Teacher data management device, teacher data management method, and program for teacher data management
The teacher data management device generates pseudo medical examination data to replace deleted data, maintaining the quality and safety of training data, thus preserving the accuracy of machine learning models.
Patent Information
- Application Number
- JP2023071811
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-04-25
- Publication Date
- 2024-11-07
AI Technical Summary
The deletion of personal information from training data used in machine learning models can impair the stability and accuracy of the learning models, as the loss of quality in the training data can lead to decreased performance.
A teacher data management device that generates pseudo medical examination data similar to the deleted data, using a pseudo-examination data generation model, and associates this data with diagnostic data to maintain the quality and safety of the training data.
The solution allows for the deletion of specific data while maintaining the stability and quality of the training data, ensuring the accuracy and safety of the learning models.
Smart Images

Figure 2024157440000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a teacher data management device, a teacher data management method, and a teacher data management program. [Background technology]
[0002] In recent years, machine learning technology using various medical data obtained in the medical field has been developed. However, such medical data may contain personal information, so there are studies on how to safely handle medical data used in machine learning from the viewpoint of protecting personal information.
[0003] For example, a providing device is known that stores patient information and medical information from a patient in a distributed manner and provides consent information of the patient when providing the information to a third party (Patent Document 1). [Prior art documents] [Patent documents]
[0004] [Patent Document 1] JP 2022-143911 A Summary of the Invention [Problem to be solved by the invention]
[0005] From the perspective of personal information protection, even if personal information has been provided for various uses with the consent of an individual, the individual may express their intention to revoke the provision (opt-out). In this case, it is necessary not to use the personal information after expressing their intention to revoke the provision.
[0006] However, in cases where data including personal information was used as training data for a learning model, etc., deleting the revoked data could result in the stability of the quality of the training data being compromised. There was a concern that the loss of stability in the quality of the training data could result in a decline in the accuracy of the learning model that used it.
[0007] The present invention aims to provide a teacher data management device, a teacher data management method, and a teacher data management program that enable specific data to be deleted while maintaining the safety and quality stability of the teacher data. [Means for solving the problem]
[0008] The teacher data management device of the present invention is equipped with a processor, which stores teacher data in a data storage unit, associating medical examination data obtained from a medical examination undertaken by an individual with diagnostic data generated based on the medical examination data, deletes specific medical examination data from the data storage unit, identifies diagnostic data associated with the deleted medical examination data, generates pseudo medical examination data that is pseudo to the deleted medical examination data based on the identified diagnostic data, and stores the pseudo medical examination data in the data storage unit in association with the identified diagnostic data.
[0009] The specific medical data is preferably medical data including information for which consent to providing information has been revoked by an individual.
[0010] The pseudo medical examination data is preferably medical examination data having similar characteristics to those of the deleted medical examination data.
[0011] The pseudo medical data is preferably medical data that does not completely match the deleted medical data.
[0012] It is preferable that the data storage unit stores teacher data that associates the personal ID, the medical examination data, and the diagnostic data by linking the personal ID of the individual who underwent the medical examination with the medical examination data using a first pseudonymous ID, and linking the first pseudonymous ID with the diagnostic data using a second pseudonymous ID.
[0013] It is preferable that the processor generates a pseudonymous reverse lookup table consisting of a personal ID and a first pseudonymous ID, a pseudonymous medical examination table consisting of the first pseudonymous ID and medical examination data, a medical examination diagnosis reverse lookup table consisting of the first pseudonymous ID and a second pseudonymous ID, and a pseudonymous diagnosis table consisting of the second pseudonymous ID and diagnostic data, and that the data storage unit stores the pseudonymous reverse lookup table, the pseudonymous medical examination table, the medical examination diagnosis reverse lookup table, and the pseudonymous diagnosis table in the data storage unit.
[0014] When the processor deletes medical examination data of a specific individual from the data storage unit, it is preferable to delete the personal ID of the specific individual in the pseudonym reverse lookup table and the specific first pseudonym ID linked thereto, the specific first pseudonym ID in the pseudonym medical examination table and the specific medical examination data linked thereto, and the specific first pseudonym ID in the medical examination diagnosis reverse lookup table and the specific second pseudonym ID linked thereto.
[0015] It is preferable that the processor generates pseudo medical data using a pseudo medical data generation model, and the pseudo medical data generation model is a trained model that has been trained to generate pseudo medical data by inputting deleted medical data and diagnostic data.
[0016] It is preferable that the processor calculates a reconstruction error indicating the degree of match between the deleted screening data and the pseudo screening data, and the pseudo screening data creation model generates the pseudo screening data so that the absolute value of the reconstruction error becomes a positive number.
[0017] It is preferable that the pseudo-examination data creation model is a model that connects a generation model that generates pseudo-examination data by inputting deleted examination data and diagnostic data, and an evaluation model that evaluates the generated pseudo-examination data.
[0018] The teacher data management method of the present invention includes the steps of storing teacher data in a data storage unit, which associates medical examination data with diagnostic data generated based on the medical examination data, deleting specific medical examination data from the data storage unit, identifying diagnostic data associated with the deleted medical examination data, generating pseudo medical examination data that is similar to the deleted medical examination data based on the identified diagnostic data, and storing the pseudo medical examination data in the data storage unit in association with the identified diagnostic data.
[0019] The teacher data management program of the present invention causes a computer to perform the following functions: storing in a data storage unit teacher data that associates medical examination data with diagnostic data generated based on the medical examination data; deleting specific medical examination data from the data storage unit; identifying diagnostic data that was associated with the deleted medical examination data; generating pseudo medical examination data that is similar to the deleted medical examination data based on the identified diagnostic data; and storing the pseudo medical examination data in the data storage unit in association with the identified diagnostic data. Effect of the Invention
[0020] According to the present invention, it is possible to delete specific data while maintaining the safety and quality stability of the teaching data. [Brief description of the drawings]
[0021] [Figure 1] A block diagram explaining the functions of the teacher data management device. [Diagram 2] 1 is an explanatory diagram illustrating the configuration of a medical information management system including a teacher data management device. [Diagram 3] 3 is an explanatory diagram illustrating various tables included in a data storage unit. FIG. [Figure 4] FIG. 13 is an explanatory diagram illustrating the contents of a kana reverse lookup table. [Diagram 5] FIG. 13 is an explanatory diagram for explaining the contents of a pseudonym medical examination table. [Figure 6] FIG. 13 is an explanatory diagram for explaining the contents of a medical examination diagnosis reverse look-up table. [Figure 7] FIG. 13 is an explanatory diagram illustrating the contents of a pseudonym diagnosis table. [Figure 8] FIG. 11 is an explanatory diagram for explaining deletion of medical examination data. [Figure 9] FIG. 4 is a block diagram illustrating the function of a pseudo medical examination data generating unit. [Figure 10] FIG. 2 is an explanatory diagram illustrating a function of a pseudo medical examination data generating unit. [Figure 11] FIG. 11 is an explanatory diagram illustrating a function of a reconstruction accuracy suppression unit. [Figure 12] 13 is a flowchart illustrating the processing flow in the teacher data management device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0022] An example of an embodiment of a teacher data management device and the like of the present invention will be described. First, the background to the creation of one of the following embodiments will be described. When generating a medical learning model using machine learning technology, the medical learning model is generated using individual medical examination data and diagnostic data based on the medical examination data as teacher data.
[0023] From the perspective of protecting personal information, even if personal information has already been provided, the provider can withdraw the provision at their will (opt-out). In this case, data containing the relevant personal information must be deleted from the accumulated training data, which may result in a decrease in the accuracy of the learning model that uses this training data for training.
[0024] Therefore, the present invention provides a teacher data management device, etc. that can delete specific data while maintaining the stability of the quality of the teacher data without reducing the quality of the teacher data, which would result in a decrease in the accuracy of the learning model, etc.
[0025] According to the teaching data management device of the present invention, pseudo medical examination data is generated from diagnostic information linked to medical examination data including personal information to be deleted, and this is used as dummy data in place of the medical examination data, which is the deleted personal information. By learning using the dummy data in place of the deleted personal information, it is possible to minimize the deterioration of the learning model quality.
[0026] In this specification, the terms teacher data, medical examination data, and diagnostic data include cases where each data is one, multiple, and a data set. In this embodiment, medical data, for example, medical examination data, diagnostic data, etc., are used as teacher data, but the present invention can use data other than medical data. For example, the present invention can also be applied to teacher data including financial data that handles personal information of customers. In addition, in this specification, the term "learning model" may include a model that has already been trained.
[0027] The teacher data management device of the present invention is a device that manages teacher data used in machine learning. As shown in Fig. 1, the teacher data management device 10 includes a data storage unit 11, a medical examination data deletion unit 12, a diagnostic data specification unit 13, a pseudo medical examination data generation unit 14, and a pseudo medical examination data replacement unit 15.
[0028] The data storage unit 11 stores teacher data that associates the medical examination data with the diagnostic data generated based on the medical examination data. It is preferable to store the medical examination data and the diagnostic data independently. This allows each of them to be edited, such as deleted. In this case, it is preferable to associate the two by providing another data that links the medical examination data with the diagnostic data.
[0029] The medical examination data is data including the results of various tests obtained by an individual undergoing a medical examination, etc., and includes what is generally referred to as medical examination data. The diagnostic data is data including the results of a diagnosis made by a doctor based on the medical examination data, etc. The teacher data is a dataset with a correct answer label used for learning by a learning model that is a machine learning algorithm. For the medical examination data, the correct answer label is the diagnostic data diagnosed by a doctor based on the medical examination data. Therefore, the teacher data is a dataset that associates the medical examination data obtained by the medical examination with the diagnostic data generated based on the medical examination data.
[0030] The teacher data may be any data that can be used for learning the learning model. Preferably, the teacher data is capable of performing learning such that, when medical examination data for which diagnostic data is unknown is input to a trained model in which the learning model is trained using the teacher data, the trained model outputs diagnostic data that is a correct answer corresponding to the medical examination data.
[0031] The medical examination data deletion unit 12 deletes specific medical examination data from the data storage unit 11. The user designates medical examination data to be deleted from the data storage unit 11, and this is set as specific medical examination data. There may be cases where the medical examination data stored in the data storage unit 11 is deleted due to reasons such as the withdrawal of consent to information provision for the medical examination data from an individual who underwent the medical examination that is the source of the medical examination data. For example, when using teacher data including the medical examination data stored in the data storage unit 11 after the consent is withdrawn, it is necessary to delete the medical examination data of the subject whose consent to information provision has been withdrawn from the data storage unit 11 before using the teacher data.
[0032] The diagnostic data specifying unit 13 specifies diagnostic data associated with the deleted medical examination data. Since the medical examination data and the diagnostic data are associated with each other, the diagnostic data associated with the deleted medical examination data can be specified from the deleted medical examination data. The specified diagnostic data is used by the pseudo medical examination data generating unit 14.
[0033] The pseudo medical data generating unit 14 generates pseudo medical data that is similar to the deleted medical data based on the specified diagnostic data. The pseudo medical data that is similar to the deleted medical data means data that is different from the deleted medical data but has similar properties and characteristics to the deleted medical data. The properties and characteristics are properties and characteristics as teacher data. Therefore, when the properties and characteristics are similar to the deleted medical data, when the pseudo medical data is used as teacher data, there is less difference in the trained model obtained compared to when the deleted medical data is used as teacher data. In other words, the pseudo medical data that is similar to the deleted medical data is preferably pseudo medical data that is similar to the trained model so that when a learning model is trained using both the teacher data including the deleted medical data and the teacher data including the pseudo medical data, it is indistinguishable from the results output by the trained model as to which teacher data was used.
[0034] As described above, the pseudo medical examination data is preferably medical examination data having similar characteristics to those of the deleted medical examination data. In this case, the characteristics specifically refer to data distribution, tendency, statistical characteristics, etc., and it is preferable that the deleted medical examination data and the pseudo medical examination data have similar data distribution, tendency, statistical characteristics, etc.
[0035] Moreover, it is preferable that the pseudo medical examination data is data that does not completely match the deleted medical examination data. If the pseudo medical examination data is data that completely matches the deleted medical examination data, it may be difficult to distinguish it from the medical examination data to be deleted, and the medical examination data may not be deleted. There may be cases where an individual can be identified by the unique characteristics of the medical examination data itself. By using the pseudo medical examination data that does not completely match the deleted medical examination data, it is possible to clearly indicate that the medical examination data desired to be deleted has been deleted.
[0036] In addition, the data that does not completely match refers to data in which all values, etc. match between the deleted medical data and the pseudo medical data when the specific medical data includes multiple values, etc. Therefore, the pseudo medical data may be partially identical to the deleted medical data.
[0037] As described above, since the deleted medical examination data and the pseudo medical examination data have similar characteristics, in a trained model in which the training data is trained using the training data before the specific medical examination data is deleted and the training data containing the pseudo medical examination data instead of the specific medical examination data, it is impossible to distinguish which training data is used from the results output by the trained model. Therefore, even if the training data containing the pseudo medical examination data is used for training, it is possible to maintain the quality of the trained model, and such training data is kept to a minimum in terms of quality degradation, so that the stability of quality is maintained.
[0038] A pseudo medical examination data generation model, which is a generative model, can be used to generate the pseudo medical examination data. The pseudo medical examination data generation model is a trained generation model that has been trained to output pseudo medical examination data by inputting deleted medical examination data and diagnostic data generated based on the deleted medical examination data. Preferably, the pseudo medical examination data generation model is trained to output pseudo medical examination data that does not completely match the deleted medical examination data but is pseudo to the deleted medical examination data. Therefore, the pseudo medical examination data generation model has a function of restoring medical examination data similar to the input medical examination data.
[0039] The pseudo-examination data generation model may adopt any learning model as long as it can output pseudo-examination data to the examination data, and may use either a learning model for supervised learning or a learning model for unsupervised learning. The pseudo-examination data generation model is preferably a neural network (NN) model because it generates superior pseudo-examination data. More preferably, a deep neural network (DNN), which is a deep network structure having many hidden layers, is preferable.
[0040] In addition, as a learning model configuration for the pseudo medical data generation model, since pseudo data can be preferably created by providing diagnostic data, which is the correct label, to the input medical data, generative models such as Conditional Generative Adversarial Networks (Conditional GAN), which are classified as supervised learning models having a generator and a discriminator, and Conditional Variational Autoencoders (Conditional VAE), which are classified as supervised learning models, may be used.
[0041] The pseudo medical examination data replacement unit 15 associates the pseudo medical examination data with the specific diagnostic data associated with the deleted medical examination data, and stores it in the data storage unit 11. That is, the pseudo medical examination data replacement unit 15 associates the pseudo medical examination data generated by the pseudo medical examination data generation unit 14 with the specific diagnostic data as a substitute for the deleted medical examination data, and stores it in the data storage unit 11. Therefore, after deleting the specific medical examination data, the pseudo medical examination data and the diagnostic data associated therewith form teacher data so as to replace the deleted medical examination data and the corresponding diagnostic data, and the teacher data is stored in the data storage unit 11.
[0042] As described above, the teacher data management device 10 can delete any specific data from the teacher data stored in the data storage unit 11, so that the teacher data does not contain inappropriate data and the safety of the teacher data can be maintained. Furthermore, even if specific data is deleted, pseudo-examination data with similar properties, characteristics, etc. can be generated and used in place of the deleted data, so the stability of the quality of the teacher data is maintained. Therefore, the teacher data management device 10 enables the deletion of specific data while maintaining the safety and stability of the quality of the teacher data.
[0043] Next, an example of an embodiment of the teacher data management device 10 etc. will be described with reference to the drawings. As shown in Fig. 2, the medical information management system 20 is a computer system that manages medical information in a medical facility such as a hospital. The medical information management system 20 includes the teacher data management device 10, a server group 21, and a client terminal 22, which are connected to each other via a network 23 so as to be able to communicate with each other.
[0044] From the viewpoint of personal information management, it is preferable that the teacher data management device 10 is installed within a medical facility as shown in Fig. 2. In this case, the network 23 is a LAN (Local Area Network) or the like installed within the medical facility. Note that the teacher data management device 10 may be installed outside the medical facility as long as it is possible to appropriately manage personal information.
[0045] The teacher data management device 10 is configured by installing an operating system program and application programs such as client programs on a computer equipped with a processor. The teacher data management device 10 stores a teacher data management program in a storage (not shown) in addition to the operating system.
[0046] The teacher data management program is an application program for causing a computer constituting the teacher data management device 10 to execute the functions of the teacher data management device 10. When the teacher data management device 10 is started, a processor (not shown) included in the teacher data management device 10 cooperates with a memory (not shown) to execute the teacher data management program, and functions as a data storage unit 11, a medical examination data deletion unit, a diagnostic data identification unit 13, a pseudo medical examination data generation unit 14, a pseudo medical examination data replacement unit 15, etc.
[0047] The server group 21 provides examination data and the like to the teacher data management device 10. The server group 21 includes an electronic medical record server 24, an image server 25, a report server 26, and the like.
[0048] The electronic medical record server 24 has a medical record database 24A that stores electronic medical records. An electronic medical record is a collection of one or more pieces of medical data. Specifically, the electronic medical record includes medical examination data and diagnostic data generated based on the medical examination data. In addition, the electronic medical record includes medical examination records, specimen test results, vital signs of patients, orders for tests, etc., treatment records, accounting data, and other medical data.
[0049] The image server 25 is a so-called PACS (Picture Archiving and Communication System) server, and has an image database 25A in which medical images are stored. Medical images are images obtained by various imaging examinations such as CT examinations, MRI examinations, X-ray examinations, ultrasound examinations, and endoscopic examinations. These medical images are recorded in a format that complies with the DICOM (Digital Imaging and Communications in Medicine) standard, for example.
[0050] The report server 26 has a report database 26A that stores medical reports such as image interpretation reports. Medical reports include image interpretation reports and detailed examination reports, and are reports that summarize images, numerical values, and findings obtained in tests such as imaging tests and specimen tests. Medical images are interpreted and image interpretation reports are created by image interpretation doctors.
[0051] The client terminal 22 is a computer (including a tablet terminal, etc.) directly operated by medical staff such as a doctor, a medical technician, or a nurse. The client terminal 22 is capable of operating various servers in the server group 21 and operating the teacher data management device 10. A doctor can access the server group 21 using the client terminal 22, check the medical examination data, and store the diagnostic data generated based on the medical examination data in the teacher data management device 10. In addition, a doctor can specify medical examination data that he or she wishes to delete from among the medical examination data stored in the data storage unit 11 of the teacher data management device 10, and issue an instruction to delete the data.
[0052] The electronic medical records, medical images, and reports are each accompanied by a personal ID (identifier) assigned to each patient. The teacher data management device 10 may obtain, as medical examination data or diagnostic data, various information obtained by analyzing various data obtained from the electronic medical record server 24, the image server 25, the report server 26, etc., by a medical support device (not shown), and the results of a doctor's diagnosis based on this information. For example, the teacher data management device 10 may obtain, as medical examination data, various examination results obtained by the medical support device analyzing CT examination images obtained from the image server 25, and may obtain diagnostic data generated by a doctor using the client terminal 22 to diagnose based on the examination results.
[0053] The teacher data management device 10 stores teacher data obtained from the server group 21 in a data storage unit 11, the teacher data associating medical examination data obtained from a medical examination undertaken by an individual with diagnostic data generated based on the medical examination data.
[0054] The data storage unit 11 preferably stores the teacher data in such a manner that the personal ID of the individual who underwent the medical examination and the medical examination data are linked by a first pseudonymous ID, and the first pseudonymous ID and the diagnostic data are linked by a second pseudonymous ID.
[0055] It is preferable that the data storage unit 11 stores these data in a plurality of tables or a database. A table is a structure for organizing and storing related data in a tabular format, and stores tabular data consisting of rows and columns. A database includes a plurality of tables, and in some cases, may store relationships between the plurality of tables. It is more preferable that the data storage unit 11 stores these data in a table, since data can be easily deleted.
[0056] The data storage unit 11 preferably stores the personal ID, the medical examination data, and the diagnostic data in separate tables. It is preferable to store the personal ID, the medical examination data, and the diagnostic data in separate tables using the first pseudonym ID and the second pseudonym ID so that they can be traced by reverse lookup from the personal ID. Furthermore, by storing each data as a separate table, even if any of the separate tables is leaked, the medical examination data and / or the diagnostic data cannot be obtained directly from the personal ID, and vice versa, so that high security can be maintained from the viewpoint of protecting personal information.
[0057] The data storage unit 11 preferably generates a pseudonymous reverse lookup table consisting of a personal ID and a first pseudonymous ID, a pseudonymous medical examination table consisting of the first pseudonymous ID and medical examination data, a medical examination diagnosis reverse lookup table consisting of the first pseudonymous ID and a second pseudonymous ID, and a pseudonymous diagnosis table consisting of the second pseudonymous ID and diagnostic data.
[0058] 3, specifically, the data storage unit 11 preferably includes a pseudonym reverse lookup table 31 in which a personal ID is linked to a first pseudonym ID, a pseudonym medical examination table 32 in which the first pseudonym ID is linked to specific measurement data that is medical examination data, a medical examination diagnosis reverse lookup table 33 in which the first pseudonym ID is linked to a second pseudonym ID, and a pseudonym diagnosis table 34 in which the second pseudonym ID is linked to diagnostic data that is a result of a doctor's diagnosis from the medical examination data. With these tables, as described above, in the teacher data, the personal ID of an individual who has undergone a medical examination is linked to the medical examination data by the first pseudonym ID, and the first pseudonym ID is linked to the diagnostic data by the second pseudonym ID.
[0059] As shown in Fig. 4, the kana reverse lookup table 31 is data in which personal IDs 35 and first kana IDs 36 are linked together. Fig. 4 shows that "Tanaka Taro" is linked to the first kana ID 36 "ano001" as the personal ID 35, and similarly, "Kimura Hanako" is linked to the first kana ID 36 "ano002" as the personal ID 35.
[0060] As shown in Fig. 5, the pseudonym medical examination table 32 is data in which the first pseudonym ID 36 is linked to medical examination data 37, which is specific measurement data. In Fig. 5, the first pseudonym ID 36 "ano001" is linked to the medical examination data 37 "height: 177, weight: 67, γGTP: 28, etc.", and similarly, the first pseudonym ID 36 "ano002" is linked to the medical examination data 37 "height: 158, weight: 124, γGTP: 93, etc.".
[0061] As shown in Fig. 6, the medical examination diagnosis reverse lookup table 33 is data in which a first pseudonym ID 36 is linked to a second pseudonym ID 38. In Fig. 6, "ano001" is linked to the second pseudonym ID 38 "unknown001", and similarly, "ano002" is linked to the first pseudonym ID 36 "unknown002" as the second pseudonym ID 38.
[0062] As shown in Fig. 7, the medical examination diagnosis reverse lookup table 33 is data in which the second pseudonym ID 38 is linked to the diagnosis data 39. In Fig. 7, the second pseudonym ID 38 "unknown001" is linked to the diagnosis data 39 "no findings", and similarly, the second pseudonym ID 38 "unknown002" is linked to the diagnosis data 39 "obesity".
[0063] The medical examination data 37 stored as described above is obtained from the individual who underwent the medical examination, with the consent of the individual to the provision of information on the medical examination data 37 for various uses. Even if consent is given once, the individual can withdraw consent to the provision of information on the medical examination data 37. After consent is withdrawn, the medical examination data 37 of the individual who has withdrawn consent cannot be used. Therefore, the medical examination data 37 of the individual who has withdrawn consent must be deleted from the data storage unit 11. Even in cases other than when an individual withdraws consent to the information provision, there may be cases where the medical examination data 37 or the diagnostic data 39 is desired to be edited, such as by deleting the medical examination data 37 or the diagnostic data 39, in order to make the data more preferable teacher data.
[0064] When deleting the medical examination data 37 of a specific individual from the data storage unit 11, the medical examination data deletion unit 12 deletes the personal ID 35 of the specific individual in the pseudonym reverse lookup table 31 and the specific first pseudonym ID 36 linked thereto, the specific first pseudonym ID 36 in the pseudonym medical examination table 32 and the specific medical examination data 37 linked thereto, and the specific first pseudonym ID 35 and the specific second pseudonym ID 38 linked thereto in the medical examination diagnosis reverse lookup table 33. This deletes the medical examination data 37 of the individual who has revoked consent to information provision. In addition, the link between the diagnostic data 39 and the personal ID 35 is released, and the diagnostic data 39 can be left anonymous so that a specific individual cannot be identified.
[0065] As shown in Fig. 8, when the medical examination data deletion unit 12 deletes the medical examination data 37 of "Kimura Hanako" because the specific individual "Kimura Hanako" has withdrawn the information provision, the medical examination data deletion unit 12 deletes the individual ID 35 "Kimura Hanako" in the kana reverse lookup table and the first kana ID 36 "ano002" linked thereto, the specific first kana ID 36 "ano002" in the kana medical examination table 32 and the specific medical examination data 37 "height: 158, weight: 124, γGTP: 93, etc." linked thereto, and the specific first kana ID 35 "ano002" in the medical examination diagnosis reverse lookup table 33 and the specific second kana ID 38 "unknown002" linked thereto. This deletes the medical examination data 37 of the individual who has withdrawn the consent to the information provision. In addition, the link between diagnostic data 39 "obesity" and personal ID 35 "Kimura Hanako" is removed, and diagnostic data 39 "obesity" can be left anonymous so that it cannot be identified as belonging to "Kimura Hanako."
[0066] In the process in which the medical examination data deleting unit 12 deletes specific medical examination data 37, the diagnostic data specifying unit 13 stores the second pseudonym ID 38 corresponding to the medical examination data 37 to be deleted, and specifies the diagnostic data 39 associated with the deleted medical examination data 37 from the second pseudonym ID 38. At this time, the diagnostic data specifying unit 13 may have a copy of the medical examination diagnosis reverse lookup table 33. This makes it possible to easily obtain the diagnostic data 39 corresponding to the stored second pseudonym ID 38, while at the same time maintaining safety from the viewpoint of protecting personal information since it does not have data that identifies an individual.
[0067] As shown in FIG. 8, when the second pseudonym ID 38 "unknown002" is deleted, the diagnostic data identification unit 13 stores the second pseudonym ID 38 and identifies the diagnostic data 39 associated therewith as "obesity" by referring to a copy of the medical examination diagnosis reverse lookup table 33.
[0068] The pseudo-examination data generating unit 14 creates pseudo-examination data for the deleted examination data 37. At this time, it is preferable to calculate a reconstruction error so that the pseudo-examination data does not completely match the deleted examination data 37. That is, it is preferable that the pseudo-examination data generating unit 14 calculates a reconstruction error indicating the degree of match between the deleted examination data 37 and the pseudo-examination data, and the pseudo-examination data generating model generates the pseudo-examination data so that the absolute value of the reconstruction error becomes a positive number.
[0069] 9, the pseudo medical examination data generating unit 14 includes a pseudo data generating neural network unit (hereinafter referred to as a pseudo data generating NN unit) 41 and a reconstruction accuracy suppressing unit 42. By including the reconstruction accuracy suppressing unit 42, the pseudo data generating NN unit 41 outputs pseudo medical examination data which is pseudo to the deleted medical examination data 37 and does not completely match the deleted medical examination data 37.
[0070] The pseudo-data generating NN unit 41 is a trained model using a neural network model, and is trained to input the medical examination data 37 and diagnostic data 39 of deleted individuals from the medical examination data 37 in the pseudonymous medical examination table 32 and the diagnostic data 39 in the pseudonymous diagnostic table 34 as input data 51, and output pseudo-medical examination data that is pseudo to the input medical examination data 37 of each individual as output data 52.
[0071] If the trained model provided in the pseudo-data generating NN unit 41 has good accuracy, there is a possibility that the trained model will reproduce the same data as the deleted medical examination data 37. Therefore, it is preferable that the reconstruction accuracy suppression unit 42 intentionally suppresses the reconstruction accuracy.
[0072] As shown in Fig. 10, the pseudo medical examination data generating unit 14 acquires two pieces of data, "height: 177, weight: 67, γGTP: 28, diagnosis: no first appearance" and "height: 158, weight: 124, γGTP: 93, diagnosis: obesity", which are the medical examination data 37 and diagnosis data 39 of the deleted individual from the medical examination data 37 in the pseudo-name medical examination table 32 and the diagnosis data 39 in the pseudo-name diagnosis table 34, as input data 51. The pseudo medical examination data generating unit 14 outputs output data 52 by the pseudo data generating NN unit 41 and the reconstruction accuracy suppressing unit 42. The output data includes pseudo medical examination data 53 for each of the two pieces of data.
[0073] The output data 52 is generated by outputting pseudo medical examination data 53 "height: 180, weight: 65, γGTP: 30" in response to input data 51 "height: 177, weight: 67, γGTP: 28, diagnosis: no first visit" in association with diagnostic data 39 "diagnosis: no first visit". Similarly, the output data 52 is generated by outputting pseudo medical examination data 53 "height: 160, weight: 120, γGTP: 100" in response to input data 51 "height: 158, weight: 124, γGTP: 93, diagnosis: obesity" in association with diagnostic data 39 "diagnosis: obesity".
[0074] These pseudo medical examination data 53 are not completely identical to the deleted medical examination data 37 included in the input data 51, and have the same characteristics as the deleted medical examination data 37. Since the pseudo medical examination data 53 have the same characteristics as the deleted medical examination data 37, when the output pseudo medical examination data 53 is used as teacher data in a learning model, the results output by the learned model are not degraded compared to when the deleted medical examination data 37 is used as teacher data in the learning model. Therefore, the teacher data including the pseudo medical examination data 53 maintains stability as teacher data compared to before the medical examination data 37 is deleted.
[0075] 11, the reconstruction error calculated by the reconstruction accuracy suppression unit 42 is calculated from the input data 51 and the output data 52. The calculated reconstruction error 54 is fed back to the pseudo-data generation NN unit 41 when the pseudo-data generation NN unit 41 learns. This enables the pseudo-data generation NN unit 41 to suitably output pseudo medical examination data 53 that does not completely match the deleted individual's medical examination data 37 included in the input data 51.
[0076] The reconstruction error 54 may be calculated by any method that calculates the error between the two data, the input data 51 and the output data 52, but it is only necessary that the reconstruction error is a positive number, that is, the error is not set to 0. As a method for calculating the error between the two data, in addition to the error based on the absolute value of the simple difference, commonly used methods such as the mean absolute error (MAE), the mean squared error (MSE), and the root mean squared error (RMSE) can be used. Regardless of the method used, the pseudo data generating NN unit 41 can be controlled so that the error does not become 0, that is, so that the reconstruction error 54 becomes a positive number, thereby preventing the input data 51 and the output data 52 from matching.
[0077] Although the reconstruction error 54 is a positive number, if the positive number is too large, the difference between the deleted screening data 37 and the output pseudo screening data 53 will become too large, and they may not have similar characteristics, so it is preferable that the reconstruction error 54 be below a preset threshold value.
[0078] The pseudo-examination data creation model itself may have the function of the reconstruction accuracy suppression unit 42. That is, the pseudo-examination data creation model may be a model in which a generation model that generates pseudo-examination data 53 by inputting deleted examination data and corresponding diagnostic data, and an evaluation model that evaluates the generated pseudo-examination data 53 are connected. This is preferable because it prevents the pseudo-examination data creation model itself from outputting output data 52 that completely matches the input data 51.
[0079] Examples of such models include Conditional Generative Adversarial Networks (Conditional GAN), which are classified as supervised learning models having a generator that is a generative model and a discriminator that is an evaluation model, and Conditional Variational Autoencoders (Conditional VAE), which are classified as supervised learning models. By using a model having a generator and a discriminator and providing diagnostic data that is a correct label to train the model, it is possible to evaluate input data 51 and output data 52 and output appropriate output data 52.
[0080] The pseudo medical examination data replacement unit 15 replaces the teacher data in which the pseudo medical examination data 53, which is the output data 52, and the diagnostic data 39 are associated with each other with teacher data including the deleted medical examination data 37, and stores the replaced teacher data in the data storage unit 11. At this time, for verification when used as teacher data, the raw data, which is the medical examination data 37, may be made distinguishable from the pseudo medical examination data 53. Labels may be attached to the pseudo medical examination data 53, labels may be attached to the medical examination data 37, and the like.
[0081] In the above embodiment, a part of the plurality of medical examination data 37 is described as the pseudo medical examination data 53, but all of the plurality of medical examination data 37 may be the pseudo medical examination data 53. In this case, the generated teacher data is not all the medical examination data 37 actually obtained, but has the characteristics of the medical examination data 37, so it is highly safe from the viewpoint of protecting personal information, and since it has the characteristics of the actual medical examination data 37, it is highly appropriate as teacher data, and is excellent teacher data.
[0082] In addition, when all the medical examination data of the teacher data managed by the teacher data management device 10 is the pseudo medical examination data 53, security from the viewpoint of protecting personal information is maintained, and the teacher data managed by the teacher data management device 10 may be provided for various uses by transferring the teacher data managed by the teacher data management device 10 from within the medical facility to a third party outside the medical facility. Therefore, the teacher data management device 10 makes it possible to provide useful teacher data.
[0083] Next, the flow of processing by the teacher data management device 10 will be described with reference to the flowchart shown in Fig. 12. When consent is obtained for various uses of medical examination data 37 obtained when an individual undergoes a medical examination, the medical examination data 37 and diagnostic data 39 generated by a doctor based on the medical examination data 37 are associated with each other, and these are regarded as teacher data. The teacher data is stored in the data storage unit 11 for purposes such as learning a learning model in machine learning (step ST110).
[0084] For example, if an individual thinks that the individual can be identified by his / her own medical examination data 37, the individual withdraws consent to the provision of information regarding the use of the medical examination data 37. When consent is withdrawn, for example, a hospital employee who performed the medical examination issues an instruction from the client terminal 22 (see FIG. 2) to delete the medical examination data stored in the data storage unit 11 (step ST120). Based on the instruction, the medical examination data deletion unit 12 deletes the medical examination data 37 from the data storage unit 11 (step ST130).
[0085] When the medical examination data deleting unit 12 deletes the medical examination data 37, the diagnostic data specifying unit 13 obtains information of the medical examination data 37 to be deleted and specifies the diagnostic data 39 corresponding to the deleted medical examination data (step ST140). The pseudo medical examination data generating unit 14 also obtains information of the deleted medical examination data 37 and information of the diagnostic data 39 corresponding thereto, and generates the pseudo medical examination data 53 (step ST150). The pseudo medical examination data 53 may be generated in a form associated with the diagnostic data 39.
[0086] The pseudo medical examination data replacement unit 15 creates teacher data that associates the pseudo medical examination data 53 with the diagnostic data 39 corresponding to the deleted medical examination data 37, and stores the created teacher data in the data storage unit 11 in place of the teacher data having the deleted medical examination data 37.
[0087] In the above embodiment, the hardware structure of the processing units that execute various processes such as the data storage unit 11, the medical examination data deletion unit 12, the diagnostic data specification unit 13, the pseudo medical examination data generation unit 14, and the pseudo medical examination data replacement unit 15 is various processors as shown below. The various processors include a CPU (Central Processing Unit), which is a general-purpose processor that executes software (programs) and functions as various processing units, a GPU (Graphical Processing Unit), a programmable logic device (PLD), which is a processor whose circuit configuration can be changed after manufacture such as an FPGA (Field Programmable Gate Array), and a dedicated electric circuit, which is a processor having a circuit configuration designed exclusively for executing various processes.
[0088] One processing unit may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same type or different types (for example, multiple FPGAs, a combination of a CPU and an FPGA, or a combination of a CPU and a GPU, etc.). Also, multiple processing units may be configured with one processor. Examples of multiple processing units configured with one processor include a multi-color form in which one processor is configured with a combination of one or more CPUs and software, as represented by computers such as clients and servers, and this processor functions as multiple processing units. A mono-color form includes a form in which a processor is used to realize the functions of the entire system including multiple processing units with one IC (Integrated Circuit) chip, as represented by System On Chip (SoC). In this way, the various processing units are configured with one or more of the above-mentioned various processors as a hardware structure.
[0089] Furthermore, the hardware structure of these various processors is, more specifically, an electric circuit (circuitry) in the form of a combination of circuit elements such as semiconductor elements, and the hardware structure of the memory unit is a storage device such as a hard disk drive (HDD) or a solid state drive (SSD).
[0090] From the above description, it is possible to understand the teacher data management device 10 described in Supplementary Items 1 to 10 below.
[0091] [Additional note 1] A processor is provided. The processor, storing teacher data in a data storage unit, the teacher data including the medical examination data obtained by the medical examination of the individual and the diagnostic data generated based on the medical examination data in association with each other; Deleting the specific medical examination data from the data storage unit; Identifying the diagnostic data associated with the deleted medical examination data; generating pseudo medical examination data for the deleted medical examination data based on the identified diagnostic data; A teacher data management device that associates the pseudo medical examination data with the identified diagnostic data and stores it in the data storage unit.
[0092] [Additional note 2] The teacher data management device described in Appendix 1, wherein the specific medical examination data is medical examination data that includes information about which consent to information provision has been withdrawn from the individual.
[0093] [Additional note 3] The teacher data management device according to claim 1 or 2, wherein the pseudo medical examination data is medical examination data having similar characteristics to those of the deleted medical examination data.
[0094] [Additional note 4] 4. The teacher data management device according to any one of claims 1 to 3, wherein the pseudo medical examination data is medical examination data that does not completely match the deleted medical examination data.
[0095] [Additional note 5] The data storage unit of the teacher data management device described in any one of appendix 1 to 4 stores the teacher data associated with the personal ID, the medical examination data, and the diagnostic data by linking the personal ID of the individual who underwent a medical examination with the medical examination data using a first pseudonymous ID, and linking the first pseudonymous ID with the diagnostic data using a second pseudonymous ID.
[0096] [Additional note 6] the processor generates a pseudonymous reverse lookup table consisting of the personal ID and the first pseudonymous ID, a pseudonymous medical examination table consisting of the first pseudonymous ID and the medical examination data, a medical examination diagnosis reverse lookup table consisting of the first pseudonymous ID and the second pseudonymous ID, and a pseudonymous diagnosis table consisting of the second pseudonymous ID and the diagnostic data; The teacher data management device described in Appendix 5, wherein the data storage unit stores the kana reverse lookup table, the kana medical examination table, the medical examination diagnosis reverse lookup table, and the kana diagnosis table in the data storage unit.
[0097] [Additional note 7] The teacher data management device described in Appendix 6, when the processor deletes the medical examination data of a specific individual from the data storage unit, deletes the personal ID of the specific individual in the pseudonym reverse lookup table and the specific first pseudonym ID linked thereto, the specific first pseudonym ID in the pseudonym medical examination table and the specific medical examination data linked thereto, and the specific first pseudonym ID and the specific second pseudonym ID linked thereto in the medical examination diagnosis reverse lookup table.
[0098] [Additional note 8] The processor generates the pseudo medical examination data using a pseudo medical examination data generation model; The teacher data management device described in any one of appendix 1 to 7, wherein the pseudo-screening data generation model is a trained model that has been trained to generate the pseudo-screening data by inputting the deleted screening data and the diagnostic data.
[0099] [Additional note 9] The processor calculates a reconstruction error indicating a degree of match between the deleted medical examination data and the pseudo medical examination data; The teacher data management device according to claim 8, wherein the pseudo medical examination data creation model generates the pseudo medical examination data so that an absolute value of the reconstruction error becomes a positive number.
[0100] [Additional Note 10] The pseudo-examination data creation model of the teacher data management device described in Appendix 8 is a model connected to a generation model that generates the pseudo-examination data by inputting the deleted examination data and the diagnostic data, and an evaluation model that evaluates the generated pseudo-examination data. [Explanation of symbols]
[0101] 10. Teacher data management device 11 Data Storage Department 12 Medical examination data deletion section 13 Diagnostic data identification section 14. Pseudo-examination data generation unit 15 Pseudo-examination data replacement section 20 Medical Information Management System 21 Servers 22 Client terminal 23 Network 24 Electronic medical record server 24A Medical Record Database 25 Image Server 25A Image Database 26 Report Server 26A Report Database 31 Kana Reverse Lookup Table 32 Kana Medical Examination Table 33 Medical Examination Diagnosis Reverse Lookup Table 34 Pseudo-Diagnostic Table 35 Personal ID 36 First Pseudonym ID 37 Medical Examination Data 38 Second Pseudonym ID 39 Diagnostic Data 41 Pseudo Data Generation NN Section 42 Reconstruction accuracy suppression unit 51 Input Data 52 Output Data 53 Pseudo-examination data 54 Reconstruction error ST110~ST160 Step
Claims
1. A processor is provided. The processor, storing teacher data in a data storage unit, the teacher data including the medical examination data obtained by the medical examination of the individual and the diagnostic data generated based on the medical examination data in association with each other; Deleting the specific medical examination data from the data storage unit; Identifying the diagnostic data associated with the deleted medical examination data; generating pseudo medical examination data for the deleted medical examination data based on the identified diagnostic data; A teacher data management device that associates the pseudo medical examination data with the identified diagnostic data and stores it in the data storage unit.
2. The teacher data management device according to claim 1 , wherein the specific medical examination data is medical examination data including information for which consent to information provision has been revoked from the individual.
3. The teacher data management device according to claim 1 , wherein the pseudo medical examination data is medical examination data having the same characteristics as those of the deleted medical examination data.
4. The teacher data management device according to claim 1 , wherein the pseudo medical examination data is medical examination data that does not completely match the deleted medical examination data.
5. The teacher data management device of claim 1, wherein the data storage unit stores the teacher data in association with the personal ID, the medical examination data, and the diagnostic data by linking the personal ID of the individual who underwent the medical examination with the medical examination data using a first pseudonymous ID, and linking the first pseudonymous ID with the diagnostic data using a second pseudonymous ID.
6. the processor generates a pseudonym reverse lookup table consisting of the individual ID and the first pseudonym ID, a pseudonym medical examination table consisting of the first pseudonym ID and the medical examination data, a medical examination diagnosis reverse lookup table consisting of the first pseudonym ID and the second pseudonym ID, and a pseudonym diagnosis table consisting of the second pseudonym ID and the diagnostic data; The teacher data management device according to claim 5 , wherein the data storage unit stores the kana reverse lookup table, the kana medical examination table, the medical examination diagnosis reverse lookup table, and the kana diagnosis table in the data storage unit.
7. The teacher data management device of claim 6, wherein when the processor deletes the medical examination data of a specific individual from the data storage unit, the processor deletes the personal ID of the specific individual in the kana reverse lookup table and the specific first kana ID linked thereto, the specific first kana ID in the kana medical examination table and the specific medical examination data linked thereto, and the specific first kana ID in the medical examination diagnosis reverse lookup table and the specific second kana ID linked thereto.
8. The processor generates the pseudo medical examination data using a pseudo medical examination data generation model; The teacher data management device according to claim 1 , wherein the pseudo medical examination data generation model is a trained model that has been trained to generate the pseudo medical examination data by inputting the deleted medical examination data and the diagnostic data.
9. The processor calculates a reconstruction error indicating a degree of match between the deleted medical examination data and the pseudo medical examination data; The teacher data management device according to claim 8 , wherein the pseudo medical examination data creation model generates the pseudo medical examination data so that an absolute value of the reconstruction error becomes a positive number.
10. The teacher data management device of claim 8, wherein the pseudo-examination data creation model is a model connected to a generation model that generates the pseudo-examination data by inputting the deleted examination data and the diagnostic data, and an evaluation model that evaluates the generated pseudo-examination data.
11. storing teacher data in a data storage unit, the teacher data correlating the medical examination data with the diagnostic data generated based on the medical examination data; deleting the specific medical examination data from the data storage unit; identifying the diagnostic data associated with the deleted medical data; generating pseudo medical examination data for the deleted medical examination data based on the identified diagnostic data; and storing the pseudo medical examination data in the data storage unit in association with the identified diagnostic data.
12. On the computer, a function of storing the medical examination data and the diagnostic data generated based on the medical examination data as teacher data in a data storage unit; A function of deleting specific medical examination data from the data storage unit; Identifying the diagnostic data associated with the deleted medical examination data; A function of generating pseudo medical examination data for the deleted medical examination data based on the identified diagnostic data; and storing the pseudo medical examination data in the data storage unit in association with the identified diagnostic data.
Citation Information
Patent Citations
Providing device, providing method, and providing program
JP2022143911A
Cited By
Information processing device, information processing method, and program
JP7855772B1
