used, by determining whether second algorithm determination information for determining the second algorithm stored in a storage unit matches the first algorithm determination information included in the learning model transfer information; the second computer notifying whether the first algorithm and the second algorithm match; and in a case where the first algorithm and the second algorithm match, the second computer setting the first parameters used with the second algorithm.

11. The transfer method of a learning model for cell image analysis according to claim 10, characterized by, comprising the steps of: the first computer generating and storing the learning model transfer information before the second computer receives input of the learning model transfer information; and the first computer outputting the stored learning model transfer information before the second computer receives input of the learning model transfer information. the first computer outputting the stored learning model transfer information before the second computer receives input of the learning model transfer information.

Images