Method and system for building interactive medical simulation scenarios based on AI artificial intelligence

Abstract

The present invention belongs to the field of medical simulation and education technology. The present invention discloses a method and system for building an interactive medical simulation scene based on AI artificial intelligence, including collecting patient physiological data, medical information data and patient feedback data; preprocessing the patient physiological data and medical information data to obtain a physiological characteristic data set and a medical information characteristic data set; fusing the patient feedback data into the physiological characteristic data set and the medical information characteristic data set to obtain a medical comprehensive characteristic data set; training a medical scene prediction model, inputting the medical comprehensive characteristic data set into the medical scene prediction model, and predicting medical scene simulation data; processing the predicted medical scene simulation data to obtain data simulation similarity error; providing more realistic data support for medical training, research and decision-making, and improving the effect and reliability of medical simulation.

Description

Technical Field

[0001] The present invention relates to the fields of medical simulation and educational technology, and more specifically, to a method and system for building an interactive medical simulation scene based on AI artificial intelligence. Background Art

[0002] As medical technology advances, the demand for medical education and training continues to grow. Existing medical simulation systems often rely on pre-set scripts and scenarios, lacking sufficient flexibility and interactivity to fully meet the teaching and training needs of complex clinical situations.

[0003] Patent application publication number CN112037629A discloses a medical surgery simulation system based on mixed reality technology, including: a data storage module for storing medical surgery information, including basic information of various medical devices and surgical information related to the various medical devices; a scene construction module for constructing a surgical simulation scene based on the medical surgery information; and a mixed reality holographic imaging module for constructing a mixed reality surgical simulation scene based on the surgical simulation scene, so that users can practice and observe surgical operations in the mixed reality surgical simulation scene. This invention can save hospitals and schools the cost and time of training medical staff and reduce the risk of operational errors for surgical interns, which is of great significance to both medical education and actual hospital work.

[0004] The existing methods for building medical simulation scenarios have the following main problems:

[0005] Evaluation methods that lack mathematical models may lead to subjectivity and inconsistency in the evaluation process. Different evaluators may make different assessments based on their personal experience and preferences, affecting the reliability and comparability of the evaluation results. This may lead to a lack of clear indicators and standards in the evaluation process. The evaluation results may lack quantitative indicators and comparable standards, making it difficult to accurately evaluate the quality and effectiveness of medical simulation scenarios. Without a clear evaluation method, it is impossible to identify problems and room for improvement in the standards through data and analysis, which limits the improvement of the quality of medical simulation scenarios.

[0006] The lack of processing of MRI image pixels makes it impossible to separate the medical structure of interest from the background, which will lead to the inability to accurately extract and utilize the characteristics of the medical structure in the subsequent model training and scene generation process, affecting the realism and accuracy of the simulation scene. This will cause the simulation scene to be unable to accurately reflect the three-dimensional form and positional relationship of the medical structure, limiting the realism and visualization effect of the simulation scene; the lack of rendering of the voxel three-dimensional model makes it impossible to simulate the propagation and interaction effects of light in the medical structure, reducing the realism and three-dimensional sense of the simulation scene.

[0007] In view of this, the present invention proposes an AI-based interactive medical simulation scenario construction method and system to solve the above problems. Summary of the Invention

[0008] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned objectives, the present invention provides the following technical solutions: a method for building an interactive medical simulation scene based on AI artificial intelligence, comprising:

[0009] S1. Collect patient physiological data, medical information data and patient feedback data;

[0010] S2. Preprocess the patient's physiological data and medical information data to obtain a physiological feature dataset and a medical information feature dataset; fuse the patient's feedback data into the physiological feature dataset and the medical information feature dataset to obtain a comprehensive medical feature dataset;

[0011] S3. Training a medical scenario prediction model, inputting the medical comprehensive feature dataset into the medical scenario prediction model, and predicting medical scenario simulation data;

[0012] S4. Processing the predicted medical scenario simulation data to obtain a data simulation similarity error; comparing the predicted data simulation similarity error with a preset data simulation similarity error threshold to determine whether the predicted medical scenario simulation data meets the medical simulation scenario construction standard;

[0013] S5. If the medical simulation scenario construction standard is met, the predicted medical scenario simulation data is processed to dynamically generate a medical simulation scenario.

[0014] Furthermore, the patient's physiological data includes the patient's blood pressure, heart rate and blood oxygen saturation; the medical information data includes case history data, video data of the medical process and treatment effect data; and the patient feedback data includes the patient's symptom change perception data and satisfaction data with the treatment effect.

[0015] Furthermore, the method of preprocessing the patient's physiological data and medical information data to obtain a physiological characteristic data set and a medical information characteristic data set includes:

[0016] A density clustering algorithm is used to identify outliers in the patient's physiological data and medical information data, and the identified outliers are eliminated to obtain the processed physiological characteristic data set and medical information characteristic data set; the obtained physiological characteristic data set and medical information characteristic data set are normalized by standard deviation, converted to standard normal distribution, and the dimensional influence between the data is eliminated to obtain the normalized physiological characteristic data set and medical information characteristic data set.

[0017] Furthermore, the method of fusing the patient feedback data into the physiological feature dataset and the medical information feature dataset to obtain a comprehensive medical feature dataset includes:

[0018] The patient feedback data is fused into the physiological feature dataset and the medical information feature dataset through a medical comprehensive feature fusion model; the training dataset is divided into a training set, a validation set, and a test set; the input data includes the patient feedback data, the physiological feature dataset, and the medical information feature dataset, and the output label is the medical comprehensive feature dataset; the medical comprehensive feature fusion model is a decision tree model;

[0019] The Gini index is selected as the loss function to measure the impurity of the data. At each node, the split feature is selected by calculating the Gini index of each feature. After the split feature is determined, the split threshold that minimizes the Gini index is selected as the optimal split point.

[0020] The Gini index is: Where G(D) is the Gini index of the medical comprehensive feature dataset D; m is the total number of different categories in the medical comprehensive feature dataset; p i is the proportion of samples belonging to the i-th category in the dataset D; D is a medical comprehensive feature dataset;

[0021] Set the minimum number of samples for each leaf node as the stopping condition to avoid over-splitting;

[0022] By setting the ccp_alpha parameter to perform post-pruning, adjust the complexity of the model, and prune the model to prevent overfitting; cross-validation is used to evaluate the impact of different ccp_alpha values ​​on model performance;

[0023] Use the training set data to train the model; use the validation set to adjust the model parameters and find the optimal model parameter settings; use the test set data to evaluate the performance of the model, using accuracy as the evaluation indicator;

[0024] A density clustering algorithm is used on the medical comprehensive feature data set to identify outliers in the medical comprehensive feature data set, and the identified outliers are eliminated to obtain the processed medical comprehensive feature data set; the obtained medical comprehensive feature data set is normalized by standard deviation, converted to standard normal distribution, and the dimensional influence between the data is eliminated to obtain the normalized medical comprehensive feature data set.

[0025] Furthermore, the training method of the medical scenario prediction model includes:

[0026] Prepare a training data set, including input data and corresponding labels; the input data is a medical comprehensive feature data set, and the output label is medical scenario simulation data; the medical scenario prediction model structure includes an input layer, a hidden layer, and an output layer; the medical scenario prediction model is a multi-layer perceptron (MLP) model;

[0027] Initialize the weights and biases of the medical scenario prediction model using a random initialization method;

[0028] For the weights of each layer: determine the dimensions of the weight matrix, with the input dimension being m1 and the output dimension being ; select the range [-b, b], where b is a positive number, which is the range of weight values; randomly generate values ​​from a uniform distribution as the initial weights;

[0029] For each layer's bias: determine the dimension of the bias vector, with the output dimension being m2; select the range [-c, c], where c is the range of bias values; and generate a random value from a uniform distribution as the initial bias.

[0030] Input the feature vector of the training data into the medical scenario prediction model; calculate the output value of each layer through the weight and bias of each layer, and pass it to the output layer; apply a nonlinear activation function to the output of the hidden layer to convert the output of the hidden layer into a nonlinear form;

[0031] Use the prediction results of the output layer and the label data to calculate the loss function to measure the error of the model prediction;

[0032] Using mean square error as the loss function, the calculation formula of the mean square error loss function is: Where y is the actual medical scenario simulation data; y′ is the predicted medical scenario simulation data;

[0033] Through the back-propagation algorithm, the gradient of each parameter with respect to the loss function is calculated, and the parameters are updated to reduce the value of the loss function. The stochastic gradient descent algorithm is used to update the parameters of the model according to the gradient, so that the loss function gradually decreases.

[0034] Adjust the learning rate to control the speed and stability of parameter updates and optimize the medical scenario prediction model. Use the Adam algorithm as the optimizer. The Adam algorithm is based on optimizing stochastic gradient descent and adaptively adjusts the learning rate.

[0035] When the training reaches the predetermined number of iterations, the final trained medical scenario prediction model is obtained; the trained medical scenario prediction model is used to predict the current medical comprehensive feature data set to obtain medical scenario simulation data.

[0036] Furthermore, the method of processing the predicted medical scenario simulation data to obtain the data simulation similarity error includes:

[0037] Collect real-time data from medical scenarios, cleanse and preprocess it to ensure consistent data format, no missing values, and no outliers, so that the predicted medical scenario simulation data and the real-time medical scenario data are similar in distribution and characteristics; select the Pearson correlation coefficient to calculate the similarity between the predicted data and the real data;

[0038] The error index is calculated by the mean square error to quantify the difference between the predicted medical scenario simulation data and the real-time medical scenario data, thereby obtaining the data simulation similarity error.

[0039] Furthermore, the method of comparing the predicted data simulation similarity error with a preset data simulation similarity error threshold to determine whether the predicted medical scenario simulation data meets the medical simulation scenario construction standard includes:

[0040] If the predicted data simulation similarity error is less than or equal to a preset data simulation similarity error threshold, it is determined that the predicted medical scenario simulation data meets the medical simulation scenario construction standard;

[0041] If the predicted data simulation similarity error is greater than a preset data simulation similarity error threshold, it is determined that the predicted medical scenario simulation data does not meet the medical simulation scenario construction standard.

[0042] Furthermore, the evaluation method of the medical simulation scenario construction standard includes:

[0043] The medical simulation scenario construction standard is evaluated by constructing a standard evaluation mathematical model, and the standard evaluation mathematical model is: Among them, SZP is the standard for building medical simulation scenarios; is the average value of the patient's physiological data; is the average value of medical information data; The average value of the patient feedback data; is the average value of the patient's physiological data, medical information data and patient feedback data that are erroneous during acquisition; t is the time required to acquire the patient's physiological data, medical information data and patient feedback data; α1 is the influencing factor of the average value of the patient's physiological data; α2 is the influencing factor of the average value of the medical information data; α3 is the influencing factor of the average value of the patient's feedback data; α4 is the influencing factor of the average value of the patient's physiological data, medical information data and patient feedback data that are erroneous during acquisition.

[0044] Furthermore, the method of processing the predicted medical scenario simulation data to dynamically generate a medical simulation scenario includes:

[0045] The medical scene simulation data is two-dimensional medical scene simulation data, including CT scan data and MRI images; the acquired two-dimensional medical scene simulation data is subjected to standard deviation normalization processing to convert it into a standard normal distribution;

[0046] Preset an image segmentation threshold, the MRI image is a grayscale image, and the MRI image pixels are divided into background and foreground parts through an image segmentation model, and the foreground part is separated from the background part using image segmentation technology;

[0047] The image segmentation model is: Among them, I SG (x, y) is the segmented MRI image; 1 is the foreground part; 0 is the background part; I(x, y) is the pixel value of the original MRI image at position (x, y); T is the image segmentation threshold;

[0048] The optimal threshold value of the image segmentation threshold T is automatically determined by the Otsu algorithm. Assuming that the total number of pixels in the MRI image is N and the gray levels are 0 and L-1, the probability of occurrence of gray levels in the image is estimated as: Among them, p(i) is the probability of occurrence of gray level; n i is the number of pixels at gray level i;

[0049] For the image segmentation threshold T, the image is divided into two categories C0 and C1; the weights of categories C0 and C1 are and Among them, ω0 and ω1 are the weights of categories C0 and C1;

[0050] Calculate the average value of the gray value as and Calculate the inter-class variance of pixel grayscale values ​​of different background or foreground categories;

[0051] The between-class variance is: R =ω0(μ0-μ T ) 2 +ω1(μ1-μ T ) 2 ; Among them, δ R is the between-class variance; μ T is the average grayscale value of all pixels in the MRI image;

[0052] Traverse all image segmentation thresholds T and select δ R The maximized image segmentation threshold T is taken as the optimal threshold;

[0053] Preset an image sequence {I1,I2,...,I n′}, each image I KFor each volume layer, a voxel 3D model is constructed, wherein the value of each voxel is determined by the image pixel value at the corresponding position; the voxel 3D model is: V(x′, y′, z′); wherein z′ is the z′th image in the image sequence;

[0054] An isosurface threshold is preset and the isosurface Ts is constructed by checking the boundary of the voxel grid. A set of triangles is generated by the MarchingCubes algorithm, and the triangles approximately represent the isosurface Ts. The vertices of each triangle are determined by linear interpolation.

[0055] The vertices of the triangle are: Where P1 and P2 are two points on the voxel grid boundary; V(P1) and V(P2) are the voxel values ​​of points P1 and P2, and P is the position of the triangle vertex;

[0056] The voxel 3D model is rendered using a ray casting algorithm to generate a medical simulation scene.

[0057] The system is built based on AI artificial intelligence interactive medical simulation scenarios, including: a data collection module for collecting patient physiological data, medical information data and patient feedback data;

[0058] The data processing module is used to pre-process the patient's physiological data and medical information data to obtain a physiological characteristic data set and a medical information characteristic data set; and fuse the patient feedback data into the physiological characteristic data set and the medical information characteristic data set to obtain a comprehensive medical characteristic data set;

[0059] The model training module is used to train the medical scenario prediction model. The medical comprehensive feature dataset is input into the medical scenario prediction model to predict the medical scenario simulation data.

[0060] A scenario construction evaluation module is used to process the predicted medical scenario simulation data to obtain a data simulation similarity error; compare the predicted data simulation similarity error with a preset data simulation similarity error threshold to determine whether the predicted medical scenario simulation data meets the medical simulation scenario construction standard;

[0061] The simulation scenario building module processes the predicted medical scenario simulation data and dynamically generates a medical simulation scenario if the medical simulation scenario building standards are met; each module is connected through wired and / or wireless means to realize data transmission between modules.

[0062] The technical effects and advantages of the AI-based interactive medical simulation scene construction method and system are as follows:

[0063] By constructing a standard evaluation mathematical model to evaluate the medical simulation scene construction standards, the evaluation process can be made more objective and consistent; the mathematical model can provide clear indicators and standards, reduce subjective interference, ensure the accuracy and comparability of the evaluation results, and standardize and normalize the medical simulation scene construction standards; it can help identify the key factors and influencing factors in the standards, and then guide the formulation and implementation of improvement measures to improve the quality and effectiveness of medical simulation scenes; through the evaluation of the standard evaluation mathematical model, the advantages and disadvantages of the medical simulation scene construction standards can be understood, thereby providing a reference and basis for decision-making and helping decision makers make wise decisions;

[0064] By using an image segmentation model to separate MRI image pixels into background and foreground parts, the medical structure of interest can be separated from the background, and more representative features can be extracted, providing a better data foundation for subsequent model training and scene generation; by converting two-dimensional medical scene simulation data into a voxel three-dimensional model, the spatial information of the medical scene can be more comprehensively represented; the voxel three-dimensional model can stack different body layers of the medical scene to form a three-dimensional model, which more realistically simulates the spatial structure of the medical scene; by rendering the voxel three-dimensional model through a ray casting algorithm, a medical simulation scene can be generated; the rendering process can simulate the propagation and interaction of light in the medical structure, producing effects such as shadows, reflections and refractions, making the medical simulation scene more realistic and three-dimensional. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of the method for building an AI-based interactive medical simulation scenario of the present invention;

[0066] Figure 2 This is a schematic diagram of the system structure for building an AI-based interactive medical simulation scenario. DETAILED DESCRIPTION

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0068] Example 1

[0069] See also Figure 1 and Figure 2 As shown, this embodiment is based on an AI artificial intelligence interactive medical simulation scenario construction method and system, including: S1, collecting patient physiological data, medical information data and patient feedback data;

[0070] S2. Preprocess the patient's physiological data and medical information data to obtain a physiological feature dataset and a medical information feature dataset; fuse the patient's feedback data into the physiological feature dataset and the medical information feature dataset to obtain a comprehensive medical feature dataset;

[0071] S3. Training a medical scenario prediction model, inputting the medical comprehensive feature dataset into the medical scenario prediction model, and predicting medical scenario simulation data;

[0072] S4. Processing the predicted medical scenario simulation data to obtain a data simulation similarity error; comparing the predicted data simulation similarity error with a preset data simulation similarity error threshold to determine whether the predicted medical scenario simulation data meets the medical simulation scenario construction standard;

[0073] S5. If the medical simulation scenario construction standard is met, the predicted medical scenario simulation data is processed to dynamically generate a medical simulation scenario.

[0074] The patient's physiological data includes the patient's blood pressure, heart rate and blood oxygen saturation; the medical information data includes case history data, video data of the medical process and treatment effect data; and the patient feedback data includes the patient's symptom change perception data and satisfaction data with the treatment effect.

[0075] The patient's blood pressure is obtained by a sphygmomanometer; the patient's heart rate is obtained by an electrocardiograph; the blood oxygen saturation is obtained by a pulse oximeter; the case history data is obtained from the patient's medical records; the video data of the medical process is obtained by the hospital's camera equipment; the treatment effect data is obtained through the doctor's observation and evaluation, and the patient's symptom improvement; the patient's symptom change perception data and satisfaction data with the treatment effect are obtained through the patient's subjective feedback;

[0076] The method of preprocessing the patient's physiological data and medical information data to obtain a physiological characteristic data set and a medical information characteristic data set includes:

[0077] A density clustering algorithm is used to identify outliers in the patient's physiological data and medical information data, and the identified outliers are eliminated to obtain the processed physiological characteristic data set and medical information characteristic data set; the obtained physiological characteristic data set and medical information characteristic data set are normalized by standard deviation, converted to standard normal distribution, and the dimensional influence between the data is eliminated to obtain the normalized physiological characteristic data set and medical information characteristic data set.

[0078] The method of fusing the patient feedback data into the physiological feature dataset and the medical information feature dataset to obtain the comprehensive medical feature dataset includes:

[0079] The patient feedback data is fused into the physiological feature dataset and the medical information feature dataset through a medical comprehensive feature fusion model; the training dataset is divided into a training set, a validation set, and a test set; the input data includes the patient feedback data, the physiological feature dataset, and the medical information feature dataset, and the output label is the medical comprehensive feature dataset; the medical comprehensive feature fusion model is a decision tree model;

[0080] The Gini index is selected as the loss function to measure the impurity of the data. At each node, the split feature is selected by calculating the Gini index of each feature. After the split feature is determined, the split threshold that minimizes the Gini index is selected as the optimal split point.

[0081] The Gini index is: Where G(D) is the Gini index of the medical comprehensive feature dataset D; m is the total number of different categories in the medical comprehensive feature dataset; p i is the proportion of samples belonging to the i-th category in the dataset D; D is a medical comprehensive feature dataset;

[0082] Set the minimum number of samples for each leaf node as the stopping condition to avoid excessive splitting;

[0083] By setting the ccp_alpha parameter to perform post-pruning, adjust the complexity of the model, and prune the model to prevent overfitting; cross-validation is used to evaluate the impact of different ccp_alpha values ​​on model performance;

[0084] Use the training set data to train the model; use the validation set to adjust the model parameters and find the optimal model parameter settings; use the test set data to evaluate the performance of the model, using accuracy as the evaluation indicator;

[0085] A density clustering algorithm is used on the medical comprehensive feature data set to identify outliers in the medical comprehensive feature data set, and the identified outliers are eliminated to obtain the processed medical comprehensive feature data set; the obtained medical comprehensive feature data set is normalized by standard deviation, converted to standard normal distribution, and the dimensional influence between the data is eliminated to obtain the normalized medical comprehensive feature data set.

[0086] The training method of the medical scenario prediction model includes:

[0087] Prepare a training data set, including input data and corresponding labels; the input data is a medical comprehensive feature data set, and the output label is medical scenario simulation data; the medical scenario prediction model structure includes an input layer, a hidden layer, and an output layer; the medical scenario prediction model is a multi-layer perceptron (MLP) model;

[0088] Initialize the weights and biases of the medical scenario prediction model using a random initialization method;

[0089] For the weights of each layer: determine the dimensions of the weight matrix, with the input dimension being m1 and the output dimension being ; select the range [-b, b], where b is a positive number, which is the range of weight values; randomly generate values ​​from a uniform distribution as the initial weights;

[0090] For each layer's bias: determine the dimension of the bias vector, with the output dimension being m2; select the range [-c, c], where c is the range of bias values; and generate a random value from a uniform distribution as the initial bias.

[0091] Input the feature vector of the training data into the medical scenario prediction model; calculate the output value of each layer through the weight and bias of each layer, and pass it to the output layer; apply a nonlinear activation function to the output of the hidden layer to convert the output of the hidden layer into a nonlinear form;

[0092] Use the prediction results of the output layer and the label data to calculate the loss function to measure the error of the model prediction;

[0093] Using mean square error as the loss function, the calculation formula of the mean square error loss function is: Where y is the actual medical scenario simulation data; y′ is the predicted medical scenario simulation data;

[0094] Through the back-propagation algorithm, the gradient of each parameter with respect to the loss function is calculated, and the parameters are updated to reduce the value of the loss function. The stochastic gradient descent algorithm is used to update the parameters of the model according to the gradient, so that the loss function gradually decreases.

[0095] Adjust the learning rate to control the speed and stability of parameter updates and optimize the medical scenario prediction model. Use the Adam algorithm as the optimizer. The Adam algorithm is based on optimizing stochastic gradient descent and adaptively adjusts the learning rate.

[0096] When the training reaches the predetermined number of iterations, the final trained medical scenario prediction model is obtained; the trained medical scenario prediction model is used to predict the current medical comprehensive feature data set to obtain medical scenario simulation data.

[0097] The method for processing the predicted medical scenario simulation data to obtain the data simulation similarity error includes:

[0098] Collect real-time data from medical scenarios, cleanse and preprocess it to ensure consistent data format, no missing values, and no outliers, so that the predicted medical scenario simulation data and the real-time medical scenario data are similar in distribution and characteristics; select the Pearson correlation coefficient to calculate the similarity between the predicted data and the real data;

[0099] The error index is calculated by the mean square error to quantify the difference between the predicted medical scenario simulation data and the real-time medical scenario data, thereby obtaining the data simulation similarity error.

[0100] For example, there is a set of real-time collected patient blood glucose data and corresponding prediction data;

[0101] patient Real-time blood sugar value Predicting blood sugar levels A 5.6 5.8 B 7.2 7.0 C 6.1 6.4

[0102] Then the mean square error calculation error index is:

[0103]

[0104] The method of comparing the predicted data simulation similarity error with a preset data simulation similarity error threshold to determine whether the predicted medical scenario simulation data meets the medical simulation scenario construction standard includes:

[0105] If the predicted data simulation similarity error is less than or equal to a preset data simulation similarity error threshold, it is determined that the predicted medical scenario simulation data meets the medical simulation scenario construction standard;

[0106] If the predicted data simulation similarity error is greater than a preset data simulation similarity error threshold, it is determined that the predicted medical scenario simulation data does not meet the medical simulation scenario construction standard.

[0107] The evaluation method of the medical simulation scenario construction standard includes:

[0108] The medical simulation scenario construction standard is evaluated by constructing a standard evaluation mathematical model, and the standard evaluation mathematical model is: Among them, SZP is the standard for building medical simulation scenarios; is the average value of the patient's physiological data; is the average value of medical information data; The average value of the patient feedback data; is the average value of the patient's physiological data, medical information data and patient feedback data that are erroneous during acquisition; t is the time required to acquire the patient's physiological data, medical information data and patient feedback data; α1 is the influencing factor of the average value of the patient's physiological data; α2 is the influencing factor of the average value of the medical information data; α3 is the influencing factor of the average value of the patient's feedback data; α4 is the influencing factor of the average value of the patient's physiological data, medical information data and patient feedback data that are erroneous during acquisition.

[0109] For example, the average value of the preset patient physiological data is 12, the average value of the medical information data is 15, the average value of the patient feedback data is 2, the average value of the patient physiological data, medical information data and patient feedback data that are erroneous during acquisition is 25, the time required to acquire the patient physiological data, medical information data and patient feedback data is 24 hours, the average value influence factor of the patient physiological data is 25%, the average value influence factor of the medical information data is 20%, the average value influence factor of the patient feedback data is 30%, and the average value influence factor of the patient physiological data, medical information data and patient feedback data that are erroneous during acquisition is 25%. Then the standard value for constructing the medical simulation scenario is approximately 0.883.

[0110] The method for processing the predicted medical scenario simulation data to dynamically generate a medical simulation scenario includes:

[0111] The medical scene simulation data is two-dimensional medical scene simulation data, including CT scan data and MRI images; the acquired two-dimensional medical scene simulation data is subjected to standard deviation normalization processing to convert it into a standard normal distribution;

[0112] Preset an image segmentation threshold, the MRI image is a grayscale image, and the MRI image pixels are divided into background and foreground parts through an image segmentation model, and the foreground part is separated from the background part using image segmentation technology;

[0113] The image segmentation model is: Among them, I SG (x, y) is the segmented MRI image; 1 is the foreground part; 0 is the background part; I(x, y) is the pixel value of the original MRI image at position (x, y); T is the image segmentation threshold;

[0114] The optimal threshold value of the image segmentation threshold T is automatically determined by the Otsu algorithm. Assuming that the total number of pixels in the MRI image is N and the gray levels are 0 and L-1, the probability of occurrence of gray levels in the image is estimated as: Among them, p(i) is the probability of occurrence of gray level; n i is the number of pixels at gray level i;

[0115] For the image segmentation threshold T, the image is divided into two categories C0 and C1; the weights of categories C0 and C1 are and Among them, ω0 and ω1 are the weights of categories C0 and C1;

[0116] Calculate the average value of the gray value as and Calculate the inter-class variance of pixel grayscale values ​​of different background or foreground categories;

[0117] The between-class variance is:R =ω0(μ0-μ T ) 2 +ω1(μ1-μ T ) 2 ; Among them, δ R is the between-class variance; μ T is the average grayscale value of all pixels in the MRI image;

[0118] Traverse all image segmentation thresholds T and select δ R The maximized image segmentation threshold T is taken as the optimal threshold;

[0119] Preset an image sequence {I1,I2,...,I n′}, each image I K For each volume layer, a voxel 3D model is constructed, wherein the value of each voxel is determined by the image pixel value at the corresponding position; the voxel 3D model is: V(x′, y′, z′); wherein z′ is the z′th image in the image sequence;

[0120] An isosurface threshold is preset and the isosurface Ts is constructed by checking the boundary of the voxel grid. A set of triangles is generated by the MarchingCubes algorithm, and the triangles approximately represent the isosurface Ts. The vertices of each triangle are determined by linear interpolation.

[0121] The vertices of the triangle are: Where P1 and P2 are two points on the voxel grid boundary; V(P1) and V(P2) are the voxel values ​​of points P1 and P2, and P is the position of the triangle vertex;

[0122] Render the voxel 3D model through ray casting algorithm to generate medical simulation scene;

[0123] The preset data simulation similarity error threshold is set by the staff, and different data simulation similarity errors are collected through the interactive medical management terminal, and the average value of multiple data simulation similarity errors is taken as the preset data simulation similarity error threshold; the preset image segmentation threshold is set similarly;

[0124] In this embodiment, by constructing a standard evaluation mathematical model to evaluate the medical simulation scenario construction standards, the evaluation process can be made more objective and consistent. The mathematical model can provide clear indicators and standards, reduce subjective interference, ensure the accuracy and comparability of evaluation results, and standardize and normalize the medical simulation scenario construction standards. It can help identify key factors and influencing factors in the standards, thereby guiding the formulation and implementation of improvement measures and improving the quality and effectiveness of medical simulation scenarios. Through the evaluation of the standard evaluation mathematical model, the strengths and weaknesses of the medical simulation scenario construction standards can be understood, thereby providing a reference and basis for decision-making and helping decision makers make wise decisions.

[0125] By using an image segmentation model to separate MRI image pixels into background and foreground parts, the medical structure of interest can be separated from the background, and more representative features can be extracted, providing a better data foundation for subsequent model training and scene generation; by converting two-dimensional medical scene simulation data into a voxel three-dimensional model, the spatial information of the medical scene can be more comprehensively represented; the voxel three-dimensional model can stack different body layers of the medical scene to form a three-dimensional model, which more realistically simulates the spatial structure of the medical scene; by rendering the voxel three-dimensional model through a ray casting algorithm, a medical simulation scene can be generated; the rendering process can simulate the propagation and interaction of light in the medical structure, producing effects such as shadows, reflections and refractions, making the medical simulation scene more realistic and three-dimensional.

[0126] Example 2

[0127] See also Figure 1 As shown, this embodiment is based on an AI artificial intelligence interactive medical simulation scenario building system, including: a data collection module for collecting patient physiological data, medical information data and patient feedback data;

[0128] The data processing module is used to pre-process the patient's physiological data and medical information data to obtain a physiological characteristic data set and a medical information characteristic data set; and fuse the patient feedback data into the physiological characteristic data set and the medical information characteristic data set to obtain a comprehensive medical characteristic data set;

[0129] The model training module is used to train the medical scenario prediction model. The medical comprehensive feature dataset is input into the medical scenario prediction model to predict the medical scenario simulation data.

[0130] A scenario construction evaluation module is used to process the predicted medical scenario simulation data to obtain a data simulation similarity error; compare the predicted data simulation similarity error with a preset data simulation similarity error threshold to determine whether the predicted medical scenario simulation data meets the medical simulation scenario construction standard;

[0131] The simulation scenario building module processes the predicted medical scenario simulation data and dynamically generates a medical simulation scenario if the medical simulation scenario building standards are met; each module is connected through wired and / or wireless means to realize data transmission between modules.

[0132] Since the electronic device introduced in this embodiment is an electronic device used to implement the method and system for building an interactive medical simulation scene based on AI artificial intelligence in the embodiment of this application, based on the method and system for building an interactive medical simulation scene based on AI artificial intelligence introduced in the embodiment of this application, technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application will not be described in detail here. As long as technical personnel in this field implement the electronic device used in the method and system for building an interactive medical simulation scene based on AI artificial intelligence in the embodiment of this application, it falls within the scope of protection of this application.

[0133] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0134] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for users of ordinary skill in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for building an interactive medical simulation scene based on AI, characterized in that: include: S1. Collect patient physiological data, medical information data and patient feedback data; S2. Preprocessing the patient's physiological data and medical information data to obtain a physiological feature dataset and a medical information feature dataset; Fusing patient feedback data into the physiological feature dataset and the medical information feature dataset to obtain a comprehensive medical feature dataset; S3. Training a medical scenario prediction model, inputting the medical comprehensive feature dataset into the medical scenario prediction model, and predicting medical scenario simulation data; S4. Processing the predicted medical scenario simulation data to obtain a data simulation similarity error; comparing the predicted data simulation similarity error with a preset data simulation similarity error threshold to determine whether the predicted medical scenario simulation data meets the medical simulation scenario construction standard; Methods for processing predicted medical scenario simulation data to obtain data simulation similarity errors include: Collect real-time data from medical scenarios, cleanse and preprocess it to ensure consistent data format, no missing values, and no outliers, so that the predicted medical scenario simulation data and the real-time medical scenario data are similar in distribution and characteristics; select the Pearson correlation coefficient to calculate the similarity between the predicted data and the real data; The error index is calculated by the mean square error to quantify the difference between the predicted medical scenario simulation data and the real-time medical scenario data, thereby obtaining the data simulation similarity error; S5. If the medical simulation scenario construction standard is met, the predicted medical scenario simulation data is processed to dynamically generate a medical simulation scenario; The method for processing the predicted medical scenario simulation data and dynamically generating a medical simulation scenario includes: The medical scene simulation data is two-dimensional medical scene simulation data, including CT scan data and MRI images; the acquired two-dimensional medical scene simulation data is normalized by standard deviation and converted into a standard normal distribution; Preset an image segmentation threshold, the MRI image is a grayscale image, and the MRI image pixels are divided into background and foreground parts through an image segmentation model, and the foreground part is separated from the background part using image segmentation technology; Preset an image sequence {I1,I2,...,I n′ }, each image I K For each volume layer, a voxel 3D model is constructed, wherein the value of each voxel is determined by the image pixel value at the corresponding position; the voxel 3D model is: V(x′, y′, z′); wherein z′ is the z′th image in the image sequence; Preset an isosurface threshold and construct the isosurface T by checking the boundary of the voxel grid s ; Generate a set of triangles through the Marching Cubes algorithm, which approximates the isosurface T s ;The vertices of each triangle are determined by linear interpolation; The vertices of the triangle are: Where P1 and P2 are two points on the voxel grid boundary; V(P1) and V(P2) are the voxel values ​​of points P1 and P2, and P is the position of the triangle vertex; The voxel 3D model is rendered using a ray casting algorithm to generate a medical simulation scene.

2. The method for building an AI-based interactive medical simulation scene according to claim 1, characterized in that: The patient's physiological data includes the patient's blood pressure, heart rate and blood oxygen saturation; the medical information data includes case history data, video data of the medical process and treatment effect data; and the patient feedback data includes the patient's symptom change perception data and satisfaction data with the treatment effect.

3. The method for building an AI-based interactive medical simulation scene according to claim 2, characterized in that: The method of preprocessing the patient's physiological data and medical information data to obtain a physiological characteristic data set and a medical information characteristic data set includes: A density clustering algorithm is used to identify outliers in the patient's physiological data and medical information data, and the identified outliers are eliminated to obtain the processed physiological characteristic data set and medical information characteristic data set; the obtained physiological characteristic data set and medical information characteristic data set are normalized by standard deviation, converted to standard normal distribution, and the dimensional influence between the data is eliminated to obtain the normalized physiological characteristic data set and medical information characteristic data set.

4. The method for building an AI-based interactive medical simulation scene according to claim 3 is characterized in that: The method of fusing the patient feedback data into the physiological feature dataset and the medical information feature dataset to obtain the comprehensive medical feature dataset includes: The patient feedback data is fused into the physiological feature dataset and the medical information feature dataset through a medical comprehensive feature fusion model; the training dataset is divided into a training set, a validation set, and a test set; the input data includes the patient feedback data, the physiological feature dataset, and the medical information feature dataset, and the output label is the medical comprehensive feature dataset; the medical comprehensive feature fusion model is a decision tree model; The Gini index is selected as the loss function to measure the impurity of the data. At each node, the split feature is selected by calculating the Gini index of each feature. After the split feature is determined, the split threshold that minimizes the Gini index is selected as the optimal split point. The Gini index is: Where G(D) is the Gini index of the medical comprehensive feature dataset D; m is the total number of different categories in the medical comprehensive feature dataset; p i is the proportion of samples belonging to the i-th category in the dataset D; D is a medical comprehensive feature dataset; Set the minimum number of samples for each leaf node as the stopping condition to avoid excessive splitting; By setting the ccp_alpha parameter to perform post-pruning, adjust the complexity of the model, and prune the model to prevent overfitting; cross-validation is used to evaluate the impact of different ccp_alpha values ​​on model performance; Use the training set data to train the model; use the validation set to adjust the model parameters and find the optimal model parameter settings; use the test set data to evaluate the performance of the model, using accuracy as the evaluation indicator; A density clustering algorithm is used on the medical comprehensive feature data set to identify outliers in the medical comprehensive feature data set, and the identified outliers are eliminated to obtain the processed medical comprehensive feature data set; the obtained medical comprehensive feature data set is normalized by standard deviation, converted to standard normal distribution, and the dimensional influence between the data is eliminated to obtain the normalized medical comprehensive feature data set.

5. The method for building an AI-based interactive medical simulation scene according to claim 4 is characterized in that: The training method of the medical scenario prediction model includes: Prepare a training data set, including input data and corresponding labels; the input data is a medical comprehensive feature data set, and the output label is medical scenario simulation data; the medical scenario prediction model structure includes an input layer, a hidden layer, and an output layer; the medical scenario prediction model is a multi-layer perceptron (MLP) model; Initialize the weights and biases of the medical scenario prediction model using a random initialization method; For the weights of each layer: determine the dimensions of the weight matrix, with the input dimension being m1 and the output dimension being ; select the range [-b, b], where b is a positive number, which is the range of weight values; randomly generate values ​​from a uniform distribution as the initial weights; For each layer's bias: determine the dimension of the bias vector, with the output dimension being m2; select the range [-c, c], where c is the range of bias values; and generate a random value from a uniform distribution as the initial bias. Input the feature vector of the training data into the medical scenario prediction model; calculate the output value of each layer through the weight and bias of each layer, and pass it to the output layer; apply a nonlinear activation function to the output of the hidden layer to convert the output of the hidden layer into a nonlinear form; Use the prediction results of the output layer and the label data to calculate the loss function to measure the error of the model prediction; Using mean square error as the loss function, the calculation formula of the mean square error loss function is: Where y is the actual medical scenario simulation data; y′ is the predicted medical scenario simulation data; Through the back-propagation algorithm, the gradient of each parameter with respect to the loss function is calculated, and the parameters are updated to reduce the value of the loss function. The stochastic gradient descent algorithm is used to update the parameters of the model according to the gradient, so that the loss function gradually decreases. Adjust the learning rate to control the speed and stability of parameter updates and optimize the medical scenario prediction model. Use the Adam algorithm as the optimizer. The Adam algorithm is based on optimizing stochastic gradient descent and adaptively adjusts the learning rate. When the training reaches the predetermined number of iterations, the final trained medical scenario prediction model is obtained; the trained medical scenario prediction model is used to predict the current medical comprehensive feature data set to obtain medical scenario simulation data.

6. The method for building an AI-based interactive medical simulation scene according to claim 5, characterized in that: The method of comparing the predicted data simulation similarity error with a preset data simulation similarity error threshold to determine whether the predicted medical scenario simulation data meets the medical simulation scenario construction standard includes: If the predicted data simulation similarity error is less than or equal to a preset data simulation similarity error threshold, it is determined that the predicted medical scenario simulation data meets the medical simulation scenario construction standard; If the predicted data simulation similarity error is greater than a preset data simulation similarity error threshold, it is determined that the predicted medical scenario simulation data does not meet the medical simulation scenario construction standard.

7. The method for building an AI-based interactive medical simulation scene according to claim 6, characterized in that: The evaluation method of the medical simulation scenario construction standard includes: The medical simulation scenario construction standard is evaluated by constructing a standard evaluation mathematical model, and the standard evaluation mathematical model is: Among them, SZP is the standard for building medical simulation scenarios; is the average value of the patient's physiological data; is the average value of medical information data; The average value of the patient feedback data; is the average value of the patient's physiological data, medical information data and patient feedback data that are erroneous during acquisition; t is the time required to acquire the patient's physiological data, medical information data and patient feedback data; α1 is the influencing factor of the average value of the patient's physiological data; α2 is the influencing factor of the average value of the medical information data; α3 is the influencing factor of the average value of the patient's feedback data; α4 is the influencing factor of the average value of the patient's physiological data, medical information data and patient feedback data that are erroneous during acquisition.

8. The method for building an AI-based interactive medical simulation scene according to claim 7, characterized in that: The image segmentation model is: Among them, I SG (x, y) is the segmented MRI image; 1 is the foreground part; 0 is the background part; I(x, y) is the pixel value of the original MRI image at position (x, y); T is the image segmentation threshold; The optimal threshold value of the image segmentation threshold T is automatically determined by the Otsu algorithm. Assuming that the total number of pixels in the MRI image is N and the gray levels are 0 and L-1, the probability of occurrence of gray levels in the image is estimated as: Among them, p(i) is the probability of occurrence of gray level; n i is the number of pixels at gray level i; For the image segmentation threshold T, the image is divided into two categories C0 and C1; the weights of categories C0 and C1 are and Among them, ω0 and ω1 are the weights of categories C0 and C1; Calculate the average value of the gray value as and Calculate the inter-class variance of pixel grayscale values ​​of different background or foreground categories; The between-class variance is: R =ω0(μ0-μ T ) 2 +ω1(μ1-μ T ) 2 ; Among them, δ R is the between-class variance; μ T is the average grayscale value of all pixels in the MRI image; Traverse all image segmentation thresholds T and select δ R The maximized image segmentation threshold T is taken as the optimal threshold.

9. An AI-based interactive medical simulation scene construction system, used to implement the AI-based interactive medical simulation scene construction method according to any one of claims 1 to 8, characterized in that: include: A data collection module, used to collect patient physiological data, medical information data and patient feedback data; A data processing module is used to pre-process the patient's physiological data and medical information data to obtain a physiological feature data set and a medical information feature data set; Fusing patient feedback data into the physiological feature dataset and the medical information feature dataset to obtain a comprehensive medical feature dataset; The model training module is used to train the medical scenario prediction model. The medical comprehensive feature dataset is input into the medical scenario prediction model to predict the medical scenario simulation data. A scenario construction evaluation module is used to process the predicted medical scenario simulation data to obtain a data simulation similarity error; compare the predicted data simulation similarity error with a preset data simulation similarity error threshold to determine whether the predicted medical scenario simulation data meets the medical simulation scenario construction standard; The simulation scenario building module processes the predicted medical scenario simulation data and dynamically generates a medical simulation scenario if the medical simulation scenario building standards are met; each module is connected through wired and / or wireless means to realize data transmission between modules.

Images