Training method and training system

Abstract

The invention provides a training method and a training system. The training method comprises the following steps: performing tagging processing on historical data; obtaining a training evaluation model based on the historical data; wherein the training evaluation model is obtained according to at least two different modes based on different labels; obtaining an operation instruction; calculating to obtain a simulation training result based on the operation instruction; the simulation training result and a result obtained by operating the operation instruction in an actual operation environment have a corresponding relationship; and the simulation training result is input into the training evaluation model, and the training evaluation model obtains an evaluation result based on the label corresponding to the training label. Through personalized data cleaning, training evaluation models in different forms are established, so that the training evaluation models have higher pertinence, and the problems that in the prior art, pertinence is not high during data processing, and the scoring effect is not ideal enough are solved.

Description

Technical Field

[0001] The present invention relates to the field of medical devices, and in particular to a training method and a training system. Background Art

[0002] Doctors need to be trained before they perform clinical operations.

[0003] The training results of current surgical training devices are manually scored and evaluated by trainers, and some can only provide qualitative conclusions of pass / fail, which are not detailed enough. Therefore, the following defects are brought about: 1. The scoring results cannot reflect the core assessment standards and indicators that specific departments and hospitals actually pay attention to. 2. The existing scoring items are relatively fixed and cannot meet the differentiated training goals of different departments. 3. Qualitative evaluation cannot reflect the user's mastery of specific skills. 4. There is a lack of processing of the original training data, and it is impossible to carry out multi-dimensional surgical evaluation. 5. Manual scoring is inefficient, and the evaluation results are affected by subjective factors.

[0004] There are some quantitative scoring schemes in the prior art. For example, after cleaning all the original data, the HMM (Hidden Markov Model) method is used to evaluate the quantitative results. However, this scheme does not perform personalized cleaning of the original data, the generated model is not very targeted, and the final scoring effect is not ideal.

[0005] That is, the existing technology has the problem that the data processing is not targeted enough and the scoring effect is not ideal. Summary of the invention

[0006] The purpose of the present invention is to provide a training method and a training system to solve the problems in the prior art of lack of pertinence in data processing and unsatisfactory scoring effect.

[0007] In order to solve the above technical problems, the present invention provides a training method, including: labeling historical data; obtaining a training evaluation model based on the historical data; wherein, based on different labels, the training evaluation model is obtained in at least two different ways; obtaining training labels; obtaining operation instructions; calculating simulation training results based on the operation instructions; the simulation training results have a corresponding relationship with the results obtained by running the operation instructions in an actual operating environment; and, the simulation training results are input into the training evaluation model to obtain an evaluation result.

[0008] Optionally, the types of labels include: at least one of: department, surgical procedure, training goal and scoring item.

[0009] Optionally, the training objectives include: at least one of basic training, incision selection, tissue separation, suturing and knotting, bleeding control and time control.

[0010] And / or, the scoring items include: at least one of: visual field, incision positioning, tissue resection, time consumption, suturing efficiency and bleeding volume.

[0011] Optionally, the step of obtaining a training evaluation model based on historical data includes: missing data processing; data format processing; logical error cleaning; removal of useless data; and correlation verification.

[0012] Optionally, the step of obtaining a training evaluation model based on historical data includes: randomly dividing the historical data into training data and verification data according to an initial ratio; training a prediction model using the training data as input; for the prediction model that has completed training, using the verification data as input, performing model verification and calculating the accuracy; if the accuracy is lower than the first accuracy, randomly dividing the historical data into the training data and the verification data again according to the initial ratio and executing subsequent steps; if the accuracy is between the first accuracy and the second accuracy, retraining the prediction model using the training data as input and executing subsequent steps; if the accuracy is higher than the second accuracy, setting the prediction model as the training evaluation model.

[0013] Optionally, when the type of the label is a scoring item, the step of training the prediction model with the training data as input includes: for each scoring item, executing the following steps: determining a scoring standard set, the scoring standard set including at least two scoring standards; determining a probability function for obtaining different outputs from different inputs under each scoring standard based on the training data; determining a reward function; a total reward calculated based on the scoring standard set, the probability function and the reward function; maximizing the calculated total reward by continuously modifying the scoring standard set; and setting the scoring standard corresponding to the maximum total reward as the training result.

[0014] Optionally, when the label is time control, the steps of training the prediction model with the training data as input include: dividing the operation process into multiple stages; setting a stage coefficient corresponding to each stage; calculating the time evaluation value based on the time consumption of the stage and the stage coefficient, and calculating the scoring result in combination with the time evaluation value; by continuously modifying the stage coefficient, the error between the calculated scoring result and the scoring result in the training data reaches a minimum value; the stage coefficient corresponding to the minimum error is set as the training result.

[0015] Optionally, the training method includes: collecting new data during the execution of the training method, and updating the training evaluation model based on the new data; the process of collecting the new data includes: collecting time information and storing it as a timestamp and UID, collecting image information and storing the storage path of the image information; collecting the training plan and storing it as training information; collecting the correction value of the evaluation result input by the outside and storing it as the actual training result; collecting custom input and storing it as supplementary information.

[0016] Optionally, the step of obtaining a training evaluation model based on historical data includes: extracting scoring items and weights of the scoring items based on departments, training objectives, scores and / or grades.

[0017] The training method includes: automatically setting the training target based on the department and the training evaluation model and / or modifying the training target based on external input; automatically setting scoring items and scoring item weights based on the training target and the training evaluation model and / or modifying the scoring items and scoring item weights based on external input.

[0018] The step of inputting the simulation training results into the training evaluation model to obtain the evaluation results includes: calculating the values ​​of the scoring items based on the simulation training results, and calculating the scoring results in combination with the weights of the scoring items.

[0019] Optionally, the step of obtaining a training evaluation model based on historical data includes: obtaining the image features of the evaluation results based on the scoring items; the step of calculating the scoring items based on the simulation training results includes: obtaining the image features based on the simulation training results; and, calculating the numerical values ​​of the scoring items based on the image features.

[0020] The image features include: tissue edge features, suturing features and field of view features.

[0021] Optionally, the training method includes: obtaining a training scheme model based on the historical data, the training scheme model including training state results and corresponding probabilities obtained under different training schemes; and selecting the training scheme with the best expected training state results as the training course based on the trainee's current training state in combination with the training scheme model.

[0022] Optionally, the training method includes: obtaining the training labels and scoring item weight information based on an interactive interface, the training labels including: department information, training target information and scoring item information; generating training content based on the department information and training target information; and obtaining the evaluation results based on the simulation training results, scoring item information and scoring item weight information.

[0023] In order to solve the above technical problems, the present invention also provides a training system, including a human-computer interaction module, a result evaluation module, a surgery training module and a training robot subsystem.

[0024] Wherein, the result evaluation module is used to obtain a training evaluation model based on historical data; wherein, the historical data is labeled; and the result evaluation module obtains the training evaluation model in at least two different ways based on different labels.

[0025] The training robot subsystem is used to obtain the operation instruction. The human-computer interaction module is used to obtain the training label; the training label has a one-to-one correspondence with the label.

[0026] The surgical training module is used to calculate a simulation training result based on the operation instruction; the simulation training result has a corresponding relationship with a result obtained by running the operation instruction in an actual operation environment.

[0027] The result evaluation module is also used to obtain an evaluation result based on the simulation training result and the label corresponding to the training label.

[0028] Optionally, the training system also includes a data module and a human-computer interaction module; wherein the data module is used to collect new data during the operation of the training system; the result evaluation module is also used to update the training evaluation model based on the new data; the surgical training module is also used to generate the virtual image required by the training robot subsystem; the training robot subsystem includes a main control terminal and a simulation result display at the physical level, the main control terminal is used to obtain the operation instructions, and the simulation result display is used to display the virtual image; the human-computer interaction module is used to obtain department information, training target information, scoring item information and scoring item weight information; the human-computer interaction module is also used to output the evaluation result.

[0029] Compared with the prior art, the present invention provides a training method and a training system, wherein the training method includes: labeling historical data; obtaining a training evaluation model based on the historical data; wherein the training evaluation model is obtained in at least two different ways based on different labels; obtaining a training label; the training label has a one-to-one correspondence with the label; obtaining an operation instruction; calculating a simulation training result based on the operation instruction; the simulation training result has a correspondence with the result obtained by running the operation instruction in an actual operation environment; and the simulation training result is input into the training evaluation model, and the training evaluation model obtains an evaluation result based on the label corresponding to the training label. Through personalized data cleaning, different forms of training evaluation models are established, making the training evaluation model more targeted, solving the problems of weak targeting and unsatisfactory scoring effect in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Those skilled in the art will appreciate that the accompanying drawings are provided for a better understanding of the present invention and do not constitute any limitation on the scope of the present invention.

[0031] Figure 1 It is a flowchart of a training method according to an embodiment of the present invention.

[0032] Figure 2 Schematic diagram of a training system according to an embodiment of the present invention.

[0033] Figure 3 It is a flowchart of the steps of collecting new data according to an embodiment of the present invention.

[0034] Figure 4 It is a flowchart of obtaining a training evaluation model according to an embodiment of the present invention.

[0035] Figure 5 It is a schematic diagram of a prediction model based on tissue resection effect training in one embodiment of the present invention.

[0036] Figure 6 4 is a schematic diagram of surgical time evaluation according to an embodiment of the present invention.

[0037] Figure 7 It is a schematic diagram of determining scoring items according to an embodiment of the present invention.

[0038] Figure 8 is a schematic diagram of image features according to an embodiment of the present invention.

[0039] Fig. 9 Schematic diagram of a training plan according to an embodiment of the present invention.

[0040] Fig.10It is a flowchart of a training method according to an embodiment of the present invention.

[0041] Fig.11a This is a function menu interface according to an embodiment of the present invention.

[0042] Fig.11b This is a course operation visual interface according to an embodiment of the present invention.

[0043] Fig.11c This is a goal setting interactive interface according to an embodiment of the present invention.

[0044] Fig.11d This is a training result report interface according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, advantages and features of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the drawings are all in a very simplified form and are not drawn to scale, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In addition, the structure shown in the drawings is often a part of the actual structure. In particular, the emphasis of each drawing is different, and sometimes different scales are used.

[0046] The core idea of ​​the present invention is to provide a training method and a training system to solve the problem in the prior art that the training results need to be manually scored and evaluated by the trainer.

[0047] The following description is given with reference to the accompanying drawings.

[0048] In order to solve the above technical problems, the present embodiment provides a training method, including: S10, labeling historical data. S20, obtaining a training evaluation model based on historical data; wherein the training evaluation model is obtained in at least two different ways based on different labels. S30, obtaining training labels; the training labels and labels have a one-to-one correspondence. S40, obtaining operation instructions. S50, calculating a simulation training result based on the operation instructions. The simulation training result has a correspondence with the result obtained by running the operation instruction in the actual operation environment. And, S60, the simulation training result is input into the training evaluation model, and the training evaluation model obtains an evaluation result based on the label corresponding to the training label, and the evaluation result includes a quantitative scoring result.

[0049] Among them, "wherein based on different labels, the training evaluation model is obtained in at least two different ways" should be understood as, if the labels are different, the training evaluation models obtained can be the same or different, but overall, there are at least two different ways to obtain the training evaluation model, and these different ways are caused by the difference in labels. For example, it will be introduced later that when the label is time control, the structure of the training evaluation model is different from other training evaluation models. "Training evaluation model" as a whole should be understood as a model with training function and evaluation function, and not as a phrase with a subject-verb structure: "obtaining an evaluation model in a training manner".

[0050] At the same time, this embodiment also provides a training system, including a result evaluation module, a surgical training module, a training robot subsystem, a data module and a human-computer interaction module.

[0051] The result evaluation module is used to obtain a training evaluation model based on historical data; and to obtain an evaluation result based on a simulated training result, wherein the evaluation result includes a quantitative scoring result.

[0052] The training robot subsystem is used to obtain operation instructions.

[0053] The surgical training module is used to calculate the simulation training results based on the operation instructions; the simulation training results correspond to the results obtained by running the operation instructions in the actual operation environment. The surgical training module is also used to generate virtual images required for training the robot subsystem.

[0054] The training robot subsystem includes a physical main control terminal and a simulation result display. The main control terminal is used to obtain operation instructions, and the simulation result display is used to display virtual images.

[0055] The data module is used to collect new data during the operation of the training system.

[0056] The result evaluation module is also used to update the training evaluation model based on new data.

[0057] The human-computer interaction module is used to obtain department information, training target information, scoring item information and scoring item weight information; wherein department information, training target information and scoring item information are all training labels; that is, the human-computer interaction module is used to obtain training labels. The human-computer interaction module is also used to output evaluation results. In one embodiment, the human-computer interaction system is a computer, tablet or smart phone with a display function, and is connected to other modules for communication.

[0058] In other embodiments, Figure 2As shown, the appearance of the training system is the same as or similar to the main operation terminal of the surgical robot system, so that trainees can adapt to the real surgical operation environment after the training. The computing components of the human-computer interaction system are arranged inside the training system, and the interactive information display 102 of the human-computer interaction system is arranged at a suitable position of the training system. Figure 2 Also shown are a simulation result display 101 and a main control terminal 103.

[0059] exist Figure 2 In the training system shown, the trainee's operation process is as follows: slide the interactive information display 102 screen by hand to interact with information, input or select parameters required for training (parameters such as department information, training target information, scoring item information and scoring item weight information), some of these parameters are training labels, and some are not. The surgical training module 104 generates a virtual training project based on the parameters input by the trainee, and generates a virtual image at the simulation result display 101. The trainee moves his head to the simulation result display 101 so that the eye area fits the simulation result display 101, and observes the virtual image in the simulation result display 101, while holding the end of the main control terminal 103 to operate, the main control terminal 103 moves under the traction of the trainee's hand, and collects the kinematic parameters of its own joints, which are the operation instructions. The virtual image in the simulation result display 101 changes at the same time (for example, the trainee simulates operating a scalpel, and the tissue displayed in the virtual image is separated along the trajectory of the scalpel movement). When the trainee's operation is completed, the simulation training results are displayed on the simulation result display 101, and the evaluation results are displayed on the interactive information display 102. The evaluation results may include only the scoring results, or may include multi-dimensional evaluation result charts, such as data radar charts, etc.

[0060] It is understandable that the above process is only used to illustrate the workflow of the training system and does not impose any limitation on the scheme of the present invention. For example, the parameters required for training may not necessarily be input through a screen, but may also be input through a keyboard, mouse, microphone or other input devices; the object of inputting parameters may not necessarily be a trainee, but may also be a trainer or a control program at a higher level. In different embodiments, the simulation result display 101 and the interactive information display 102 may be different physical objects, or may be the same display, but the simulation result display function and the interactive information display function are implemented at different times or in different display areas.

[0061] The surgical training module includes an instruction processing submodule and a graphic processing submodule.

[0062] The instruction processing submodule, after receiving the training instructions set by the trainer and the operation instructions of the training robot subsystem, processes and obtains the key parameters for constructing the simulation training environment, such as scene information, surgical instrument posture information, field of view information, etc. The instruction processing submodule is also used to output motion instructions to drive the main control terminal 103 to actively move, for example, resetting the robot arm.

[0063] The graphics processing submodule, after receiving the key parameters of the instruction processing submodule, generates a real-time simulated training surgical field of view, and sends it to the display unit of the training robot subsystem through the transmission device; and finally displays it through the simulation result display.

[0064] The surgical training module may include an interactive submodule for reminding the user of the current training status through sound and light. Of course, the above functions may also be implemented through the human-computer interaction module.

[0065] The training robot subsystem consists of the following parts: main control terminal, sensor, central control unit, and data processing module.

[0066] The central control unit is used to respond to the motion command output by the command processing submodule to drive the main control end to actively move.

[0067] The main control end is used to follow the external force and obtain the actual main hand joint information mθ through the sensor and transmit it to the central processing unit for processing. The main control end is also used to perform active movements, such as resetting, under the control of the central control unit.

[0068] The central processing unit calculates the expected slave joint information sθ from the actual master hand joint information mθ through master-slave mapping and motion planning, and sends it to the surgical training module. It can be understood that if the expected slave joint information sθ is input to a slave device, the slave device will move, for example, drive a surgical instrument to move to the target position. However, in this training system, there is no slave device on the physical level, only a simulated slave device generated by the surgical training module.

[0069] The data processing module is mainly used to pre-process the relevant data, such as filtering, unit conversion, etc., and is also used to set the transmission of the data to be transmitted according to the appropriate protocol between systems.

[0070] Finally, the signal output by the training robot subsystem can control the movement of each virtual operating arm of the surgical robot simulation slave end.

[0071] In different embodiments, since the conversion process of the actual master hand joint information mθ and the expected slave end joint information sθ can be implemented by the training robot subsystem or the surgical training module, the actual master hand joint information mθ and the expected slave end joint information sθ can be understood as operation instructions, because both contain all the information related to the movement, but the data format is different.

[0072] The data module includes a data acquisition submodule and a data processing submodule.

[0073] After receiving the training start signal from the robot subsystem, the data acquisition submodule obtains the motion information of the virtual robot arm and the training screen of the simulation result display.

[0074] The data processing submodule receives the original data, processes the data, and outputs the processed feature information to the result evaluation module.

[0075] After the result evaluation module obtains the evaluation result, it is manually confirmed or modified to obtain the final result and sent to the data acquisition submodule; the unit data finally obtained is used to iteratively update the training evaluation model.

[0076] Furthermore, the data acquisition submodule is as follows Figure 3 The new data is collected in the manner shown. That is, the training method includes: collecting new data during the execution of the training method, and updating the training evaluation model based on the new data; the process of collecting new data includes: collecting time information and storing it as a timestamp and UID (User Identify, user ID, the specific encoding format of UID can be set according to actual needs, and will not be described in detail here), collecting image information and storing the storage path addr of the image information; collecting training plans and storing them as training information; collecting correction values ​​of evaluation results inputted from the outside and storing them as actual training results; collecting custom inputs and storing them as supplementary information.

[0077] Furthermore, the data processing submodule receives the original data and performs data processing, and the data cleaning process includes but is not limited to the following steps:

[0078] S101, missing data processing: For individual samples in the original data, check the completeness of the sample data and supplement the data based on experience or other information, such as the gender of the missing operator in a sample.

[0079] S102, data format processing: preliminary processing is performed on the format of the relevant data, the data content contains characters that should not exist, the content and field meanings do not match, etc., such as inconsistent time format, etc.

[0080] S103, logical error cleaning: process duplicate values, unreasonable values, and contradictory values. For example, if there are two columns of data, sex and gender, in the data sample, duplicates can be removed based on experience.

[0081] S104, remove useless data: that is, remove data that is completely useless in the training, such as the number of family members of the operator in a certain sample data, which can be removed; such as the storage location of the original record of the sample data, which can be removed.

[0082] S105, correlation verification: If there are multiple sources of data, correlation verification is required. For example, for a surgical training of the same operator, the surgical information collected by two nurses respectively needs to be verified, that is, the operator information, surgical procedure, surgical time, etc. of the surgery should be consistent.

[0083] In the above raw data processing, you can refer to historical data or communicate with people involved in data collection.

[0084] Meanwhile, in the training method, the step of obtaining the training evaluation model based on historical data includes: missing data processing; data format processing; logic error cleaning; removing useless data; and correlation verification. The step of collecting new data may also include: missing data processing; data format processing; logic error cleaning; removing useless data; and correlation verification.

[0085] The human-computer interaction module includes a training decision submodule, an evaluation decision submodule and a display and interaction submodule, and the user display submodule specifically includes an interactive information display.

[0086] The training decision submodule is used for users to confirm training instructions and send them to the surgical training module to generate training projects.

[0087] The evaluation decision submodule is used to set user-defined evaluation items and weights, and confirm the final evaluation results, which are sent to the data module for storage.

[0088] The display and interaction submodule is used to output interactive information through sound interaction and visual interaction (for example, when a user clicks a button, the sound of the button being clicked is played), and after the result evaluation module completes the training evaluation, the evaluation result report is sent to the interactive information display for presentation.

[0089] The steps of obtaining a training evaluation model based on historical data executed in the result evaluation module include: randomly dividing the historical data into training data and verification data according to an initial ratio; training the prediction model with the training data as input; for the prediction model that has completed training, using the verification data as input, performing model verification and calculating the accuracy; if the accuracy is lower than the first accuracy, randomly dividing the historical data into training data and verification data according to the initial ratio again and executing subsequent steps; if the accuracy is between the first accuracy and the second accuracy, re-training the prediction model with the training data as input and executing subsequent steps; if the accuracy is higher than the second accuracy, setting the prediction model to a training evaluation model.

[0090] Figure 4 A scheme for obtaining a training evaluation model.

[0091] In the information processing module, there are a large number of data samples sample(f,a), including the cleaned feature information f and the surgical evaluation result a corresponding to its image. The historical data is randomly divided into training data Sample_t and verification data Sample_v according to the initial ratio p (which can be any ratio from 0 to 1).

[0092] Take Sample_t data as input to train a model for predicting training evaluation results. For the prediction model output this time, use Sample_v data as input and output the evaluation result a' calculated by this model. Compare the actual evaluation result a with the model output result a' to get the model accuracy x', which is between 0 and 1.

[0093] After verifying all Sample_v, the overall accuracy x is obtained; if it is lower than the minimum tolerance accuracy, the data grouping is considered unreasonable and re-randomized; if the accuracy x is between the minimum tolerance accuracy and the passing accuracy, the model is retrained for verification; if the accuracy x is higher than the passing accuracy, the output model is used to calculate the labels for new data.

[0094] Whenever historical data is updated, repeat the above verification operation to iteratively upgrade the model. Optionally, the evaluation result a has different decision strategies. It can be based on different department needs, such as time optimization, effect optimization and global optimization, and can be processed based on manually set weights.

[0095] Specifically, the types of labels include: department, procedure, training goal, and scoring item. In some embodiments, some of the above types may also be included.

[0096] The training objectives include: basic training, incision selection, tissue separation, suturing and knotting, bleeding control and time control. In some embodiments, part of the above contents may also be included.

[0097] The scoring items include: visual field, incision location, tissue resection, time consumption, suturing efficiency and bleeding volume. In some embodiments, part of the above contents may also be included.

[0098] In a preferred embodiment, when the type of label is a scoring item, the step of training the prediction model with training data as input includes: for each scoring item, executing the following steps: determining a scoring standard set, the scoring standard set includes at least two scoring standards; determining a probability function of different outputs obtained from different inputs under each scoring standard based on the training data; determining a reward function; calculating a total reward based on the scoring standard set, the probability function and the reward function; maximizing the calculated total reward by continuously modifying the scoring standard set; and setting the scoring standard corresponding to the maximum total reward as the training result.

[0099] Scoring items include tissue resection effect, suturing effect, visual field score, etc.

[0100] like Figure 5 As shown in the figure, taking the tissue resection effect as an example, the decision-making process of finding the optimal score is introduced; in the existing data, the tissue resection effect S can adopt several evaluation criteria A to obtain different tissue resection effects S'. sample(s,a,s') means that under a certain tissue resection effect s, a certain evaluation criterion A is adopted, and the obtained tissue resection effect is S', T(s,a,s') represents the probability of the above sample(s,a,s'); rule π(s) represents the rule of the evaluation criterion A that I currently adopt under state s, R(s,π(s),s') represents the reward function for achieving a certain (s,π(s),s') based on rule π(s), then the evaluation scheme of the final decision plan is:

[0101]

[0102] For all data in a database, there is an optimal rule π*(s) that satisfies the following formula.

[0103]

[0104] At this time, there is a globally optimal allocation rule, and the evaluation result V(s) reaches the maximum value V*(S).

[0105] In the above text, "π*" and "V*" are both integral symbols, and "*" does not represent multiplication.

[0106] Preferably, when the label is time control, the training data is used as input, and the steps of training the prediction model include: dividing the operation process into multiple stages; setting the stage coefficient corresponding to each stage; calculating the time evaluation value based on the time consumption of the stage and the stage coefficient, and calculating the scoring result in combination with the time evaluation value; by continuously modifying the stage coefficient, the error between the calculated scoring result and the scoring result in the training data is minimized; the stage coefficient corresponding to the minimum error is set as the training result.

[0107] like Figure 6 As shown in the figure, for surgical time control, the surgical time can be divided into different stages, such as start-up time, transfer time, master-slave time, suturing time, closing time, adjustment time, etc. Then the time of each stage is scored, and the coefficient of each stage is adjusted to make the final result consistent with the actual situation. For different surgical procedures, the stages of an operation are divided differently, according to the surgical procedure and department requirements; at the same time, the stage coefficient can also be changed by changes in parameters such as surgical procedure, department, doctor proficiency, and surgical difficulty.

[0108] In other words, different scoring items will lead to different structures of the training evaluation model, and thus different ways of obtaining the training evaluation model.

[0109] In a preferred embodiment, the step of obtaining a training evaluation model based on historical data includes: extracting scoring items and scoring item weights based on departments, training goals, scores and / or grades.

[0110] The training method includes: automatically setting training goals based on departments and training evaluation models and / or modifying training goals based on external input; automatically setting scoring items and scoring item weights based on training goals and training evaluation models and / or modifying scoring items and scoring item weights based on external input.

[0111] The step of inputting the simulation training results into the training evaluation model to obtain the evaluation results includes: calculating the values ​​of the scoring items based on the simulation training results, and obtaining the scoring results in combination with the scoring item weights.

[0112] Please refer to Figure 7 , the historical data includes different training targets for different departments. Under each training target, there are scores and grades of different training sessions; each training session contains information that affects the scoring and evaluation, such as suturing status, operation time, etc. Based on these historical data, the data is processed to obtain the weight information of different training targets for different departments, and the weight information of each feature under different training targets. In subsequent training, after selecting a department, the priority training target can be automatically assigned based on the weight information of the training target. Training target assignment automatically assigns the scoring items with a larger weight corresponding to the current training target.

[0113] The step of obtaining a training evaluation model based on historical data includes: obtaining image features of evaluation results based on scoring items; the step of calculating scoring items based on simulation training results includes: obtaining image features based on simulation training results; and, calculating the values ​​of scoring items based on image features.

[0114] like Figure 8 As shown, the image features include: tissue edge feature 201 , suture feature 202 and field of view feature 203 .

[0115] Specifically, the steps of acquiring image features based on the simulation training results include: using Haar-like features for detection, Haar-like is a classic feature extraction algorithm, and those skilled in the art can undoubtedly know the algorithm object it refers to based on Haar-like; using integral graphs to accelerate the evaluation of Haar-like features; using AdaBoost (Adaptive Boosting) algorithm to train strong classifiers that distinguish between tissues and instruments; and using screening cascades to cascade strong classifiers together to identify tissue edge features, suture features and field of view features.

[0116] The above process uses the convolutional neural network method to extract feature information from the image, perform preliminary cleaning and classification on the original data, and reduce space complexity and overfitting problems.

[0117] Furthermore, the training method includes: obtaining a training scheme model based on historical data, the training scheme model including training state results and corresponding probabilities obtained under different training schemes; and selecting a training scheme with the best expected training state results as a training course based on the trainee's current training state and the training scheme model.

[0118] like Fig. 9 As shown in the figure: Taking the stitching efficiency S as an example, several course training schemes A can be adopted to obtain different stitching efficiencies S'. sample(s,a,s') means that under a certain stitching efficiency s, a certain training scheme A is adopted, and the stitching efficiency obtained is S', T(s,a,s') represents the probability of the above sample(s,a,s'); rule π(s) represents the law of the scheme A currently adopted under state s, R(s,π(s),s') represents the reward function for achieving a certain (s,π(s),s') based on rule π(s), then the evaluation scheme of the final decision scheme is:

[0119]

[0120] For all data in a database, there is an optimal rule π*(s) that satisfies the following formula.

[0121]

[0122] At this time, there is a globally optimal allocation rule, and the evaluation result V(s) reaches the maximum value V*(S).

[0123] In the above text, "π*" and "V*" are both integral symbols, and "*" does not represent multiplication.

[0124] Other training items such as suturing, operation time control, and optimal field of view can all be evaluated through this decision-making process. After obtaining the evaluation results, the weights of each evaluation item (the weights can be manually confirmed) are used for evaluation to arrive at the best training course schedule.

[0125] The training method includes: obtaining department information, training target information, scoring item information and scoring item weight information based on an interactive interface; generating training content based on department information and training target information; and obtaining evaluation results based on simulation training results, scoring item information and scoring item weight information.

[0126] Among them, the training target information, scoring item information and scoring item weight information can be automatically generated based on the training evaluation model, or can be manually confirmed and intervened, and the final scoring result can also be modified through manual confirmation.

[0127] The above process can refer to Fig.10 To understand. Fig.10 It can be seen that if there is no human confirmation, the training evaluation model can automatically select and assign training objectives, scoring items, scoring item weights, and automatically give evaluation results. But at the same time, these processes can be manually confirmed and intervened, so that the training evaluation model can be continuously improved and optimized during the operation of the system. Fig.10 In FIG. 1 , the step of generating new data to update the training evaluation model is also shown. After the training is completed, the data generated by this training is added to the historical data as new data.

[0128] Please refer to Fig.11a , Fig.11b , Fig.11c and Fig.11d ,in, Fig.11a This is a function menu interface of an embodiment of the present invention; Fig.11b This is a course operation visual interface of an embodiment of the present invention; Fig.11c is a goal setting interactive interface of an embodiment of the present invention; Fig.11d This is a training result report interface of an embodiment of the present invention. The function menu interface will display various functions of the training system, and the user can select different functions, for example, can select basic action training to enter the training project. Fig.11bA training process is demonstrated, which is tissue stripping training. The operation process and related information will be displayed on the interface. In the goal setting interactive interface, departments, training goals, scoring items and weights of each scoring item can be set. When there is no external input, the training system will give appropriate values ​​based on its own training evaluation model. When there is external input, the training process will be carried out according to the input results. The training result report interface can be set according to actual needs. In this embodiment, the scores of each scoring item are given and displayed in the form of a radar chart, which has good readability and is conducive to guiding trainees to adjust themselves according to the results.

[0129] This embodiment also provides a training software, which is used to perform the following steps:

[0130] S11, obtain historical data with labels. The historical data is obtained by two methods. The first method is that the data is collected by a related device not involved in this embodiment, and its label is automatically obtained when the device collects the data, or is added by manual identification. The second method is based on the training system / training method involved in this embodiment. At this time, the historical data is already labeled with a format that meets the requirements.

[0131] And, steps S20, S30, S40, S50 and S60.

[0132] The training software can also perform other extended steps involved in the above-mentioned training method.

[0133] This embodiment also provides a readable storage medium, on which a program is stored, and when the program is run, step S11 and steps S20, S30, S40, S50 and S60 are executed. The program can also execute other extended steps involved in the above training method.

[0134] In summary, this embodiment provides a training method and a training system. The training method includes: labeling historical data; obtaining a training evaluation model based on the historical data; wherein, based on different labels, the training evaluation model is obtained in at least two different ways; obtaining a training label; the training label has a one-to-one correspondence with the label; obtaining an operation instruction; calculating a simulation training result based on the operation instruction; the simulation training result has a correspondence with the result obtained by running the operation instruction in an actual operation environment; and, the simulation training result is input into the training evaluation model, and the training evaluation model obtains an evaluation result based on the label corresponding to the training label. Through personalized data cleaning, different forms of training evaluation models are established, making the training evaluation model more targeted, solving the problems of weak targeting during data processing and unsatisfactory scoring effect in the prior art.

[0135] The above description is only a description of the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any changes or modifications made by a person skilled in the art in the field of the present invention based on the above disclosure shall fall within the scope of protection of the technical solution of the present invention.

Claims

1. A training method, characterized in that: The training method comprises: Label the historical data; Obtaining a training evaluation model based on the historical data; wherein the training evaluation model is obtained in at least two different ways based on different labels; Acquire a training label; the training label has a one-to-one correspondence with the label; Get operation instructions; A simulation training result is calculated based on the operation instruction; the simulation training result has a corresponding relationship with a result obtained by running the operation instruction in an actual operation environment; and The simulation training results are input into the training evaluation model, and the training evaluation model obtains an evaluation result based on the label corresponding to the training label.

2. The training method according to claim 1, characterized in that: The types of labels include: at least one of department, procedure, training goal and scoring item; Wherein, the training objectives include: at least one of basic training, incision selection, tissue separation, suture knotting, bleeding control and time control; and / or, The scoring items include: at least one of visual field, incision positioning, tissue resection, time consumption, suturing efficiency and bleeding volume.

3. The training method according to claim 1, characterized in that: The step of obtaining a training evaluation model based on historical data includes: Missing data processing; data format processing; logical error cleaning; removal of useless data; and, correlation verification.

4. The training method according to claim 1, characterized in that: The step of obtaining a training evaluation model based on historical data includes: Randomly divide the historical data into training data and verification data according to an initial ratio; Training a prediction model using the training data as input; For the prediction model after training, using the verification data as input, performing model verification and calculating the accuracy; If the accuracy rate is lower than the first accuracy rate, the historical data is randomly re-divided into the training data and the verification data according to the initial ratio and subsequent steps are executed; if the accuracy rate is between the first accuracy rate and the second accuracy rate, the prediction model is re-trained with the training data as input and subsequent steps are executed; if the accuracy rate is higher than the second accuracy rate, the prediction model is set as the training evaluation model.

5. The training method according to claim 4, characterized in that: When the type of the label is a scoring item, the step of training the prediction model with the training data as input includes: For each scoring item, the following steps are performed: determining a scoring standard set, wherein the scoring standard set includes at least two scoring standards; determining a probability function of obtaining different outputs from different inputs under each of the scoring standards based on the training data; determining a reward function; calculating a total reward based on the scoring standard set, the probability function and the reward function; maximizing the calculated total reward by continuously modifying the scoring standard set; and setting the scoring standard corresponding to the maximum total reward as the training result.

6. The training method according to claim 4, characterized in that: When the label is time control, the step of training the prediction model with the training data as input includes: Divide the operation process into multiple stages; Set the phase coefficient corresponding to each of the phases; Calculating the time evaluation value based on the time consumption of the stage and the stage coefficient, and calculating the scoring result in combination with the time evaluation value; By continuously modifying the stage coefficient, the error between the calculated scoring result and the scoring result in the training data reaches a minimum value; The stage coefficient corresponding to the minimum error is set as the training result.

7. The training method according to claim 1, characterized in that: The step of obtaining a training evaluation model based on historical data includes: extracting scoring items and weights of the scoring items based on departments, training objectives, scores and / or grades; The training method comprises: automatically setting the training target based on the department and the training evaluation model and / or modifying the training target based on external input; automatically setting the scoring items and the scoring item weights based on the training target and the training evaluation model and / or modifying the scoring items and the scoring item weights based on external input; The step of inputting the simulation training results into the training evaluation model to obtain the evaluation results includes: calculating the values ​​of the scoring items based on the simulation training results, and calculating the scoring results in combination with the weights of the scoring items.

8. The training method according to claim 7, characterized in that: The step of obtaining the training evaluation model based on historical data includes: obtaining the image features of the evaluation results based on the scoring items; the step of calculating the scoring items based on the simulation training results includes: obtaining the image features based on the simulation training results; and calculating the values ​​of the scoring items based on the image features; The image features include: tissue edge features, suturing features and field of view features.

9. The training method according to any one of claims 1 to 8, characterized in that: The training method comprises: Acquire a training scheme model based on the historical data, wherein the training scheme model includes training state results and corresponding probabilities obtained under different training schemes; Based on the trainee's current training status and the training program model, the training program with the best expected training status result is selected as the training course.

10. The training method according to any one of claims 1 to 8, characterized in that: The training method comprises: Acquire the training label and scoring item weight information based on the interactive interface, wherein the training label includes: department information, training target information and scoring item information; Generate training content based on the department information and training target information; and, The evaluation result is obtained based on the simulation training result, the scoring item information and the scoring item weight information.

11. A training system, characterized in that: It includes human-computer interaction module, result evaluation module, surgical training module and training robot subsystem, among which, The result evaluation module is used to obtain a training evaluation model based on historical data; wherein the historical data is labeled; the result evaluation module obtains the training evaluation model in at least two different ways based on different labels; The training robot subsystem is used to obtain the operation instruction; The human-computer interaction module is used to obtain training labels; the training labels have a one-to-one correspondence with the labels; The result evaluation module is also used to obtain an evaluation result based on the simulation training result and the label corresponding to the training label; The surgical training module is used to calculate a simulation training result based on the operation instruction; the simulation training result has a corresponding relationship with a result obtained by running the operation instruction in an actual operation environment.

Images