Learning system, learning device, learning method, and program

Abstract

To increase the accuracy of a model.SOLUTION: A learning system of an embodiment has an acquisition unit, a selection unit, and an update unit. The acquisition unit acquires a first data distribution related to a first dataset of datasets based on a first cohort. The selection unit selects a second cohort used for updating a first model from a plurality of second cohorts on the basis of the first data distribution acquired by the acquisition unit. The update unit updates the first model on the basis of at least part of a second dataset of datasets based on the second cohort selected for the first model on the basis of at least part of the datasets based on the second cohort.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The embodiments disclosed in this specification and the drawings relate to a learning system, a learning device, a learning method, and a program. [Background technology]

[0002] In the medical field, data held by clients such as websites is highly confidential, so distributed learning (online learning), which allows models to be built without each client directly sharing their data, is a learning method that is highly compatible with the medical field. In distributed learning, for example, a model (local model) trained by each client based on a data set based on a cohort such as a patient is used.

[0003] However, in medical facilities such as small clinics and small hospitals, the number of factors in the training data for subjects and other subjects is small, and the training data for training the local model may be insufficient. If the training data is insufficient, it is difficult to generate a model with good accuracy even if the client learns the local data. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Special Publication No. 2017-519282 [Patent Document 2] Special Publication No. 2017-520824 [Patent Document 3] Special Publication No. 2020-530917 Summary of the Invention [Problem to be solved by the invention]

[0005] The problem to be solved by the embodiments disclosed in this specification and the drawings is to improve the accuracy of the model. However, the problem to be solved by the embodiments disclosed in this specification and the drawings is not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described later can also be positioned as other problems. [Means for solving the problem]

[0006] A learning system according to an embodiment includes an acquisition unit, a selection unit, and an update unit. The acquisition unit acquires a first data distribution for a first dataset from a dataset based on a first cohort. The selection unit selects a second cohort to be used for updating a first model from among a plurality of second cohorts based on the first data distribution acquired by the acquisition unit. The update unit updates the first model based on at least a portion of the dataset based on the second cohort, based on at least a portion of the second dataset from the dataset based on the selected second cohort. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a configuration diagram of a learning system 1 according to a first embodiment. [Diagram 2] FIG. 1 is a configuration diagram of a target site 100A. [Diagram 3] FIG. 13 is a configuration diagram of a candidate site 100B. [Figure 4] FIG. 2 is a diagram showing the configuration of a central server 200. [Diagram 5] FIG. 2 is a sequence diagram showing the flow of processing in the learning system 1. [Figure 6] 11 is a flowchart showing a procedure for selecting a selection site 300. [Figure 7] FIG. 2 is a diagram showing an image of selecting a selection site 300. [Figure 8] FIG. 11 is a block diagram of a target site 100A in the learning system of the second embodiment. [Figure 9] FIG. 1 is a configuration diagram of a managed site 100D. [Figure 10]FIG. 13 is a configuration diagram of a central server 200 according to a third embodiment. [Figure 11] FIG. 13 is a sequence diagram showing a process flow in the learning system of the third embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] Hereinafter, a learning device, a learning system, a learning method, and a program according to the embodiments will be described with reference to the drawings. In the following description, components having the same functions are denoted by the same reference numerals, and the description thereof may be omitted.

[0009] (First embodiment) 1 is a configuration diagram of a learning system 1 of the first embodiment. The learning system 1 includes, for example, a plurality of sites 100 and a central server 200. The plurality of sites 100 are connected to the central server 200 via a communication network NW so as to be able to transmit and receive data. The central server 200 generates a global model by distributed learning using learning data acquired by the plurality of sites 100, and provides the global model to the plurality of sites 100.

[0010] The communication network NW refers to any information and communication network that uses electrical communication technology. The communication network NW includes wireless / wired LANs such as hospital backbone LANs (Local Area Networks) and the Internet, as well as telephone communication line networks, optical fiber communication networks, cable communication networks, and satellite communication networks.

[0011] The site 100 includes a learning device installed in, for example, a medical facility. The site 100 collects a cohort-based dataset and operates a trained model. By operating the model, the site 100 collects or provides information on, for example, treatment or diagnosis in a medical facility. The site 100 includes, for example, a target site 100A, a candidate site 100B, and a plurality of other sites 100C. The target site 100A is, for example, a small site installed in a medical facility such as a small clinic or a small hospital. The target site 100A is an example of a first site.

[0012] The candidate site 100B is, for example, a large-scale site installed in a large medical facility such as a general hospital. The site 100 includes a plurality of candidate sites 100B. The other sites 100C are either the target site 100A or the candidate sites 100B or sites of a different scale.

[0013] In the following description, when distinguishing between the candidate sites 100B, they are assigned sub-numbers and referred to as a first candidate site 100B-1, a second candidate site 100B-2, a third candidate site 100B-3, .... One selected from among the multiple candidate sites 100B becomes the selected site 300. The candidate site 100B is an example of a second site.

[0014] The central server 200 is provided, for example, in a facility other than a medical facility. The central server 200 collects information from the sites 100, analyzes the collected information, generates information to be provided to the medical facility, and transmits the generated information to the site 100 to provide it to the medical facility. The central server 200 may be provided in the medical facility.

[0015] 2 is a configuration diagram of the target site 100A. The target site 100A includes, for example, a communication interface 110, an input interface 120, a processing circuit 130, and a memory 140. The communication interface 110 communicates with an external device, for example, a central server 200, via a communication network NW. The communication interface 110 includes, for example, a communication interface such as a NIC (Network Interface Card).

[0016] The input interface 120 accepts various input operations by an operator, and outputs an electrical signal indicating the content of the accepted input operation to the processing circuit 130. The input interface 120 is realized by, for example, a mouse, a keyboard, a touch panel, a drag ball, a switch, a button, a joystick, a camera, an infrared sensor, a microphone, etc.

[0017] In this specification, the input interface is not limited to an interface having physical operation parts such as a mouse, a keyboard, etc. For example, an example of the input interface also includes an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the device and outputs the electrical signal to a control circuit.

[0018] The processing circuit 130 has, for example, a processor such as a CPU (Central Processing Unit). The processing circuit 130 controls the overall operation of the site 100. The processing circuit 130 includes, for example, a collection function 131, a data distribution calculation function 132, an update request function 133, and an acceptance function 134. The processing circuit 130 realizes these functions by, for example, a hardware processor executing a program stored in a storage device (storage circuit).

[0019] The hardware processor refers to a circuit such as a central processing unit (CPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a programmable logic device (e.g., a simple programmable logic device (SPLD) or a complex programmable logic device (CPLD), or a field programmable gate array (FPGA). Instead of storing a program in a storage device, the program may be directly embedded in the circuit of the hardware processor. In this case, the hardware processor realizes a function by reading and executing the program embedded in the circuit. The hardware processor is not limited to being configured as a single circuit, and may be configured as one hardware processor by combining multiple independent circuits to realize each function. The storage device may be a non-transitory (hardware) storage medium. Furthermore, multiple components may be integrated into one hardware processor to realize each function.

[0020] Each component of the processing circuit 130 may be distributed and realized by multiple pieces of hardware. The processing circuit 130 may not be a component of the site 100, but may be realized by a processing device capable of communicating with the site 100. Each function included in the processing circuit 130 may be distributed among multiple circuits, or may be made available by starting application software stored in the memory 140.

[0021] The memory 140 is included in a storage device. The memory 140 stores a medical information database (hereinafter, DB) 141 and a model information DB 142. The medical information DB 141 includes medical data, which is information obtained by medical treatment of a patient. Factors of the medical data include, for example, attributes of the patient such as the patient's age, sex, and physical build, the doctor in charge, the name of the disease, symptoms, date and time, and season. The model information DB 142 includes information on a local model (hereinafter, first model) that is used in the target site 100A and operated in the medical facility in which the target site 100A is installed.

[0022] The collection function 131 collects medical data of a first cohort associated with the target site 100A (hereinafter, first medical data). The first medical data is collected, for example, when a patient is examined or treated in a medical facility where the target site 100A is installed. The collection function 131 adds the collected first medical data to the first medical data included in the medical information DB 141 in the memory 140 and stores it. The first medical data is an example of a first data set.

[0023] When the first data distribution calculation condition is satisfied, the data distribution calculation function 132 calculates a first data distribution for the first dataset among the datasets based on the first cohort, based on the first medical data included in the medical information DB 141 stored in the memory 140. The data distribution calculation function 132 is an example of a data distribution calculation unit. The first dataset is, for example, data suitable for updating a first model operated in a medical facility where the target site is installed, but when the medical facility is, for example, small, the amount of data may be insufficient for updating the first model.

[0024] The data distribution calculation function 132 transmits the calculated first data distribution to the central server 200 using the communication interface 110. The data distribution calculation function 132 transmits an ID, which is identification information for identifying its own site, together with the first data distribution to the central server 200. The data distribution calculation function 132 may store the calculated first data distribution in the memory 140.

[0025] The first data distribution calculation condition may be any condition. For example, the first data distribution calculation condition may be that the first medical data is newly collected by the collection function 131 of the target site 100A, that the number of first medical data added to the medical information DB 141 reaches a predetermined number, or that a certain period of time, for example, 10 days, has passed.

[0026] When a first model update condition included in the model information DB stored in the memory 140 is satisfied, the update request function 133 transmits update request information to the central server 200 using the communication interface 110. When transmitting the update request information to the central server 200, the update request function 133 transmits the first model together with the update request information to the central server. The update request function 133 is an example of an update request unit.

[0027] The first model update condition may be any condition. For example, the first model update condition may be that the operator has performed an input operation to the input interface 120 to request an update of the first model. The first model update condition may be that a predetermined update period has elapsed. The first model update condition may be that the amount of newly collected first medical data has reached a predetermined number.

[0028] The acceptance function 134 accepts the updated first model transmitted by the central server 200. The acceptance function 134 stores the accepted updated first model in the memory 140. The target site 100A, for example, operates the accepted updated first model. The acceptance function 134 is an example of an acceptance unit.

[0029] 3 is a configuration diagram of the candidate site 100B. Like the target site 100A, the candidate site 100B includes, for example, a communication interface 110, an input interface 120, a processing circuit 130, and a memory 140. Compared to the processing circuit of the target site 100A, the processing circuit 130 of the candidate site 100B does not include an update request function 133, but instead includes an extraction function 135, an update function 136, a verification function 137, and a transmission function 138. In the following description, the local model stored as a part of the model information DB 142 in the memory 140 of the candidate site 100B is referred to as a second model.

[0030] The collection function 131 of the candidate site 100B collects medical data of a second cohort (hereinafter, second medical data) associated with the candidate site 100B. The collection function 131 adds the collected second medical data to the second medical data included in the medical information DB 141 of the memory 140 and stores it.

[0031] The data distribution calculation function 132 of the candidate site 100B calculates a second data distribution for a second data set among data sets based on a second cohort, which is a cohort of the candidate site 100B, based on second medical data included in the medical information DB 141 stored in the memory 140 of the candidate site 100B. The second medical data is an example of the second data set.

[0032] The candidate site 100B is a large-scale site, and the target site 100A is a small-scale site. Therefore, the amount of data of factors (features) stored in the memory 140 of the candidate site 100B is greater than the amount of data of factors stored in the memory 140 of the target site. The amount of data of factors stored in the memory 140 of the target site is the amount of data of the first cohort, and the amount of data of factors stored in the memory 140 of the candidate site 100B is the amount of data of the second cohort.

[0033] The data distribution calculation function 132 transmits the calculated second data distribution to the central server 200 using the communication interface 110. The data distribution calculation function 132 transmits the ID of its own site and information indicating the data amount of the factor stored in the memory 140 (hereinafter, data amount information) together with the second data distribution to the central server 200. The data distribution calculation function 132 may store the calculated second data distribution in the memory 140.

[0034] The extraction function 135 extracts a data set to be used for updating the first model from the second medical data based on the first data distribution transmitted by the central server 200 together with the update request. The data set to be used for updating the first model is a part of the second medical data. The data set to be used for updating the first model may be the entire second medical data. The extraction function 135 is an example of an extraction unit.

[0035] The update function 136 updates the second model included in the model information DB 142 in the memory 140 of the candidate site 100B by learning such as machine learning using the second medical data stored in the memory 140 as learning data, to generate an updated second model. The update function 136 stores the generated updated second model in the model information DB 142 as a new second model.

[0036] In response to an update request from the central server 200, the update function 136 updates the first model transmitted by the central server 200 by learning such as machine learning to generate an updated first model. The update function 136 uses at least a part of the second medical data stored in the memory 140 as learning data. The update function 136 is an example of an update unit.

[0037] The verification function 137 verifies the updated first model updated by the update function 136. When verifying the updated first model, the verification function 137 uses, for example, a data set of the second medical data other than the data set used to update the first model. The verification function 137 is an example of a verification unit.

[0038] The sending function 138 uses the communication interface 110 to send the updated first model generated by the update function 136 and verified by the verification function 137 to the central server 200. The sending function 138 is an example of a sending unit. The sending function 138 may send to the central server 200 the updated first model generated by the update function and not verified by the verification function 137.

[0039] 4 is a configuration diagram of the central server 200. The central server 200 is provided, for example, in a management facility that centrally manages information provided by or to the multiple sites 100. The central server 200 includes, for example, a communication interface 210, an input interface 220, a processing circuit 230, and a memory 240. The communication interface 210 communicates with an external device, for example, the multiple sites 100, via a communication network NW.

[0040] The communication interface 210 includes a communication interface such as a NIC. Distribution data and provided models are transmitted from the multiple sites 100 to the central server 200. Therefore, multiple distribution data and provided models are transmitted to the central server 200. The input interface 220 accepts various input operations by an operator, and outputs an electrical signal indicating the contents of the accepted input operation to the processing circuit 230.

[0041] The processing circuitry 230 has a processor such as a CPU. The processing circuitry 230 controls the overall operation of the central server 200. The processing circuitry 230 has, for example, an acquisition function 231 and a selection function 232. The processing circuitry 230 realizes these functions by, for example, a hardware processor executing a program stored in a storage device (storage circuit).

[0042] Each component of the processing circuit 230 may be distributed and realized by multiple pieces of hardware. The processing circuit 230 may not be a component of the central server 200, but may be realized by a processing device capable of communicating with the central server 200. Each function included in the processing circuit 230 may be distributed to multiple circuits, or may be made available by starting application software stored in the memory of the central server 200. A distribution data DB 241 is stored in the memory 240.

[0043] The acquisition function 231 acquires a first data distribution transmitted by the target site 100A and received by the communication interface 110. The acquisition function 231 is an example of an acquisition unit. The acquisition function 231 further acquires a second data distribution transmitted by the candidate site 100B.

[0044] The acquisition function 231 adds and stores the second data distribution from among the acquired data distributions to the second data distribution included in the distribution data DB 241 in the memory 240. The acquisition function 231 acquires the second distribution data transmitted by each of the multiple candidate sites 100B. The distribution data DB 241 includes multiple pieces of second distribution data.

[0045] The acquisition function 231 may acquire the first data distribution and the second data distribution at any timing. For example, the acquisition function 231 may periodically request the target site 100A and the candidate site 100B to transmit data distributions, and acquire the data distributions transmitted by each site 100 in response to the request. The central server 200 transmits a global model to the target site 100A and the candidate site 100B, and may request each site 100 to transmit a data distribution when transmitting the global model, and acquire the data distributions transmitted by each site 100 in response to the request.

[0046] The acquisition function 231 may acquire the received data distribution when each site 100 periodically calculates a data distribution and transmits the data distribution to the central server 200. The acquisition function 231 may acquire the received data distribution when each site 100 calculates a data distribution for each timing at which medical data is collected and transmits the data distribution to the central server 200.

[0047] The acquisition function 231 acquires an updated first model that is selected by the selection function 232 and transmitted by a candidate site 100B (hereinafter, selected site 300) that is suitable for updating the first model operated at the target site 100A. The acquisition function 231 transmits the acquired updated first model to the target site 100A using the communication interface 210.

[0048] The selection function 232 selects a selection site 300 from among the multiple candidate sites 100B based on the first data distribution acquired by the acquisition function 231 and the second distribution data stored in the memory 240. The selection function 232 selects the selection site 300 based on, for example, the similarity between the first data distribution and the second data distribution. The selection function 232 is an example of a selection unit.

[0049] The selection function 232 transmits a model update request for requesting an update of the first model to the selected selection site 300. When transmitting the model update request to the selection site 300, the selection function 232 transmits the first model transmitted by the target site 100A to the selection site 300 together with the model update request.

[0050] Next, a description will be given of the processing of the learning system 1. Fig. 5 is a sequence diagram showing the processing flow in the learning system 1. In the learning system 1, the target site 100A requests the central server 200 to update the first model, and the central server 200 transmits the updated first model (updated first model) to the target site 100A.

[0051] In processing the learning system 1, first, the target site 100A extracts and acquires the local model (first model) from the model information DB 142 stored in the memory 140 as model information when the update condition for the first model is met using the update request function 133 (step S101).

[0052] When the update request function 133 acquires the first model, the data distribution calculation function 132 calculates a first data distribution based on the first medical data of the medical information DB 141 stored in the memory 140 (step S103). Next, the update request function 133 and the data distribution calculation function 132 transmit the acquired first model and the calculated first data distribution to the central server 200 using the communication interface 110 (step S105).

[0053] On the other hand, when the second data distribution calculation condition is satisfied, the candidate site 100B calculates the second data distribution based on the second medical data in the medical information DB 141 stored in the memory 140 in the data distribution calculation function 132 (step S201). The second data distribution calculation condition may be any condition. The second data distribution calculation condition may be, for example, that new second medical data has been collected by the collection function 131 of the candidate site 100B, that the number of second medical data added to the medical information DB 141 has reached a predetermined number, or that a certain period of time, for example, 10 days, has passed.

[0054] The candidate site 100B transmits the second data distribution calculated by the data distribution calculation function 132 to the central server 200. When transmitting the second data distribution, the candidate site 100B also transmits data amount information to the central server 200 (step S203). The candidate site 100B repeats the process of step S201 to step S203 every time the second data distribution calculation condition is satisfied. The central server 200 adds and stores the second data distribution transmitted by the candidate site 100B to the second data distribution included in the distribution data DB 241 in the memory 240 (step S301).

[0055] The central server 200, which has received the model update request, extracts a plurality of second data distributions from the distribution data DB 241 stored in the memory 240 by the selection function 232 (step S303). Next, the selection function 232 selects a selection site 300 from among the plurality of candidate sites 100B based on the first data distribution transmitted by the target site 100A (step S305).

[0056] In selecting the selected site 300, the selection function 232 compares the first data distribution transmitted by the target site 100A with each of the multiple second data distributions transmitted by the multiple candidate sites 100B. The selection function 232 selects the candidate site 100B that transmitted the second data distribution that is closest to the first data distribution from among the multiple candidate sites 100B as the selected site 300.

[0057] 6 is a flowchart showing a procedure for selecting a selected site 300. The selection function 232 acquires a first data distribution, a second data distribution of the multiple candidate sites 100B, and an amount of data available for updating the first model held by the multiple candidate sites 100B (step S401).

[0058] Next, the selection function 232 compares the first data distribution with the second data distribution of each of the multiple selection site candidates, and calculates the candidate site 100B having the second data distribution closest to the first data distribution as the first selection site candidate (step S403). The selection function 232 uses, for example, KL divergence when calculating the first selection site candidate. FIG. 7 is a diagram showing an image of selecting the selection site candidate. Here, the first data distribution and the second data distribution are both distributions that represent any of multiple types of factors (age in this example) using histograms.

[0059] 7, the first to third large scale histograms HA to HC and the small scale histogram are all histograms with the subject's age on the horizontal axis and frequency on the vertical axis. The first large scale histogram HA represents the second data distribution of the first candidate site 100B-1, the second large scale histogram HB represents the second data distribution of the second candidate site 100B-2, and the third large scale histogram HC represents the second data distribution of the third candidate site 100B-3. The small scale histogram HD represents the first data distribution of the target site 100A.

[0060] The selection function 232 that performs the KL divergence selects an index value δ(p 1 ,p 2 ) is calculated using the following formula (1). 1 ,p 2 ) indicates the distance between each of the first to third large scale histograms HA to HC and the small scale histogram HD.

[0061]

number

[0062] In the above formula (1), p 1 is the probability density function of the first data distribution, p 2 is the probability density function of the second data distribution, and ξ is the variable of the probability density function for each factor.

[0063] The selection function 232 selects an index value δ(p 1 ,p 2 ) and calculate the calculated index value δ(p 1 ,p 2) is calculated as the index value δ of each candidate site 100B. After the selection function 232 calculates the index value δ of one candidate site 100B, it also calculates the index values ​​δ of the other candidate sites 100B using the above formula (1). For example, the first index value δ_A is the index value calculated for the first candidate site 100B-1, the second index value δ_B is the index value calculated for the second candidate site 100B-2, and the third index value δ_C is the index value calculated for the third candidate site 100B-3. The selection function 232 uses the calculated index values ​​of each of the multiple sites to calculate the first selection site candidate using the following formula (2).

[0064]

number

[0065] In the above formula (2), N is the number of types of factors in the candidate site 100B, Sj is the candidate site 100B, j is the identification (ID) of the candidate site 100B, and S (hat) is the target site. The index value δ calculated for each candidate site 100B has a relationship representing the similarity with the first data distribution. For example, the smaller the index value δ of the candidate site 100B, the higher the similarity between the second data distribution and the first data distribution in the candidate site 100B.

[0066] In the example shown in FIG. 7, the first index value δ_A of the first candidate site 100B-1 is 0.1, the second index value δ_B of the second candidate site 100B-2 is 0.7, and the third index value δ_C of the third candidate site 100B-3 is 0.8. In this case, the second data distribution of the first candidate site 100B-1 has the highest similarity to the first data distribution. The second candidate site 100B-2 and the third candidate site 100B-3 have the highest similarity to the first data distribution. As a result, the selection function 232 calculates the first candidate site 100B-1 as the first dam site candidate.

[0067] Next, the selection function 232 determines whether the first selection site candidate, for example, the first candidate site 100B-1, has the amount of data required to update the first model (hereinafter, the required data amount) (step S405). If it is determined that the first selection site candidate has the required data amount, the selection function 232 selects the first selection site candidate as the selected site 300 (step S407), and ends the process shown in FIG.

[0068] If the selection function 232 determines that the first selected site candidate does not have the required data volume, it sets the next candidate site, for example, the second candidate site 100B-2, as the selected site candidate (step S409), and proceeds to step S405. Thereafter, the selection function 232 proceeds with the processes from step S405 to step S409, and finally selects the selected site candidate determined in step S405 to have the required data volume as the selected site 300, and ends the process shown in FIG.

[0069] 5, the selection function 232 transmits a model update request to the selected selection site 300 (step S307). When transmitting the model update request to the selection site 300, the selection function 232 transmits the first data distribution and the first model together with the model update request.

[0070] The selected site 300, which has been selected from among the multiple candidate sites 100B and has received the model update request, extracts a data set to be used for updating the first model from the second medical data using the extraction function 135 (step S211). Next, the update function 136 updates the first model transmitted by the central server 200 by learning such as machine learning using the second medical data as learning data (step S213). Next, the verification function 137 verifies the updated first model generated by updating the first model using the update function 136 (step S215).

[0071] Next, the transmission function 138 in the selection site 300 transmits the updated first model verified by the verification function 137 to the central server 200 (step S217). Next, the central server 200 transmits the updated first model transmitted by the selection site 300 to the target site 100A as is (step S311). The target site 100A accepts the updated first model transmitted by the central server 200 by the acceptance function 134 (step S111). In this way, the learning system 1 ends the processing shown in FIG. 5.

[0072] In the learning system 1 of the first embodiment, the first model operated in the target site 100A is updated in the candidate site 100B. Therefore, the first model can be updated by a second data set based on the second cohort in the candidate site 100B. Here, the first model is updated using a second data set of a second data distribution similar to the first data distribution. Therefore, the accuracy of the first model (model) can be improved.

[0073] Second Embodiment Next, a second embodiment will be described. Fig. 8 is a block diagram of a target site 100A in a learning system of the second embodiment. The learning system of the second embodiment is mainly different from the learning system 1 of the first embodiment in that a verification function 137 is provided in the processing circuit 130 of the target site 100A.

[0074] In the learning system 1 of the second embodiment, similarly to the first embodiment, the central server 200 selects a selection site 300 based on the first data distribution and the second data distribution, and transmits a first model to the selected selection site 300. The selection site 300 updates the first model based on the received first model and the second medical data stored in the memory 140 to generate an updated first model, and transmits the generated updated first model to the central server 200. The central server 200 transmits the received updated first model to the target site 100A.

[0075] The collection function 131 in the target site 100A stores the first medical data (hereinafter, new first medical data) collected after the update request function 133 transmits the first model to the central server 200 in the memory 140. The verification function 137 verifies the suitability of the updated first model updated and generated by the update function 136 of the selection site 300 for the first cohort.

[0076] The verification function 137 receives the updated first model transmitted by the central server 200 using the communication interface 110. The verification function 137 verifies the received updated first model using new first medical data stored in the memory 140. The verification function 137 is an example of a verification unit.

[0077] The learning system of the second embodiment has the same operational effects as the learning system 1 of the first embodiment. Furthermore, the learning system of the second embodiment verifies the updated first model generated by the selection site 300 with new first medical data. Therefore, it is possible to determine the suitability of the updated first model when used at the site itself.

[0078] (Third embodiment) Next, a third embodiment will be described. The learning system of the third embodiment further includes a plurality of managed sites 100D as the sites 100. The central server 200 transmits and distributes the first model operated at the plurality of managed sites 100D to each of the managed sites 100D. The central server 200 manages updates of the first model at the managed sites 100D, and when updating the first model, requests a selected site 300 selected from among the plurality of candidate sites 100B to update the first model.

[0079] Fig. 9 is a configuration diagram of the managed site 100D. Fig. 10 is a configuration diagram of the central server 200 of the third embodiment. The managed site 100D, like the target site 100A and the candidate site 100B of the first embodiment, includes, for example, a communication interface 110, an input interface 120, a processing circuit 130, and a memory 140. The processing circuit 130 of the managed site 100D includes a collection function 131 and a data distribution calculation function 132, but does not include an update request function or an update function.

[0080] The central server 200 of the third embodiment, like the central server of the first embodiment, includes, for example, a communication interface 210, an input interface 220, a processing circuit 230, and a memory 240. The processing circuit 230 in the central server 200 of the third embodiment includes, for example, an acquisition function 231, a selection function 232, and a management function 233. The acquisition function 231 and the selection function 232 have the same functions as those of the first embodiment.

[0081] The management function 233 manages the first models in the multiple managed sites 100D. As part of the management of the first models, the management function 233 updates the first models in each of the multiple managed sites 100D as necessary. The management function 233 identifies a target site 100A for which the first model is to be updated from among the multiple managed sites 100D.

[0082] Next, the processing of the learning system of the third embodiment will be described. Fig. 11 is a sequence diagram showing the flow of processing in the learning system of the third embodiment. In processing of the learning system 1, first, when the first data distribution calculation condition is established, the managed site 100D calculates the first data distribution based on the first medical data stored in the medical information DB 141 in the data distribution calculation function 132 (step S501).

[0083] Next, the data distribution calculation function 132 transmits the calculated first data distribution to the central server 200 using the communication interface 110 (step S503). The managed site 100D repeats the processes of steps S501 and S503 every time the first data distribution calculation condition is satisfied.

[0084] On the other hand, when the second data distribution calculation condition is satisfied, the candidate site 100B uses the data distribution calculation function 132 to calculate a second data distribution based on the second medical data stored in the medical information DB 141 (step S601). The second data distribution calculation condition is the same as that in the first embodiment. Then, the data distribution calculation function 132 transmits the calculated second data distribution to the central server 200 (step S603). The candidate site 100B repeats the processes of steps S601 and S603 every time the second data distribution calculation condition is satisfied.

[0085] The central server 200 stores the first data distribution transmitted by the managed site 100D and the second data distribution transmitted by the candidate site 100B in the distribution data DB 241 (step S701). Next, the central server 200 identifies a target site for which the target site identification condition is met (step S703).

[0086] The target site identification condition may be any condition. For example, the target site identification condition may be that the first data distribution transmitted by the managed site 100D has been received, that a predetermined number of first data distributions have been received, or that a certain period of time has passed. The target site identification condition is determined for each of the multiple managed sites 100D, and the central server 200 identifies the managed site 100D for which the target site identification condition is met as the target site 100A.

[0087] Next, the selection function 232 selects a selection site 300 from among the multiple candidate sites 100B based on the first data distribution transmitted by the identified target site 100A (step S705). The procedure for selecting the selection site 300 is the same as in the first embodiment. Next, the selection function 232 transmits a model update request to the selected selection site 300 (step S707). When transmitting the model update request to the selection site 300, the selection function 232 transmits the first model of the identified target site 100A together with the model update request.

[0088] The selection site 300 that has received the model update request extracts the second medical data stored in the memory 140 by the extraction function 135 (step S611). Then, the update function 136 updates the first model transmitted by the central server 200 by learning such as machine learning using the second medical data as learning data (step S613). Then, the verification function 137 verifies the updated first model generated by updating the first model by the update function 136 (step S615).

[0089] Next, the transmission function 138 in the selection site 300 transmits the updated first model verified by the verification function 137 to the central server 200 (step S617). The central server 200 receives the updated first model transmitted by the selection site 300 and stores it in the memory 240 (step S711). Next, the central server 200 transmits the received updated first model to the target site 100A identified in step S703 (step S713). The target site 100A accepts the updated first model transmitted by the central server 200 by the acceptance function 134 (step S511). In this way, the learning system ends the process shown in FIG. 11.

[0090] The learning system of the third embodiment has the same effect as the first embodiment. In addition, the learning system of the third embodiment manages the first model operated at the managed sites 100D such as multiple small sites in the central server 200, and transmits the updated first model to each managed site 100D. At this time, the selected site 300 such as a large site is requested to update the first model, so that the model of the managed site can be updated with high accuracy.

[0091] In the above embodiment, the target site 100A is provided with the update request function 133, and the candidate site 100B is provided with the update function 136. However, at least one of the update request function 133 and the update function 136 may be provided in the site 100, regardless of the size of the site 100.

[0092] For example, the target site 100A may have an update function to receive the updated first model and update the first model at its own site. In this case, the target site 100A may transmit the first data distribution to the central server 200 when it transmits the updated first model to the central server 200.

[0093] In the above embodiment, the first data distribution is transmitted to the selection site 300 and the updated first model is transmitted to the target site 100A via the central server 200, but information may be transmitted and received between the sites without going through the central server 200. For example, the target site 100A may have a data distribution calculation function and a selection function, transmit the first data distribution to the selected selection site 300, and include an acceptance unit that accepts the updated first model transmitted by the selection site 300. In this case, the selection site 300 may have an acquisition function that acquires the first data distribution, an update function that updates the first model to generate an updated first model, and a transmission function that transmits the updated first model to the target site 100A.

[0094] In the above second embodiment, the verification function 137 is provided in the target site 100A having the update request function 133. In contrast, the verification function may be provided in the candidate site 100B having the update function 136. In this case, at least one of the first medical data and the first data distribution may be transmitted to the candidate site 100B by the target site 100A.

[0095] In the above embodiment, the target site 100A transmits the first data distribution to the central server 200, and the central server compares the first data distribution with the second data distribution to select the selection site 300. However, the increase in the first data distribution may be used instead of or in addition to the first data distribution. In this case, the increase in the second data distribution may be used instead of or in addition to the second data distribution to select the selection site 300. The increase in the first data distribution may be calculated by the target site 100A and transmitted to the central server 200, or may be calculated by the central server 200. Similarly, the increase in the second data distribution may be calculated by the candidate site 100B and transmitted to the central server 200, or may be calculated by the central server 200.

[0096] According to at least one of the embodiments described above, the accuracy of the model can be improved by having an acquisition unit that acquires a first data distribution for a first dataset from a dataset based on a first cohort, a selection unit that selects a second cohort to be used for updating a first model from among a plurality of second cohorts based on the first data distribution acquired by the acquisition unit, and an update unit that updates the first model based on at least a portion of the second dataset from a dataset based on the selected second cohort.

[0097] Although some embodiments have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope of the invention and its equivalents as described in the claims, as well as in the scope and spirit of the invention. [Explanation of symbols]

[0098] 100A…Target site 100B…Candidate site 100C...Other sites 100D…Managed site 110,210…Communication interface 120,220…Input interface 130, 230... Processing circuit 131...Collection function 132…Data distribution calculation function 133…Update request function 134…Reception facility 135…Extraction function 136…Update function 137…Verification function 138…Transmission function 140,240…Memory 141…Medical information DB 142…Model information DB 200: Central server 231…Acquisition function 232...Selection function 233…Management function 241…Distribution data DB 300...Selected sites HA...First large scale histogram HB: Second large scale histogram HC: Third large scale histogram HD: Small scale histogram

Claims

1. an acquisition unit that acquires a first data distribution for a first dataset from a dataset based on a first cohort; a selection unit that selects a second cohort to be used for updating a first model from among a plurality of second cohorts based on the first data distribution acquired by the acquisition unit; An update unit that updates the first model based on a second dataset among the datasets based on the selected second cohort. Learning system.

2. The selection unit selects the second cohort based on a similarity between the first data distribution and a second data distribution for the second data set based on each of the plurality of second cohorts. The learning system according to claim 1 .

3. An extracting unit is further configured to extract a data set to be used for updating the first model from the second data set based on the first data distribution. A learning system according to claim 1 or 2.

4. The amount of data in the dataset based on the second cohort is greater than the amount of data in the dataset based on the first cohort. A learning system according to claim 1 or 2.

5. Further comprising a verification unit that verifies suitability of the first model updated by the update unit for the first cohort. A learning system according to any one of claims 1 to 4.

6. Multiple sites collect cohort-based datasets and run trained models; a central server that obtains a data distribution of the data set collected by each of the multiple sites; the plurality of sites includes a first site and a second site; The second site is a data distribution calculation unit that calculates a first data distribution for a first data set among data sets based on a first cohort associated with the first site; and an update unit configured to update a first model utilized at the first site based at least in part on a second data set based on a second cohort associated with the second site; The central server comprises: an acquisition unit that acquires the first data distribution calculated by the data distribution calculation unit; A selection unit that selects the second cohort to be used for updating the first model from among a plurality of second cohorts based on the first data distribution acquired by the acquisition unit. Learning system.

7. The first site further includes a verification unit that verifies suitability of the first model updated by the update unit for the first cohort. The learning system according to claim 6.

8. The first site further includes an update request unit that requests an update of the first model. A learning system according to claim 6 or 7.

9. A learning device included in a second site, an extraction unit that extracts a dataset to be used for updating from a second dataset among data based on a second cohort associated with the second site, based on a first data distribution for a first dataset used for training a first model among datasets based on a first cohort calculated by a data distribution calculation unit included in the first site; an update unit that updates the first model based at least in part on a second data set based on a second cohort associated with the second site; A transmission unit that transmits the first model updated by the update unit to the first site. Learning device.

10. The computer obtaining a first data distribution for a first dataset, the first dataset being based on the first cohort; selecting a second cohort to be used for updating the first model from among the second cohorts based on the obtained first data distribution; updating the first model based at least in part on a second data set based on the selected second cohort; How to learn.

11. On the computer, obtaining a first data distribution for a first dataset, the first dataset being based on the first cohort; selecting a second cohort to be used for updating the first model from among the second cohorts based on the acquired first data distribution; updating the first model based at least in part on a second data set based on the selected second cohort; program.

Images