Method for supporting development of medical device, method for machine learning, medical device development supporting device, and computer program

Abstract

To provide a method for supporting development of a medical device which enables grasping the combination of physical characteristics of a medical device according to a treatment site to treat and the treatment record of a lesion.SOLUTION: The method for supporting development of a medical device includes: acquiring physical characteristic data showing plural physical characteristics of a medical device used for a surgery, patient data including data showing a treatment site of a patient on which a surgery was performed by using the medical device and a lesion, and treatment record data showing the treatment record using the medical device; machine-learning the relation among the acquired physical characteristic data, patient data, and treatment history data; and presenting plural pieces of physical characteristic data of the medical device suitable for a specific treatment site and a specific lesion on the basis of the result of machine learning.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a medical instrument development support method, a machine learning method, a medical instrument development support device, and a computer program. [Background technology]

[0002] 2. Description of the Related Art Intravascular treatments using medical instruments such as guidewires, catheters, balloon catheters, and stents are being performed as minimally invasive treatments for lesions such as stenosis and blockages that occur in blood vessels. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2019 / 106882 Summary of the Invention [Problem to be solved by the invention]

[0004] In the development of medical devices for intravascular treatment, the performance of the developed devices is evaluated using a vascular model that mimics the treatment site and lesion. However, it is difficult to quantitatively estimate the treatment outcome when the developed devices are used on patients from the results of evaluation using the vascular model. In particular, it is difficult to grasp the correlation between the various physical properties of the devices, such as surface slipperiness, tip load, and size, and the treatment outcome when the devices are used on patients.

[0005] An object of the present disclosure is to provide a medical instrument development support method, a machine learning method, a medical instrument development support device, and a computer program that enable understanding of physical property data of a medical instrument according to the treatment site and treatment outcome of the lesion to be treated. [Means for solving the problem]

[0006] The present invention provides a medical instrument development support method, which (1) prepares a learning model that uses machine learning to identify the relationship between physical property data indicating multiple physical properties of a medical instrument used in surgery, patient data including data indicating the treatment site and lesion of a patient who underwent surgery using the medical instrument, and treatment outcome data indicating the treatment outcome using the medical instrument, and uses the learning model to present the multiple physical property data of the medical instrument suitable for a specific treatment site and lesion.

[0007] (2) The above-mentioned (1) medical instrument development support method preferably includes acquiring physical property data indicating a plurality of physical properties of a medical instrument to be used in surgery, patient data including data indicating a treatment site and a lesion of a patient on whom surgery was performed using the medical instrument, and treatment outcome data indicating the outcome of treatment using the medical instrument, and generating the learning model by machine learning based on the acquired physical property data, the patient data, and the treatment outcome data. (3) In the medical device development support method of (1) or (2) above, it is preferable to use the learning model to present data indicating treatment sites and lesions suitable for the medical device having a specific number of physical properties. (4) The medical instrument development support method described in any one of (1) to (3) above preferably includes the steps of: setting a treatment area and a lesion for which the medical instrument to be developed is to be targeted; presenting multiple physical property data of the medical instrument suitable for the set treatment area and lesion using the learning model; prototyping the medical instrument based on the presented multiple physical property data of the medical instrument; and presenting treatment areas and lesions suitable for the medical instrument having the multiple physical properties of the prototype medical instrument using the learning model. (5) The medical instrument development support method described in any one of (1) to (4) above preferably involves machine learning of the relationship between physical property data indicating multiple physical properties of each of multiple types of the medical instruments to be used in combination, patient data, and treatment outcome data, and using the learning model, presenting multiple physical property data of each of the multiple types of the medical instruments suitable for a specific treatment site and lesion. (6) In the medical instrument development support method described in any one of (1) to (5) above, it is preferable to present a plurality of physical property data of the medical instrument that satisfy a constraint condition that constrains a relationship between the plurality of physical property data. (7) In the medical instrument development support method according to any one of (1) to (6) above, it is preferable that the patient data includes data indicating a treatment site of a coronary artery and a coronary artery lesion. (8) In the medical device development support method described in any one of (1) to (7) above, it is preferable that the medical device includes a catheter, and the physical property data of the catheter includes data indicating surface slipperiness, tip load, outer diameter, inner diameter, length, stiffness of the base end, or tensile strength. (9) In the medical device development support method described in any one of (1) to (8) above, it is preferable that the medical device includes a guidewire, and the physical property data of the guidewire includes data indicating surface slipperiness, tip load, tip shape, torque transmissibility, stiffness of the base end, or tensile strength. (10) In the medical device development support method described in any one of (1) to (9) above, it is preferable that the medical device includes a balloon catheter, and the physical property data of the balloon catheter includes data indicating a balloon size, a pressure resistance, or an expansion pressure. (11) In the medical device development support method described in any one of (1) to (10) above, it is preferable that the medical device includes a stent, and the physical property data of the stent includes data indicating a stent diameter, a shaft diameter of a delivery portion, stent flexibility, stent retention strength, or an expansion diameter.

[0008] The present invention provides (12) a machine learning method for acquiring physical property data indicating multiple physical properties of a medical instrument used in surgery, patient data including data indicating a treatment site and a lesion of a patient on whom surgery was performed using the medical instrument, and treatment outcome data indicating the outcome of treatment using the medical instrument, and generating a learning model by machine learning based on the acquired physical property data, the patient data, and the treatment outcome data.

[0009] The present invention provides (13) a medical instrument development support device comprising: a learning model that has been machine-learned to learn the relationship between physical property data indicating a plurality of physical properties of a medical instrument used in surgery, patient data including data indicating a treatment site and a lesion of a patient on whom surgery was performed using the medical instrument, and treatment outcome data indicating the treatment outcome using the medical instrument; and a presentation unit that uses the learning model to present the plurality of physical property data of the medical instrument suitable for a specific treatment site and lesion.

[0010] The present invention also provides (14) a computer program for causing a computer to execute a process of reading a learning model that has been machine-learned to relate the relationship between physical property data indicating a plurality of physical properties of a medical instrument used in surgery, patient data including data indicating a treatment site and a lesion of a patient on whom surgery was performed using the medical instrument, and treatment outcome data indicating the treatment outcome using the medical instrument, and presenting the plurality of physical property data of the medical instrument suitable for a specific treatment site and lesion using the learning model. Effect of the Invention

[0011] According to the present disclosure, it is possible to obtain physical property data of a medical instrument according to the treatment site and treatment outcome of a lesion to be treated. [Brief description of the drawings]

[0012] [Figure 1] 1 is a block diagram showing an example of the configuration of a medical instrument development support device according to a first embodiment. [Diagram 2] 4 is a flowchart showing a processing procedure of the machine learning method according to the first embodiment. [Diagram 3] 3 is a flowchart showing the processing steps of a medical instrument development support method according to the first embodiment. [Figure 4] FIG. 2 is a schematic diagram showing an optimum property display screen 4 on which the estimation results of optimum physical property data of a medical instrument are displayed. [Diagram 5] FIG. 13 is a schematic diagram showing a suitable treatment site, etc. display screen displaying the estimation results of data indicating suitable treatment sites and lesions for a medical instrument having specific physical properties. [Figure 6] FIG. 11 is a block diagram showing an example of the configuration of a first learning model according to the second embodiment. [Figure 7] A block diagram showing an example of the configuration of a second learning model in embodiment 2. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] A medical instrument development support method, a machine learning method, a medical instrument development support device, and a computer program according to embodiments of the present disclosure will be described below with reference to the drawings. Note that the present disclosure is not limited to these examples, but is indicated by the claims, and is intended to include all modifications within the meaning and scope equivalent to the claims. In addition, at least a part of the embodiments described below may be arbitrarily combined.

[0014] (Embodiment 1) 1 is a block diagram showing an example of the configuration of a medical instrument development support device 1 according to embodiment 1. The medical instrument development support device 1 is a computer, and includes a processing unit 11, a storage unit 12, a display unit (presentation unit) 13, an operation unit 14, and a communication unit (acquisition unit) 15. Each unit is connected via a bus. Note that the medical instrument development support device 1 may be a multi-computer consisting of multiple computers, or may be a virtual machine virtually constructed by software.

[0015] The processing unit 11 is an arithmetic processing device or processor having one or more central processing units (CPUs), micro-processing units (MPUs), graphics processing units (GPUs), general-purpose computing on graphics processing units (GPGPUs), tensor processing units (TPUs), etc. The processing unit 11 reads out and executes a computer program P (program product) stored in the storage unit 12, thereby implementing the medical instrument development support method according to the first embodiment.

[0016] The memory unit 12 has, for example, a main memory unit and an auxiliary memory unit. The main memory unit is a temporary storage area such as a static random access memory (SRAM), a dynamic random access memory (DRAM), or a flash memory, and temporarily stores data required for the processing unit 11 to execute arithmetic processing. The auxiliary memory unit is a storage device such as a hard disk or an electrically erasable programmable ROM (EEPROM). The auxiliary memory unit stores the computer program P executed by the processing unit 11, the learning model 2, and various data required for other processing. Details of the computer program P and the learning model 2 will be described later.

[0017] The auxiliary storage unit may be an external storage device connected to the medical instrument development support device 1. The computer program P and the learning model 2 may be written to the auxiliary storage unit during the manufacturing stage of the medical instrument development support device 1, or the medical instrument development support device 1 may acquire the computer program P and the learning model 2 through communication and store them in the auxiliary storage unit after distributing them from the external medical instrument development support device 1. The computer program P and the learning model 2 may be recorded in a readable manner on a recording medium 1a such as a magnetic disk, an optical disk, or a semiconductor memory, or may be read from the recording medium 1a by a reading device and stored in the auxiliary storage unit.

[0018] The display unit 13 is, for example, a display device such as a liquid crystal panel or an organic EL (Electro Luminescence) display.

[0019] The operation unit 14 is an input device such as a hardware keyboard, a mouse, a touch panel, etc. A user can operate the medical instrument development support device 1 using the operation unit 14 to input information required for designing a medical instrument and to display physical property data of the medical instrument, etc.

[0020] The communication unit 15 includes a communication circuit for performing communication processing with an external device such as the database server 3. The communication unit 15 of the present embodiment 1 accesses the database server 3 and acquires learning data for generating the learning model 2 according to the present embodiment 1.

[0021] The database server 3 stores, as learning data, data that associates multiple physical property data of the medical instruments used in surgery, patient data including data indicating the treatment site and lesion of the patient on whom surgery was performed using the medical instruments, and treatment outcome data indicating the treatment outcome using the medical instruments.

[0022] The medical device is, for example, a catheter, a guidewire, a balloon catheter, a stent, etc. The physical property data of the catheter is, for example, data indicating surface slipperiness, tip load, outer diameter, inner diameter, length, stiffness of the base end, tensile strength, etc. The physical property data of the guidewire is, for example, data indicating surface slipperiness, tip load, tip shape, torque transmission, stiffness of the base end, tensile strength, etc. The physical property data of the balloon catheter is, for example, data indicating balloon size (length, diameter), pressure resistance, expansion pressure, etc. The physical property data of the stent is, for example, data indicating stent diameter, shaft diameter of the delivery part, stent flexibility, stent retention strength, expansion diameter, etc. The physical property data of the above-mentioned medical device is data obtained by ISO or ASTM or other arbitrarily set test method.

[0023] Among the physical properties related to the above physical property data, the catheter tip load refers to the load that the sheet receives when the tip of the catheter is hung vertically and pressed against the sheet perpendicularly. The tip load of the guidewire has the same meaning. The rigidity of the proximal end refers to the rigidity of the catheter or guidewire in the vicinity of the portion that is gripped by the surgeon. Generally, catheters and guidewires have low rigidity at the distal end and high rigidity at the proximal end. Torque transmission refers to the ability of the guidewire to transmit torque applied by the operator to the proximal end of the guidewire to the distal end of the guidewire.

[0024] The patient data includes at least data indicating the treatment site and the lesion. The data indicating the treatment site and the lesion is, for example, information on the treatment site and the coronary artery lesion of the coronary artery represented by the coronary artery number. Coronary artery lesions include, for example, chronic total occlusion, calcified lesions, bifurcation lesions, etc. The coronary artery lesions also include information on the blood vessel diameter. Furthermore, the patient data preferably includes data indicating the age, sex, surgical history, health condition, and the condition of the affected area of ​​the surgical target. Furthermore, the patient data preferably includes a medical history at the hospital, family composition, and the results of the health check. The results of the health check include, for example, information such as height, weight, and blood pressure.

[0025] The treatment outcome data is data indicating whether or not a surgery performed using a medical instrument was successful. The treatment outcome data includes, for example, information such as treatment success rates obtained from literature, the number of incidents reported to the government, and treatment success rates accumulated at hospitals.

[0026] 2 is a flowchart showing a processing procedure of the machine learning method according to the first embodiment. The processing unit 11 of the instrument development support device 1 acquires, as learning data, physical property data of the medical instrument, patient data, and treatment outcome data (step S11) from the database server 3. Then, the processing unit 11 performs machine learning on the relationship between the acquired physical property data, patient data, and treatment outcome data (step S12).

[0027] For example, the processing unit 11 learns the relationship between the physical characteristic data, the patient data, and the treatment outcome data by using a multiple regression model as the learning model 2. The multiple regression model is expressed by, for example, the following formula (1). S = Σai × xi + Σbi × yi + c … (1) however, S: Treatment outcome data xi: Patient data (including at least data showing the treatment site and lesion) yi: Physical property data for each of one or more medical devices ai,bi,c: weighting coefficients

[0028] The processing unit 11 generates the learning model 2 by optimizing the weighting coefficients so as to minimize an objective function such as a square sum error. The memory unit 12 stores data such as the weighting coefficients of the function expressed by the above formula as the learning model 2. By using the learning model 2 stored in the memory unit 12, the processing unit 11 becomes able to specify physical property data of a medical instrument that is optimal for a specific treatment site and lesion, or a combination of multiple physical property data. Furthermore, by using the learning model 2, the processing unit 11 becomes able to specify a treatment site and lesion for which a medical instrument having specific physical properties is suitable.

[0029] Here, a multiple regression model has been described as an example of learning model 2, but the relationship between the physical property data, patient data, and treatment outcome data may be learned by machine learning using a model configured as a neural network, a SVM (Support Vector Machine), a Bayesian network, or a regression tree, etc.

[0030] For convenience of explanation, an example has been described in which the medical instrument development support device 1 performs machine learning or generation of the learning model 2, but the learning process of the learning model 2 may be executed using another computer.

[0031] Furthermore, a learning model 2 may be prepared for each of multiple types of medical instruments, such as guidewires, catheters, balloon catheters, and stents, or one learning model 2 may be configured to have variables related to the physical property data of each of the multiple types of medical instruments.

[0032] 3 is a flowchart showing a processing procedure of the medical instrument development support method according to the first embodiment. The processing unit 11 of the medical instrument development support device 1 sets data indicating a treatment site and a lesion, etc., for which the medical instrument to be developed is to be a target (step S31). Specifically, a user operates the operation unit 14 to input data indicating a treatment site and a lesion, etc., for which the medical instrument to be developed is to be a target, as well as other patient data, to the medical instrument development support device 1. The processing unit 11 temporarily stores the data indicating the treatment site and the lesion, etc., input by the operation unit 14 in the storage unit 12.

[0033] Next, the processing unit 11 identifies and presents the physical properties and physical property data of the medical instrument suitable for the treatment area and lesion indicated by the set data (step S32). Specifically, the processing unit 11 uses the data indicating the set treatment area, lesion, etc., and the learning model 2 to determine a plurality of physical property data that maximizes the treatment outcome data. For example, the processing unit 11 may identify a combination of a plurality of physical property data that results in a differential value of the learning model 2 for each of the plurality of physical properties being zero. The processing unit 11 then presents the identified combination of a plurality of physical property data that maximizes the treatment outcome data. The combination of the plurality of pieces of physical property data is displayed on the display unit 13. In addition, when there are multiple points where the treatment outcome data is maximized, that is, when there are multiple candidate combinations of multiple physical property data, the processing unit 11 may be configured to rank these combinations of physical property data according to the treatment outcome data and present them. Furthermore, the processing unit 11 may identify and present optimal physical property data of a medical instrument for each attribute such as age, sex, etc. Specifically, the processing unit 11 may change the values ​​of the variable indicating age and the variable indicating sex included in the above formula (1) to identify the physical property data of a medical instrument, and present the identified physical property data.

[0034] 4 is a schematic diagram showing an optimum characteristic display screen 4 displaying the estimation result of the optimum physical characteristic data of the medical instrument. The optimum characteristic display screen 4 includes a lesion etc. display section 41 and an optimum specific display section 42. The lesion etc. display section 41 is a section that displays data indicating the treatment site and the lesion set in step S31. The optimum specific display section 42 is a section that displays the physical characteristic data of the medical instrument specified in step S32. The configuration of the optimum characteristic display screen 4 shown in Fig. 4 is one example, and the method of displaying the information specified in step S32 is not particularly limited. Data indicating the treatment area and the lesion can be displayed by characters, figures, symbols, numbers, or other methods. Similarly, physical characteristic data can be displayed by characters, figures, symbols, numbers, or other methods. Also, although an example of displaying the information specified in step S32 on the display unit 13 has been described, it may be configured to be output as a data file.

[0035] Next, the user prototypes the medical instrument by referring to the physical property data presented by the medical instrument development support device 1 (step S33). For example, the user prototypes the medical instrument by referring to the presented physical property data, and checks its performance using a blood vessel model that simulates a treatment site and a lesion. Based on the evaluation results of the blood vessel model, the user corrects or changes the physical properties of the medical instrument to be prototyped next. In this way, the user creates a prototype of the medical instrument. The physical property data indicated by the physical properties of the prototype is usually different from that indicated by the physical property data presented in step S32.

[0036] Next, the processing unit 11 of the medical instrument development support device 1 acquires physical property data of the prototype medical instrument designed in step S33 (step S34). Specifically, the user inputs a plurality of pieces of physical property data of the designed medical instrument by operating the operation unit 14. The processing unit 11 acquires the plurality of pieces of physical property data input via the operation unit 14. The processing unit 11 may be configured to acquire the physical property data of the prototype detected by using a sensor.

[0037] Next, the processing unit 11 of the medical instrument development support device 1 identifies a treatment site and a lesion suitable for the designed medical instrument, and presents data indicating the identified treatment site and the lesion (step S35). Specifically, the processing unit 11 estimates treatment outcome data by substituting the multiple physical property data acquired in step S34 and patient data including data indicating any treatment site and lesion into the learning model 2.

[0038] Since the expected treatment sites and lesions are known, the combinations are also known, and the number of combinations is finite. For example, if there are 15 treatment sites and three types of lesions, the number of combinations is 45. The processing unit 11 estimates treatment outcome data for each of the multiple combinations of treatment sites and lesions using the learning model 2, and displays data indicating the treatment site and lesion with the highest treatment outcome on the display unit 13. The processing unit 11 also displays a list of combinations of data indicating treatment sites and lesions in a ranking format in order of highest treatment outcome.

[0039] FIG. 5 is a schematic diagram showing a suitable treatment site display screen 5 that displays the estimated results of data indicating suitable treatment sites and lesions for medical instruments having specific physical properties. The suitable treatment site display screen 5 includes a physical property data display section 51 and a suitable treatment site display section 52. The physical property data display section 51 is a section that displays the physical property data acquired in step S34. The suitable treatment site display section 52 is a section that displays data indicating suitable treatment sites and lesions identified in step S35. The suitable treatment site display section 52 includes a list that displays combinations of data indicating treatment sites and lesions in order of high treatment results. The configuration of the suitable treatment site display screen 5 shown in FIG. 5 is one example, and the display method of the information identified in step S35 is not particularly limited. The data indicating suitable treatment sites and lesions can be displayed by characters, figures, symbols, numbers, or other methods. In addition, an example of displaying the information identified in step S35 on the display section 13 has been described, but it may be configured to output it as a data file.

[0040] Next, the user evaluates the validity of the physical characteristics of the medical device designed in step S33 by comparing the data showing the treatment site and lesion targeted by the medical device to be developed with the results of the data showing the treatment site and lesion presented in step S35 (step S36).The user adjusts the physical characteristics of the medical device by repeating the processes of steps S33 to S36 as necessary.

[0041] According to the medical instrument development support device 1 configured in this manner, a user of the medical instrument development support device 1 can grasp a combination of physical property data of a medical instrument according to a treatment site and a treatment result of a lesion. This allows the user of the medical instrument development support device 1 to carry out design and development with the physical property data presented by the medical instrument development support device 1 in mind, thereby contributing to shortening development time and reducing research and development costs. In addition, the user can confirm whether the prototype is suitable for the treatment site and lesion for which the prototype is being developed by inputting the physical property data of the prototype into the medical instrument development support device 1. Furthermore, the user can understand data indicating the treatment site and lesion for which the prototype can provide the most value by inputting the physical property data of the prototype into the medical instrument development support device 1.

[0042] Furthermore, by inputting the physical property data of the medical device for sale into the medical device development support device 1, the user can specify data indicating treatment sites and lesions that are expected to produce good treatment results, which were not anticipated at the time of initial sale. This enables the user to promote sales and rebrand the medical device.

[0043] (Embodiment 2) The medical instrument development support device 1 according to the second embodiment differs from the first embodiment in that it includes a first learning model 21 and a second learning model 22 trained using a neural network model. Since the other configurations of the medical instrument development support device 1 are similar to those of the medical instrument development support device 1 according to the first embodiment, the same reference numerals are used for the similar parts and detailed description will be omitted.

[0044] FIG. 6 is a block diagram showing a configuration example of the first learning model 21 according to the second embodiment. The first learning model 21 is a trained model that has been trained by machine learning using teacher data to output an optimal combination of multiple physical property data of medical instruments when data indicating a treatment site and a lesion is input. The first learning model 21 may be configured to output the physical property data of one medical instrument, or may be configured to output the physical identification data of multiple medical instruments. Here, a description will be given assuming that the first learning model 21 is generated for each of multiple medical instruments.

[0045] The first learning model 21 performs a predetermined calculation on input values ​​and outputs the calculation results, and the memory unit 12 stores data such as coefficients of functions that define this calculation and thresholds as the first learning model 21.

[0046] The first learning model 21 is, for example, a neural network model having an input layer 21a to which data indicating the treatment area and the lesion is input, an intermediate layer 21b that extracts features of the data indicating the treatment area and the lesion, and an output layer 21c that outputs physical property data calculated based on the extracted features.

[0047] The input layer 21a of the neural network has a plurality of neurons (nodes) to which data on the treatment site and lesion is input, and passes each input data to the intermediate layer 21b.

[0048] The intermediate layer 21b has multiple layers each made up of multiple neurons (nodes). Each layer extracts features related to the physical properties of the medical instrument from the input data, and passes them on to the next layer in turn, until the last layer passes them on to the output layer 21c.

[0049] The output layer 21c includes a plurality of neurons (nodes) corresponding to a plurality of physical properties, and each neuron outputs physical property data.

[0050] A learning method of the first learning model 21 will be described. First, the medical instrument development support device 1 collects the physical property data of the medical instrument, the patient data, and the treatment outcome data as the teacher data. Then, the medical instrument development support device 1 generates learning data by adding the physical property data with good treatment outcome as the teacher data to the data indicating the treatment site and the lesion included in the patient data. Next, the medical instrument development support device 1 generates the first learning model 21 by performing machine learning or deep learning on the pre-learning neural network model using the generated learning data.

[0051] Specifically, the medical instrument development support device 1 inputs data indicating a treatment site and a lesion included in the learning data into a pre-learning neural network model, and acquires physical data output from the output layer 21c through arithmetic processing in the intermediate layer 21b. The medical instrument development support device 1 then compares the physical data output from the output layer 21c with physical property data indicated by the teacher data, and optimizes parameters used in the arithmetic processing in the intermediate layer 21b so that the physical property data output from the output layer 21c approaches the value of the teacher data. The parameters are, for example, weights (coupling coefficients) between neurons. There is no particular limitation on the method of optimizing the parameters, but for example, the medical instrument development support device 1 optimizes various parameters using the steepest descent method or the like.

[0052] The medical instrument development support device 1 obtains a trained first learning model 21 by repeatedly performing the above process using the learning data.

[0053] 7 is a block diagram showing a configuration example of the second learning model 22 according to embodiment 2. The second learning model 22 is a trained model that has been trained by machine learning using teacher data to output treatment outcome data for a combination of data indicating a plurality of treatment sites and lesions when a plurality of pieces of physical property data of a medical instrument are input.

[0054] The second learning model 22 performs a predetermined calculation on the input value and outputs the calculation result, and the memory unit 12 stores data such as the coefficients of the function that defines this calculation and threshold values ​​as the second learning model 22.

[0055] The second learning model 22 is, for example, an input model in which a plurality of physical property data of a medical instrument is input. The neural network model has an input layer 22a, an intermediate layer 22b that extracts features of the physical property data, and an output layer 22c that outputs treatment outcome data calculated based on the extracted features.

[0056] The input layer 22a of the neural network corresponds to each of the multiple physical property data of the medical instrument, has multiple neurons (nodes) to which the physical property data is input, and passes each input data to the intermediate layer 22b.

[0057] The intermediate layer 22b has multiple layers each made up of multiple neurons (nodes). Each layer extracts features related to treatment outcome data from the input data and passes them on to the next layer in sequence, and the last layer passes them on to the output layer 22c.

[0058] The output layer 22c has a plurality of neurons (nodes) corresponding to a combination of a plurality of treatment sites and lesions, and each neuron outputs treatment outcome data when a medical instrument having the input physical characteristics is used on the treatment site and lesion.

[0059] For convenience of explanation, an example has been described in which the medical instrument development support device 1 performs machine learning or generation of the first learning model 21 and the second learning model 22, but the machine learning may be performed using another computer. The trained second learning model 22 and data related to the second learning model 22 may be provided in a form of distribution via a communication network, similar to the computer program P, or may be provided in a form recorded on the recording medium 1a.

[0060] A learning method of the second learning model 22 will be described. First, the medical instrument development support device 1 collects the physical property data of the medical instrument, the patient data, and the treatment outcome data as the teacher data. Then, the medical instrument development support device 1 generates learning data by adding the treatment outcome data as teacher data to the physical property data. The learning data is prepared for each combination of treatment site and lesion. Next, the medical instrument development support device 1 generates the second learning model 22 by performing machine learning or deep learning on the pre-learning neural network model using the generated learning data.

[0061] Specifically, the medical instrument development support device 1 inputs the physical property data included in the learning data into the pre-learning neural network model, and obtains the treatment outcome data output from the output layer 22c through arithmetic processing in the intermediate layer 22b. The medical instrument development support device 1 then compares the treatment outcome data output from the output layer 22c with the treatment outcome data indicated by the teacher data, and optimizes the parameters used in the arithmetic processing in the intermediate layer 22b so that the treatment outcome data output from the output layer 22c approaches the values ​​of the teacher data. The parameters are, for example, weights (coupling coefficients) between neurons. The method of optimizing the parameters is not particularly limited, but for example, the medical instrument development support device 1 optimizes various parameters using the steepest descent method or the like.

[0062] The medical instrument development support device 1 obtains a trained second learning model 22 by repeatedly performing the above process using the learning data.

[0063] The processing unit 11 of the medical instrument development support device 1 can execute a calculation process to obtain optimal physical property data of the medical instrument from data indicating a treatment site and a lesion by reading the data stored as the first learning model 21. That is, the processing unit 11 can execute the process of step S32 according to the first embodiment by using the first learning model 21.

[0064] Similarly, the processing unit 11 of the medical instrument development support device 1 stores as a second learning model 22 By reading the data, it becomes possible to execute a calculation process for obtaining treatment outcome data for each combination of data indicating a plurality of treatment sites and lesions from the physical property data of the medical instrument. That is, the processing unit 11 can execute the process of step S35 according to the first embodiment by using the second learning model 22.

[0065] According to the medical instrument development support device 1 according to the second embodiment configured as described above, it is possible to grasp the combination of the physical properties of the medical instrument and the physical property data according to the treatment result of the treatment site and the lesion, as in the first embodiment. The medical instrument development support device 1 can accurately present the physical property data of the medical instrument optimal for the treatment site and the lesion by using the first learning model 21 and the second learning model 22 configured as neural network models. In addition, the medical instrument development support device 1 can present data indicating the treatment site and the lesion for which good treatment result can be expected by using the medical instrument having the specific physical property.

[0066] (Embodiment 3) The medical instrument development support device 1 according to the third embodiment differs from the second embodiment in that the first learning model 21 outputs physical property data for each of a plurality of types of medical instruments. Since the other configurations of the medical instrument development support device 1 are similar to those of the medical instrument development support device 1 according to the second embodiment, the same reference numerals are used for the similar parts and detailed description will be omitted.

[0067] The first learning model 21 according to the third embodiment is a model that has been machine-learned to determine the relationship between physical property data indicating multiple physical properties of each of multiple types of medical instruments used in combination, patient data, and treatment outcome data.

[0068] Each of the multiple types of medical instruments has multiple appropriate physical properties and physical property data. The output layer 21c of the first learning model 21 has neurons (nodes) corresponding to the multiple physical properties of each of the multiple types of medical instruments, and each node outputs the multiple physical property data. The configurations of the input layer 21a and the intermediate layer 21b are the same as those of the second embodiment.

[0069] A learning method of the first learning model 21 of the third embodiment will be described. As in the second embodiment, the medical instrument development support device 1 collects a plurality of physical property data of a plurality of medical instruments, patient data, and treatment outcome data. Then, the medical instrument development support device 1 generates learning data by adding a plurality of physical property data of a plurality of medical instruments, which are a combination that produces a good treatment outcome, as teacher data to data indicating a treatment site and a lesion included in the patient data. Next, the medical instrument development support device 1 generates the first learning model 21 by performing machine learning or deep learning on the pre-learning neural network model using the generated learning data.

[0070] According to the medical instrument development support device 1 of embodiment 3 configured in this manner, it is possible to identify and present multiple physical property data for multiple types of medical instruments that are optimal for the treatment site and lesion that are the subject of development.

[0071] (Embodiment 4) The medical instrument development support device 1 according to the fourth embodiment differs from the first to third embodiments in that it presents a plurality of pieces of physical property data of a medical instrument that satisfy a predetermined constraint condition that constrains the relationship between the plurality of pieces of physical property data. Since the other configurations of the medical instrument development support device 1 are similar to those of the medical instrument development support device 1 according to the first to third embodiments, the same reference numerals are used for the similar parts and detailed description will be omitted.

[0072] A first example of a constraint condition is information for restricting an upper limit, a lower limit, or both of predetermined physical property data possessed by a medical device.

[0073] A second example of a constraint condition is information for restricting the relationship between a plurality of predetermined physical property data of a medical device to a certain range. For example, there is a constraint that the stiffness of the distal end of the guidewire is lower than the stiffness of the proximal end. Therefore, the processing unit 11 restricts the physical property data of the guidewire to be presented such that the stiffness of the distal end is lower than the stiffness of other parts.

[0074] More specifically, for example, a guidewire for coronary arteries is designed so that the stiffness of the portion from the tip to about 400 mm is lower than the stiffness of the portion on the proximal side. Therefore, the medical instrument development support device 1 limits the output of the processing unit 11 so that the stiffness at about 400 mm from the tip is lower than the stiffness of the portion on the proximal side.

[0075] In a microcatheter for coronary arteries, the rigidity of the portion from the tip to about 300 mm is designed to be lower than the rigidity of other portions. Therefore, the medical instrument development support device 1 limits the output of the processing unit 11 so that the rigidity of the portion from the tip to about 300 mm is lower than the rigidity of the portion on the proximal side.

[0076] For example, the learning model 2 according to the first embodiment may be configured to have a weighting coefficient that makes the stiffness of the tip portion lower than the stiffness of other portions. In other words, the processing unit 11 may train the learning model 2 such that the objective function increases when the stiffness of the tip portion of the guidewire becomes higher than the stiffness of other portions. The learning models 2 of the second and third embodiments may be configured in a similar manner.

[0077] In addition, if the stiffness of the tip of the guidewire obtained using learning model 2 is higher than the stiffness of other parts, the processing unit 11 may correct the stiffness of the tip of the guidewire to be lower than the stiffness of other parts and display it.

[0078] Furthermore, the processing unit 11 may be configured to issue a warning when the stiffness of the tip portion of the guidewire obtained using the learning model 2 is higher than the stiffness of other portions.

[0079] A third example of the constraint condition is information for defining the relationship between a plurality of physical property data possessed by a plurality of medical instruments. In other words, the constraint condition can be said to be information for defining the relationship between the physical property data possessed by a first medical instrument and the physical property data possessed by a second medical instrument. For example, there is a constraint that the inner diameter of the catheter is smaller than the outer diameter of the guidewire. Therefore, the processing unit 11 may train the learning model 2 so that the inner diameter of the catheter is smaller than the outer diameter of the guidewire. The processing unit 11 may also be configured to correct the inner diameter of the catheter output from the learning model 2 to be smaller than the outer diameter of the guidewire and present it.

[0080] The medical instrument development support device 1 of embodiment 4 configured as described above can identify and present multiple physical property data for multiple types of medical instruments that are optimal and feasible for the treatment site and lesion that are the subject of development. [Explanation of symbols]

[0081] 1: Medical device development support device 1a: Recording medium 2: Learning model 3: Database server 4: Optimal characteristics display screen 5: Screen showing suitable treatment areas, etc. 11: Processing section 12: Storage section 13: Display section 14:Operation section 15: Communications Department 21: First learning model 22: Second Learning Model 41: Lesion display section 42: Optimal specific display section 51: Physical property data display section 52: Suitable treatment area, etc. display section P: Computer program

Claims

1. A learning model is prepared by machine learning the relationship between physical property data indicating a plurality of physical properties of a medical instrument used in a surgical operation, patient data including data indicating a treatment site and a lesion of a patient in whom a surgical operation was performed using the medical instrument, and treatment outcome data indicating the treatment outcome using the medical instrument; Using the learning model, a plurality of physical property data of the medical device suitable for a particular treatment site and lesion is presented. Methods for supporting medical device development.

2. Acquiring physical property data indicating a plurality of physical properties of a medical instrument used in a surgical operation, patient data including data indicating a treatment site and a lesion of a patient on whom a surgical operation was performed using the medical instrument, and treatment outcome data indicating a treatment outcome using the medical instrument; The learning model is generated by machine learning based on the acquired physical property data, the patient data, and the treatment outcome data. The medical instrument development support method according to claim 1.

3. Using the learning model to present data indicative of treatment sites and lesions suitable for use with the medical device having a particular plurality of physical characteristics. The medical instrument development support method according to claim 1.

4. A treatment site and a lesion to be treated by the medical device to be developed are set, Presenting a plurality of physical property data of the medical instrument suitable for the set treatment site and lesion using the learning model; Prototype the medical device based on the presented physical property data of the medical device; Using the learning model, data indicating treatment sites and lesions suitable for use with the medical device having a plurality of physical characteristics of the prototype medical device is presented. The medical instrument development support method according to claim 1.

5. Machine learning is performed to obtain a relationship between physical property data indicating a plurality of physical properties of each of the plurality of types of medical instruments used in combination, patient data, and treatment outcome data; Using the learning model, present a plurality of physical property data for each of a plurality of types of the medical instruments suitable for a particular treatment site and lesion. The medical instrument development support method according to claim 1.

6. Presenting a plurality of physical property data of the medical device that satisfy a constraint condition that constrains a relationship between the plurality of physical property data. The medical instrument development support method according to claim 1.

7. The patient data includes data indicative of a coronary artery treatment site and a coronary artery lesion. The medical instrument development support method according to claim 1.

8. the medical device comprises a catheter; The physical property data of the catheter includes data indicating surface slipperiness, tip load, outer diameter, inner diameter, length, stiffness of the proximal end, or tensile strength. The medical instrument development support method according to claim 1.

9. the medical device includes a guidewire; The physical property data of the guidewire includes surface slipperiness, tip load, tip shape, torque Includes data showing transmissibility, proximal stiffness, or tensile strength The medical instrument development support method according to claim 1.

10. the medical device includes a balloon catheter; The physical property data of the balloon catheter includes data indicative of a balloon size, a pressure resistance, or an expansion pressure. The medical instrument development support method according to claim 1.

11. the medical device comprises a stent; The physical property data of the stent includes data indicative of a stent diameter, a shaft diameter of a delivery portion, a stent flexibility, a stent retention strength, or an expansion diameter. The medical instrument development support method according to claim 1.

12. Acquiring physical property data indicating a plurality of physical properties of a medical instrument used in a surgical operation, patient data including data indicating a treatment site and a lesion of a patient on whom a surgical operation was performed using the medical instrument, and treatment outcome data indicating a treatment outcome using the medical instrument; A learning model is generated by machine learning based on the acquired physical property data, the patient data, and the treatment outcome data. Machine learning methods.

13. a learning model that performs machine learning on the relationship between physical property data indicating a plurality of physical properties of a medical instrument used in a surgery, patient data including data indicating a treatment site and a lesion of a patient who has undergone surgery using the medical instrument, and treatment outcome data indicating the treatment outcome using the medical instrument; a presentation unit that uses the learning model to present a plurality of physical property data of the medical instrument suitable for a specific treatment site and lesion; A medical instrument development support device equipped with:

14. A learning model is read out, which is a machine learning model of the relationship between physical property data indicating a plurality of physical properties of the medical instrument used in the surgery, patient data including data indicating a treatment site and a lesion of a patient in whom the surgery was performed using the medical instrument, and treatment outcome data indicating the treatment outcome using the medical instrument; Using the learning model, a plurality of physical property data of the medical device suitable for a particular treatment site and lesion is presented. A computer program for causing a computer to execute a process.

Images