Medical training analog simulation system based on big data analysis

Abstract

The invention relates to a medical training analog simulation system based on big data analysis, in particular to the field of medical training analog simulation, the medical training analog simulation system provides an efficient, accurate and dynamically adaptive training environment through cooperative work of a plurality of modules; the data processing module ensures high quality of input data, the construction module improves the generalization ability of the model through transfer learning and multi-task learning, the dynamic optimization module improves the model prediction precision through self-supervised learning and learning rate adjustment, and the real-time feedback module monitors student operation and provides personalized tutoring; the whole system combines high-quality data, intelligent training and real-time feedback, provides accurate and personalized support for medical education and practical training, and improves operation skills and coping ability of students.

Description

Technical Field

[0001] The present invention relates to the field of medical training simulation, and more specifically, to a medical training simulation system based on big data analysis. Background Art

[0002] Existing medical training simulation systems based on big data analysis primarily rely on medical data extracted from a large number of historical case histories, using advanced algorithms to simulate patients. These systems aim to assist doctors in skill training and decision-making rehearsal by providing highly realistic virtual patients. Leveraging medical big data, the systems can generate a diverse range of simulated cases encompassing various age groups, genders, and disease types, providing doctors with a training experience that closely resembles real-world clinical scenarios. Current simulation systems apply traditional machine learning methods to image recognition, disease prediction, and treatment decision-making, which can improve doctors' diagnostic capabilities and clinical response speed to a certain extent.

[0003] However, existing medical training simulation systems still have some problems, especially in terms of model accuracy and training data quality. Due to the diversity and complexity of medical data itself, existing training data often has problems such as missing labels, mislabeling, or insufficient data volume, resulting in the trained model may not accurately reflect the diversity of diseases or the actual conditions of patients. In addition, traditional supervised learning methods often encounter problems of overfitting or underfitting when processing large-scale, multi-source medical data, and cannot effectively improve the generalization ability of the model. Furthermore, existing data cleaning and feature extraction technologies have limited effects and often cannot fully utilize the potential value of the data. Therefore, how to improve the quality of training data, optimize the accuracy of the model, and still achieve high training results when data is scarce has become the main challenge facing existing medical training simulation systems. Summary of the Invention

[0004] In response to the technical problems existing in the prior art, the present invention provides a medical training simulation system based on big data analysis. By comprehensively applying technologies such as multimodal data processing, transfer learning, multi-task learning, data enhancement and generative adversarial networks, it can provide effective support in solving data scarcity, improving data quality and optimizing model accuracy, thereby overcoming the main challenges in existing medical training simulation systems.

[0005] The technical solution of the present invention to solve the above technical problems is as follows: specifically comprising: a data processing module, a construction module, a dynamic optimization module and a real-time feedback module;

[0006] Data processing module: This module acquires multimodal medical data from medical data sources, including medical images, patient records, diagnostic reports, and time-series vital signs data. It uses a denoising convolutional neural network to remove artifacts from the medical image data. It then uses an autoencoder to reconstruct the time-series vital signs data and remove outliers. Finally, the processed medical data is sent to the construction module via an AES-256 encrypted wireless transmission protocol.

[0007] Construction module: After receiving the processed medical data, it is used as input to start building a medical training simulation model. When the amount of data is small, the model that has been pre-trained on a large medical dataset is borrowed and transferred to the new medical task using the transfer learning method. At the same time, a multi-task learning framework is constructed using the multi-task learning method to learn multiple medical tasks simultaneously through a shared network structure. After the medical training model is constructed, if the training data is insufficient to support the training of the medical training simulation model, the existing data is image enhanced to generate different variants to increase the diversity of the dataset. At the same time, the generative adversarial network technology is used to generate pathology-specific synthetic data to expand the training set, thereby obtaining a trained medical training simulation model.

[0008] Dynamic Optimization Module: Continuously optimizes medical training simulation model parameters through self-supervised learning and dynamic learning rate adjustment mechanisms. Real-time Feedback Module: Monitors student operation steps in real time. When the step sequence differs from the standard medical process by a degree d greater than 0.4, an audio-visual warning signal is triggered, and a remedial teaching video for the associated case is pushed. Based on the prediction results of the medical training simulation model, the lesion area is visualized through a gradient-weighted class activation heat map, and the operation error warning is associated with the lesion heat map. When the warning is triggered, the confidence level of the corresponding misdiagnosed area in the heat map is automatically highlighted.

[0009] In a preferred embodiment, in the data processing module, the medical data sources include open data platforms and hospitals, and the denoising convolutional neural network includes five residual blocks. Each residual block consists of two 3×3 convolutional layers, a batch normalization layer, and a ReLU activation function. Its loss function is:

[0010]

[0011] in, represents the loss function value of the denoising convolutional neural network, represents the double summation of all pixel positions (h, w) of the medical image, I(h, ω) represents the pixel value of the original medical image at position (h, w), Represents the pixel value of the medical image at position (h,w) after denoising, where H and W represent the height and width of the medical image respectively. It means taking the average value of the total pixels H·W;

[0012] The residual blocks of the denoising convolutional neural network use skip connections, and the number of convolution kernels in the last layer is 64.

[0013] In a preferred embodiment, the medical task includes disease classification, symptom regression, and pathology report generation, and they share a feature extraction layer. The feature extraction layer uses a Transformer architecture and fuses image features and text medical record features through a multi-head attention mechanism. Its multi-task loss function is:

[0014] η MTL =αη class +βη reg +γη text ;

[0015] Among them, η MTL represents the total loss value of multi-task learning, α, β, γ represent the weight coefficients of each task, η class represents the cross entropy loss function for the disease classification task, η reg represents the loss function of the symptom regression task, η text Represents the negative log-likelihood loss function for the pathology report generation task.

[0016] In a preferred embodiment, the image enhancement process specifically includes rotation, scaling, and grayscale value disturbance.

[0017] In a preferred embodiment, the generator objective function of the adversarial generative network is:

[0018]

[0019] in, Denotes the total loss function of the generator, E z,c represents the expected value of the joint distribution of the noise vector z and the pathological condition label c, G(z,c) represents the synthetic data generated by the generator based on the noise z and the pathological condition c, D(G(z,c)) represents the discrimination probability of the discriminator on the generator, log(1-D(G(z,c))) represents the adversarial loss term, λ represents the L1 regularization coefficient, and 10≤λ≤100, ||yG(z,c)||1 represents the L1 reconstruction loss.

[0020] In a preferred embodiment, the dynamic optimization module adopts a mask self-supervised learning strategy to randomly mask p% areas of the input data, and updates the medical training simulation model parameters by predicting the masked content.

[0021] In a preferred embodiment, the dynamic learning rate adjustment mechanism is specifically as follows:

[0022] When the validation set loss does not decrease for three consecutive cycles, the learning rate is adjusted to In a preferred embodiment, in the real-time feedback module, the difference d is calculated by a dynamic time warping algorithm, and the specific calculation formula is:

[0023]

[0024] Where T represents the total number of time steps of the operation step, t represents the time index of the operation step, Indicates the cumulative sum of the differences from the 1st to the Tth step, represents the student's operational feature vector at step t, Represents the standard operational feature vector of step t.

[0025] In a preferred embodiment, the heat map calculation formula is:

[0026]

[0027] Among them, H(x,y) represents the activation intensity value of (x,y) in the heat map, ReLU represents the linear rectification function, Indicates the weighted sum of the activation values ​​of K feature maps, A k represents the kth feature map, A k (x,y) represents the activation value of the kth feature map at position (x,y), ω k Represents the classification weight.

[0028] In a preferred embodiment, the real-time feedback module automatically matches the heat map display mode according to the warning type, specifically including: S1, if the warning is a misdiagnosis, the heat map focuses on the missed lesion area;

[0029] S2. If the operation times out and a warning is issued, the thermal map will mark the key anatomical structures to guide the path.

[0030] The beneficial effects of the present invention are: the medical training simulation system provides an efficient, accurate and dynamically adaptable training environment through the collaborative work of multiple modules; the data processing module ensures the high quality of input data, the construction module improves the generalization ability of the model through transfer learning and multi-task learning, the dynamic optimization module improves the model prediction accuracy through self-supervised learning and adjustment of the learning rate, and the real-time feedback module monitors student operations and provides personalized guidance; the overall system combines high-quality data, intelligent training and real-time feedback to provide accurate and personalized support for medical education and training, and improve students' operational skills and coping capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Flow chart of the method of the present invention;

[0032] Figure 2 This is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0034] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0035] In the description of this application, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art will recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in this application.

[0036] Example 1

[0037] This embodiment provides Figure 1-2 A medical training simulation system based on big data analysis is shown, which specifically includes: a data processing module, a construction module, a dynamic optimization module and a real-time feedback module;

[0038] Data processing module: This module acquires multimodal medical data from medical data sources. Multimodal medical data includes medical images, patient medical records, diagnostic reports, and time-series vital signs data. A denoising convolutional neural network is used to remove artifacts from the medical image data to retain key features. This operation can improve the accuracy of medical images in diagnosis, especially in identifying lesions. Furthermore, because time-series vital signs data often contain noise and sudden outliers, an autoencoder is used to reconstruct the time-series vital signs data and remove outliers, thereby ensuring the stability of the training data. Finally, the processed medical data is sent to the construction module via the AES-256 encrypted wireless transmission protocol. This module's design ensures data quality in practical applications, laying a solid foundation for subsequent training and model optimization.

[0039] Construction module: After receiving the processed medical data, it is used as input and starts to build a medical training simulation model. When the amount of data is small, the model that has been pre-trained on a large medical data set is borrowed and transferred to a new medical task using the transfer learning method. For example, a convolutional neural network model trained on a large amount of chest X-ray data is transferred to the medical task of lung disease detection to reduce dependence on large-scale data. At the same time, a multi-task learning framework is constructed using the multi-task learning method. By sharing the network structure, multiple medical tasks are learned at the same time to improve the generalization ability of the model. After the medical training model is built, In teaching, especially medical imaging data, there are often limited samples. If the training data is insufficient to support the training of medical training simulation models, the existing data is enhanced to generate different variants to increase the diversity of the data set. At the same time, the generative adversarial network technology is used to generate pathology-specific synthetic data to expand the training set. For practical training, it can generate diverse simulation data to help students practice and test in different situations, and then obtain a trained medical training simulation model. This module can provide diverse training data as needed to ensure that the medical training simulation system can cope with different training scenarios.

[0040] Dynamic Optimization Module: This module continuously optimizes the parameters of the medical training simulation model through self-supervised learning and a dynamic learning rate adjustment mechanism. This module can improve the simulation system's prediction accuracy and adaptability to dynamic changes, and is very effective for the ever-changing data streams in real applications.

[0041] Real-time feedback module: real-time monitoring of students' operation steps. When the difference d between the step sequence and the standard medical process is greater than 0.4, the difference threshold d>0.4 is set based on the experience of clinical experts, triggering an audio-visual warning signal and pushing a remedial teaching video of the related case. According to the prediction results of the medical training simulation model, the lesion area is visualized through the gradient weighted class activation heat map, and the operation error warning is associated with the lesion heat map. When the warning is triggered, the confidence level of the corresponding misdiagnosed area in the heat map is automatically highlighted. This module design can improve the user experience, especially in education and practical training environments.

[0042] In this embodiment, the data processing module is specifically described. The medical data sources include the open data platform and hospitals. The denoising convolutional neural network contains 5 residual blocks. Each residual block consists of two 3×3 convolutional layers, a batch normalization layer, and a ReLU activation function. Its loss function is:

[0043]

[0044] in, Represents the loss function value of the denoising convolutional neural network, which is used to measure the denoised image The difference from the original medical image I. The smaller the value, the better the denoising effect. It represents the double summation of all pixel positions (h, w) of the medical image, covering all pixels from the 1st row and the 1st column to the Hth row and the Wth column. I(h, ω) represents the pixel value of the original medical image at the position (h, w). Represents the pixel value of the denoised medical image at position (h, w), that is, the output of the denoising convolutional neural network. H and W represent the height and width of the medical image, respectively. Indicates taking the average value of the total pixels H·W; skip connections are used between the residual blocks of the denoising convolutional neural network, and the number of convolution kernels in the last layer is 64;

[0045] Before using the autoencoder, it is necessary to first preprocess the time series life feature data. The specific steps include: S1, using linear interpolation of the interpolation method to process missing data points;

[0046] S2. Normalize the data so that the autoencoder can effectively process input data of different dimensions;

[0047] S3, Sliding Window: For the time series of vital signs data, the sliding window method is used to segment the data to facilitate model training. The vital signs data of the time series in each window will be used as an input sample of the autoencoder;

[0048] Secondly, the goal of the autoencoder is to reconstruct the input data. The input data here refers to the vital signs data of the time series. Therefore, the difference between the input data and the reconstructed data (i.e., the reconstruction error) can be used to detect outliers. In this application, the mean square error between the input data and the reconstructed data at each time point is calculated. If the reconstruction error at a certain moment is greater than the preset threshold, the point is considered to be an outlier. The calculation formula of the mean square error is:

[0049]

[0050] Among them, x i Represents the raw life characteristic data of the time series, Represents the reconstructed output data, and N represents the number of data points. At the same time, in order to perform outlier detection more flexibly and accurately, this application designs an adaptive threshold formula based on the distribution of reconstruction error. This formula will dynamically adjust the threshold according to the error distribution of training data, thereby improving the accuracy of the model. Specifically: the reconstruction error follows a normal distribution, and a threshold is dynamically set by calculating the mean and standard deviation of the error. The formula for the adaptive threshold is:

[0051] Threshold=μ+ασ;

[0052] Among them, μ represents the mean reconstruction error of the training data, σ represents the standard deviation of the reconstruction error, and α represents an adjustable hyperparameter that controls the detection sensitivity. In addition, in this formula, the larger the α value, the higher the threshold and the fewer outliers are detected; the smaller the α value, the lower the threshold and the more outliers are detected. The most appropriate value of α is selected through cross-validation.

[0053] In this embodiment, the construction module needs to be specifically explained. The medical tasks include disease classification, symptom regression, and pathology report generation, and they share a feature extraction layer. The feature extraction layer adopts the Transformer architecture and integrates image features and text medical record features through a multi-head attention mechanism. Its multi-task loss function is:

[0054] η MTL =αη class +βη reg +γη text ;

[0055] Among them, η MTL represents the total loss value of multi-task learning, α, β, γ represent the weight coefficients of each task, which are used to adjust the contribution ratio of different tasks to the total loss and satisfy α+β+γ1 to ensure weight normalization and avoid the training process being dominated by a single task, η class Represents the cross entropy loss function of the disease classification task. The specific calculation formula is:

[0056]

[0057] Among them, y c Represents the one-hot encoding of the true label, where if the disease category is c, then y c =1, otherwise 0, p c represents the disease category probability predicted by the medical training simulation model;

[0058] η reg The loss function for the symptom regression task is represented by the mean square error in this application. The specific calculation formula is:

[0059]

[0060] Among them, s i represents the true symptom score of the i-th sample, such as dyspnea index, pain level, represents the symptom score predicted by the medical training simulation model;

[0061] η text represents the negative log-likelihood loss function for the pathology report generation task. The specific calculation formula is:

[0062]

[0063] Among them, ω t represents the tth word in the pathology report, P(ω t |ω 1:t-1 , X) represents the medical training simulation model according to the preceding word ω 1:t-1 And the probability of generating the current word from the input data X;

[0064] Image enhancement processing specifically includes rotation (θ∈[-15°,15°]), scaling (s∈[0.9,1.1]) and grayscale value perturbation (Δ∈[-10,10]); the generator objective function of the adversarial generative network is:

[0065]

[0066] in, Represents the total loss function of the generator, which is used to guide the optimization of the generator so that the synthetic data it generates can deceive the discriminator and be close to the real data distribution. z,crepresents the expected value of the joint distribution of the noise vector z and the pathological condition label c, G(z,c) represents the synthetic data generated by the generator based on the noise z and the pathological condition c, D(G(z,c)) represents the probability of the discriminator to distinguish the generator from the noise z and the pathological condition c, where D(·)∈[0,1] represents the confidence of the discriminator that the data is true, log(1-D(G(z,c))) represents the adversarial loss term, the purpose of which is to force the discriminator's probability of distinguishing the synthetic data D(G(z,c)) to approach 0 by minimizing this term, that is, the discriminator has difficulty distinguishing between real and synthetic data, λ represents the L1 regularization coefficient, and 10≤λ≤100, the purpose of which is to prevent the generator from over-optimizing the adversarial loss and ignoring the authenticity of the data, ||yG(z,c)||1 represents the L1 reconstruction loss, which is used to calculate the pixel-level absolute difference between the generated data G(z,c) and the real data y, where

[0067] In this embodiment, the dynamic optimization module specifically needs to be explained. Since medical data is constantly updated, the medical training simulation model also needs to continuously adapt to the new data. Therefore, the dynamic optimization module adopts a masked self-supervised learning strategy to randomly mask a p% area of ​​the input data, where the value range of p is: 10≤p≤30, and updates the parameters of the medical training simulation model by predicting the masked content;

[0068] The dynamic learning rate adjustment mechanism is as follows:

[0069] When the validation set loss does not decrease for three consecutive cycles, the learning rate is adjusted to Attenuation, where n represents the number of attenuation times. In this embodiment, it is specifically necessary to explain the real-time feedback module, and the difference d is calculated by the dynamic time warping algorithm. The specific calculation formula is:

[0070]

[0071] Among them, T represents the total number of time steps of the operation steps, t represents the time index of the operation step, from t = 1 to t = T, it represents the t-th step, Indicates the cumulative sum of the differences from the 1st to the Tth step, The feature vector representing the student's operation at step t contains operation data in multiple dimensions, where typical features include tool selection, operation position, operation force, and operation duration; The standard operation feature vector representing step t, which is defined by expert experience or clinical guidelines;

[0072] The heat map calculation formula is:

[0073]

[0074] Among them, H(x,y) represents the activation intensity value of (x,y) in the heat map, and also represents the importance of the location to the model prediction result. The higher the value, that is, the warmer the color, the greater the contribution of the lesion area to the model decision. ReLU represents the linear rectification function, which is positioned as ReLU(z)=max(0,z). The purpose is to filter out negative values ​​and retain only positive activations, ensuring that the heat map only highlights areas that have positive contributions to the classification. Indicates the weighted sum of the activation values ​​of K feature maps, A k represents the kth feature map, A k (x,y) represents the activation value of the kth feature map at position (x,y), ω k Indicates classification weight; the real-time feedback module automatically matches the heat map display mode according to the warning type, including:

[0075] S1. If a misdiagnosis warning is issued, the heat map will focus on the missed lesion area;

[0076] S2. If the operation times out and a warning is issued, the thermal map will mark the key anatomical structures to guide the path.

[0077] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0078] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the functions described in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0079] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0081] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0082] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A medical training simulation system based on big data analysis, characterized in that: Specifically include: Data processing module, construction module, dynamic optimization module and real-time feedback module; The data processing module acquires multimodal medical data from medical data sources, including medical images, patient records, diagnostic reports, and time-series vital sign data. It uses a denoising convolutional neural network to remove artifacts from the medical image data. It then uses an autoencoder to reconstruct the time-series vital sign data and remove outliers. The processed medical data is then sent to the construction module via an AES-256 encrypted wireless transmission protocol. Construction module: After receiving the processed medical data, it is used as input to start building a medical training simulation model. When the amount of data is small, the model that has been pre-trained on a large medical dataset is borrowed and transferred to the new medical task using the transfer learning method. At the same time, a multi-task learning framework is constructed using the multi-task learning method to learn multiple medical tasks simultaneously through a shared network structure. After the medical training model is constructed, if the training data is insufficient to support the training of the medical training simulation model, the existing data is image enhanced to generate different variants to increase the diversity of the dataset. At the same time, the generative adversarial network technology is used to generate pathology-specific synthetic data to expand the training set, thereby obtaining a trained medical training simulation model. Dynamic optimization module: Continuously optimizes the parameters of the medical training simulation model through self-supervised learning and dynamic learning rate adjustment mechanism; Real-time feedback module: Real-time monitoring of students' operation steps. When the difference d between the step sequence and the standard medical process is greater than 0.4, an audio-visual warning signal is triggered, and a remedial teaching video of the related case is pushed. According to the prediction results of the medical training simulation model, the lesion area is visualized through the gradient weighted class activation heat map, and the operation error warning is associated with the lesion heat map. When the warning is triggered, the confidence level of the corresponding misdiagnosed area in the heat map is automatically highlighted.

2. The medical training simulation system based on big data analysis according to claim 1 is characterized by: In the data processing module, the medical data sources include open data platforms and hospitals. The denoising convolutional neural network contains five residual blocks, each of which is composed of two 3×3 convolutional layers, a batch normalization layer, and a ReLU activation function. Its loss function is: in, represents the loss function value of the denoising convolutional neural network, represents the double summation of all pixel positions (h, w) of the medical image, I(h, ω) represents the pixel value of the original medical image at position (h, w), Represents the pixel value of the medical image at position (h,w) after denoising, where H and W represent the height and width of the medical image respectively. It means taking the average value of the total pixels H·W; The residual blocks of the denoising convolutional neural network use skip connections, and the number of convolution kernels in the last layer is 64.

3. The medical training simulation system based on big data analysis according to claim 1 is characterized by: The medical tasks include disease classification, symptom regression, and pathology report generation, and share a feature extraction layer. The feature extraction layer uses a Transformer architecture and fuses image features with text medical record features through a multi-head attention mechanism. Its multi-task loss function is: or MTL =ai class +ve reg +earth text ; Among them, η MTL represents the total loss value of multi-task learning, α, β, γ represent the weight coefficients of each task, η class represents the cross entropy loss function for the disease classification task, η reg represents the loss function of the symptom regression task, η text Represents the negative log-likelihood loss function for the pathology report generation task.

4. The medical training simulation system based on big data analysis according to claim 1 is characterized by: The image enhancement processing specifically includes rotation, scaling and grayscale value disturbance.

5. The medical training simulation system based on big data analysis according to claim 1 is characterized by: The generator objective function of the adversarial generative network is: in, Denotes the total loss function of the generator, E z,c represents the expected value of the joint distribution of the noise vector z and the pathological condition label c, G(z,c) represents the synthetic data generated by the generator based on the noise z and the pathological condition c, D(G(z,c)) represents the discrimination probability of the discriminator on the generator, log(1-D(G(z,c))) represents the adversarial loss term, λ represents the L1 regularization coefficient, and 10≤λ≤100, ||yG(z,c)||1 represents the L1 reconstruction loss.

6. The medical training simulation system based on big data analysis according to claim 1 is characterized by: The dynamic optimization module adopts a mask self-supervised learning strategy to randomly mask p% areas of the input data and update the medical training simulation model parameters by predicting the masked content.

7. The medical training simulation system based on big data analysis according to claim 1 is characterized by: The dynamic learning rate adjustment mechanism is specifically as follows: When the validation set loss does not decrease for three consecutive cycles, the learning rate is set to θ = θ0 × 0.5 n Decay, where n is the number of decays.

8. The medical training simulation system based on big data analysis according to claim 1 is characterized by: In the real-time feedback module, the difference d is calculated using the dynamic time warping algorithm. The specific calculation formula is: Where T represents the total number of time steps of the operation step, t represents the time index of the operation step, Indicates the cumulative sum of the differences from the 1st to the Tth step, represents the student's operational feature vector at step t, Represents the standard operational feature vector of step t.

9. The medical training simulation system based on big data analysis according to claim 1 is characterized by: The heat map calculation formula is: Among them, H(x,y) represents the activation intensity value of (x,y) in the heat map, ReLU represents the linear rectification function, Indicates the weighted sum of the activation values ​​of K feature maps, A k represents the kth feature map, A k (x,y) represents the activation value of the kth feature map at position (x,y), ω k Represents the classification weight.

10. The medical training simulation system based on big data analysis according to claim 1 is characterized by: The real-time feedback module automatically matches the heat map display mode according to the warning type, specifically including: S1. If a misdiagnosis warning is issued, the heat map will focus on the missed lesion area; S2. If the operation times out and a warning is issued, the thermal map will mark the key anatomical structures to guide the path.

Images