Orthognathic surgery treatment suggestion generation system, method, equipment and medium

Abstract

The invention belongs to the technical field of artificial intelligence and biomedicine, and provides an orthognathic surgery treatment suggestion generation system, method, equipment and medium in order to assist doctors in generating orthognathic surgery suggestions. Inputting the preprocessed orthognathic medical data of the patient into a prediction model to obtain orthognathic surgical treatment suggestion auxiliary information of the patient; wherein the prediction model is obtained by training a Transform network model; in the training process of the network model, feed-forward network parameters of the r network model are adjusted based on a bidirectional coupling grey wolf optimization algorithm, and in the process of optimizing the feed-forward network parameters by the bidirectional coupling grey wolf optimization algorithm, a search strategy is dynamically adjusted based on the importance of orthognathic features; and the performance of the network model in processing high-dimensional, nonlinear and dynamically changing orthognathic data is improved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence and biomedicine related technologies, and in particular, relates to a system, method, device and medium for generating orthognathic surgery treatment suggestions. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous development of the medical field, especially orthognathic surgery treatment technology, doctors' diagnosis and treatment decisions are increasingly dependent on the analysis of large amounts of medical data. Traditional orthognathic surgery treatment plans often rely on the doctor's experience and judgment, but when faced with a large number of patients and complex medical data, empirical decisions are prone to deviations and inaccuracies, especially when there are large individual differences among patients, and the personalized needs of treatment plans are more prominent. Therefore, how to improve the accuracy, standardization and personalization of treatment plans through technical means has become an urgent problem to be solved.

[0004] Orthognathic surgery is a complex surgical procedure that involves extremely complex medical data, including multi-dimensional information such as the patient's clinical data, imaging examinations, genetic factors, etc., which are often highly individual. Existing technologies mainly rely on doctors' subjective experience and basic diagnosis and treatment standards, but due to the lack of sufficient standardized and personalized intelligent decision-making tools, doctors often face a huge workload when formulating treatment plans for patients, and the accuracy and effectiveness of the plans are difficult to fully guarantee.

[0005] There are at least the following problems in the generation of treatment recommendations for existing orthognathic surgery: First, medical data in the field of orthognathic surgery is usually limited, especially the relevant data for generating personalized treatment recommendations is relatively scarce, resulting in the inability of traditional models to fully learn high-quality personalized treatment plans during the training process. Insufficient data samples often lead to poor generalization ability of the model and prone to overfitting problems.

[0006] Second, traditional orthognathic surgery treatment recommendations often rely on the doctor's personal experience, which may involve subjective biases, and the treatment plans are less standardized, making it difficult to provide accurate personalized treatment recommendations for each patient.

[0007] 3. In traditional feedforward network training, the optimization algorithm mostly uses gradient descent. However, under high-dimensional data, gradient descent may face the risk of gradient vanishing or falling into a local optimal solution, resulting in low training efficiency.

[0008] 4. Traditional medical data processing often treats data as static tables and lacks in-depth understanding and expression of complex semantic relationships in the data, resulting in low data utilization. Summary of the invention

[0009] In order to overcome the deficiencies of the above-mentioned prior art, the present invention provides a system, method, device and medium for generating orthognathic surgery treatment recommendations, which optimizes the feedforward network based on the bidirectional coupling grey wolf optimization algorithm, improves the performance of the Transformer network model in processing high-dimensional, nonlinear and dynamically changing orthognathic data, and enables the generated surgical plan to assist doctors in their diagnosis and treatment decisions.

[0010] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a system for generating orthognathic surgery treatment recommendations, the system comprising: An acquisition module, used for acquiring and preprocessing the orthognathic medical data of the patient; A suggestion generation module, used for inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment suggestion for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

[0011] In a second aspect, the present invention provides a method for generating orthognathic surgery treatment recommendations, the method comprising: Obtain the patient's orthognathic medical data and perform preprocessing; Inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment recommendation for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

[0012] In a third aspect, the present invention provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the following steps are performed: Obtain the patient's orthognathic medical data and perform preprocessing; Inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment recommendation for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

[0013] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the following steps are performed: Obtain the patient's orthognathic medical data and perform preprocessing; Inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment recommendation for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

[0014] One or more of the above technical solutions have the following beneficial effects: In the present invention, the feedforward network is optimized based on the bidirectional coupling gray wolf optimization algorithm. By dynamically adjusting the search range and intensity, the convergence speed of the Transformer network model is accelerated, the stability and adaptability of the feedforward network are improved, and the performance of the Transformer network model in processing high-dimensional, nonlinear and dynamically changing orthognathic data is improved.

[0015] In the present invention, a generative adversarial network based on latent space decoupling is used to expand training samples, which solves the problem of insufficient samples in traditional training. By generating diverse and personalized samples, the diversity of the data set is enhanced, and the generalization ability and accuracy of the prediction model are improved. A large amount of orthognathic medical data is used for training, so that the Transformer network model can automatically generate auxiliary treatment plans and make customized suggestions based on the individual differences of patients.

[0016] In the present invention, by structured processing of orthognathic medical data and converting it into a natural language format, a more accurate and standardized data input method is provided, which can be conveniently input into the Transformer network model for training, and the word segmentation and word embedding processing of the data enhances the data representation ability.

[0017] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0019] Figure 1 A block diagram of a system for generating orthognathic surgery treatment recommendations in Embodiment 1 of the present invention; Figure 2 Schematic diagram of the Transformer network structure in Example 1 of the present invention. DETAILED DESCRIPTION

[0020] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0021] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0022] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0023] Embodiment 1 like Figure 1 As shown, this embodiment discloses a system for generating orthognathic surgery treatment suggestions, including: An acquisition module, used for acquiring and preprocessing the orthognathic medical data of the patient; A suggestion generation module, used for inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment suggestion for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

[0024] In this embodiment, when training a prediction model for personalized treatment of orthognathic surgery based on a Transformer network, the goal is to use a large amount of patient information, medical orthognathic diagnosis, and treatment plans to train the prediction model so that it can generate personalized treatment recommendations based on the input patient information. The collected orthognathic medical training data is structured data containing patient information, diagnosis, treatment plans, etc. The sources of orthognathic medical training data collection mainly include medical institutions, research literature, doctor records, and other channels, and the orthognathic medical data is stored in JSON format. In one embodiment, an example of orthognathic medical training data is as follows: {"Patient Information":{"Patient ID":"1001","Age":25,"Gender":"Male","Medical History":["Orthodontic treatment (dental correction)","None"],"Symptoms":["Protruding jaw","Malocclusion"]},"Diagnosis":{"Orthognathic condition":"Class III malocclusion","CT scan results":"CT showed mild overgrowth of the mandible and slight facial asymmetry","X-ray results":"X-ray The patient showed malocclusion and needed surgery to adjust the relationship between the upper and lower jaws."},"Treatment plan":{"Recommended surgery":"Bimaxillary surgery (maxillary LeFort I + mandibular double-segment osteotomy)","Postoperative care":"Regular follow-up is required within 3 months after surgery, and orthodontic treatment is required for 6 months after surgery"},"Doctor's advice":["The patient is suitable for bimaxillary surgery and needs 6 months of orthodontic preparation before surgery.","Pay attention to the control of facial swelling after surgery and follow-up regularly."]}.

[0025] The properties of orthognathic data include the following: Patient information: includes the patient's basic information (such as age, gender, etc.), medical history and symptoms. The medical history field records whether the patient has had other relevant treatment experiences, and the symptom field records the main problems currently faced by the patient.

[0026] Diagnostic results: including diagnosed orthognathic conditions such as Class II, Class III, etc., as well as detailed results obtained through imaging examinations (CT, X-rays).

[0027] Treatment plan: This includes the recommended type of surgery, such as single-jaw surgery or double-jaw surgery, as well as the postoperative care plan. This part is key information for the predictive model to learn how to generate treatment recommendations.

[0028] Doctor's Advice: Provides specific personalized advice from your doctor based on diagnosis and treatment options.

[0029] Since the Transformer network processes textual orthognathic medical data, it is necessary to convert the structured JSON orthognathic medical data into a natural language format, that is, convert the fields of different parts into suitable sentence forms, and then splice them into text input. In this embodiment, the orthognathic medical data in JSON format are translated into sentence forms in sequence, and each field becomes a descriptive natural language sentence.

[0030] Furthermore, all textual orthognathic medical data needs to be processed by word segmentation. Specifically, existing Chinese word segmentation tools such as Jieba word segmentation and BERT word segmentation can be used. The word segmentation process divides the text into smaller units such as words or characters, so that the model can understand and process these inputs, such as: Input text: "Patient ID is 1001, age is 25, gender is male, medical history includes orthodontic treatment, current symptoms are protruding jaw and malocclusion." After word segmentation, it becomes: ["Patient","ID","Yes","1001","","Age","25","Years","","Sex","Male","","Medical History","Includes","Dental","Orthodontic","","Current","Symptoms","There are","Jaw","Protrusion","and","Occlusion","Male"].

[0031] Furthermore, when the orthognathic medical data after word segmentation is input into the Transformer network, it is further converted into word embedding. Each word or subword is mapped to a vector. The model understands the semantic relationship of the text through these vectors. The specific input format is as follows: Input sequence: The input received by the Transformer network is a sequence of word vectors after word segmentation; Position encoding: Since the Transformer network does not have the temporal nature of a recurrent neural network, position encoding is needed to mark the position of each word in the sequence; Batch processing: Combine multiple orthognathic medical data into one batch and input the model to improve training efficiency.

[0032] The Transformer network consists of two parts: an encoder and a decoder. The encoder maps the input features into a series of context-aware vector representations, and the decoder gradually decodes these vector representations into target features. Its architecture is as follows: Figure 2 shown.

[0033] Specifically, the encoder part is composed of The encoder is composed of 2 identical encoding layers, where each encoding layer consists of two sub-layers: a multi-head attention mechanism sub-layer and a feed-forward network sub-layer. The encoder adds residual connections and normalization operations after each sub-layer.

[0034] The multi-head attention mechanism sublayer is implemented through the multi-head self-attention mechanism, which is the core module of the encoder. It is responsible for learning the importance of each position from the input features and combining the information of all positions to obtain a global context representation, thereby improving the performance and generalization ability of the model. Specifically, the multi-head self-attention mechanism is implemented as follows: the input features pass through three different linear transformation layers to generate "query", "key" and "value" vectors respectively, and the linear transformation is achieved by multiplying the weight matrix; further, the generated query, key and value vectors are divided into multiple smaller parts, namely "heads", so that the model can process multiple different representation subspaces in parallel, thereby capturing the diversity information in the input orthognathic medical data; further, for each head, the dot product of the query and key vectors is calculated respectively to obtain the attention score, which reflects the relative importance of different positions in the sequence, and in order to make the numerical value of the attention score stable, the score is divided by a scaling factor, which can be the square root of the key vector dimension; further, the Softmax function is used to process the attention score to obtain the attention weight of each position. The Softmax function ensures that all weights add up to 1, so that it can be processed as a probability distribution, expressed as: (1) In the formula, For the The attention weight of each head, For the The query vector of the head, For the The key vector of the head, For the The transpose of the key vector of the head, is the dimension of the input features, is the number of attention heads, For the The value vector of the heads; for function.

[0035] Furthermore, the obtained attention weights are used to perform weighted summation on the value vector to obtain the output vector; the output vectors of all heads are concatenated together, and then the output of the final multi-head self-attention layer is generated through a linear transformation layer.

[0036] The feedforward network sublayer consists of two fully connected layers and an activation function, which is used to perform nonlinear transformation on the vector at each position. In this sublayer, the input vector first passes through the first fully connected layer for linear transformation, then passes through a ReLU activation function for nonlinear transformation, and finally passes through the second fully connected layer for linear transformation to improve the expressiveness and generalization ability of the model, so that the model can better handle complex input orthognathic medical data.

[0037] Specifically, the decoder consists of The decoder layer is composed of the same decoding layers. Unlike the encoding layer, each decoding layer has an additional multi-head attention mechanism sublayer, that is, the decoding layer contains two multi-head attention mechanism sublayers and one feedforward network mechanism sublayer. In addition, the second multi-head attention mechanism sublayer of the decoding layer is implemented by the multi-head cross attention mechanism. The multi-head cross attention mechanism is a mechanism for information exchange between the encoder and the decoder. It takes the output features of the encoder and the input features of the decoder as input, and uses the attention mechanism to perform cross-attention calculations on the input, thereby obtaining a set of weighted decoder output features.

[0038] It should be noted that in the encoder and decoder of the Transformer network, the output of each sub-layer is the result of layer normalization and residual connection, expressed as: (2) In the formula, is the layer normalization function, represents the output of the residual connected sublayer, Represents the input of the sublayer; is the normalized output result.

[0039] The feedforward network of the Transformer network is a neural network structure composed of neurons, which plays a role in enhancing the nonlinear processing ability of the model. The neural network structure includes weights and bias parameters. In order to avoid the gradient vanishing and gradient descent phenomena caused by the traditional neural network using the gradient descent method to update parameters, and to avoid the phenomenon that the parameters fall into the local optimal solution after the training is completed, this embodiment adopts a bidirectionally coupled gray wolf optimization algorithm to train the feedforward network during the training process of the feedforward network. On the basis of the traditional gray wolf algorithm, this embodiment adopts a dynamic adjustment strategy so that the algorithm can adaptively adjust the search range and intensity according to the local optimal solution obtained during the iteration process, thereby accelerating the convergence speed and improving the algorithm stability, and improving the data adaptive extraction capability of the feedforward network in complex feedforward orthognathic medical data, especially when processing high-dimensional, nonlinear and dynamically changing feedforward network orthognathic medical data, it can more accurately capture key features.

[0040] Specifically, the training process of the feedforward network algorithm based on bidirectional coupled gray wolf optimization is as follows: Step 11: Initialize the weights and biases of the feedforward network, as well as the positions of each individual in the gray wolf optimization algorithm, to ensure the diversity of the starting points.

[0041] In one embodiment, the initialization method is expressed as: (3) (4) In the formula, and They represent the first The weights and biases of the layers, Indicates compliance with a specific distribution; It means the mean is 0 and the variance is Normal distribution of Initialize the variance for the parameters of the feedforward network; is a normal distribution.

[0042] Step 12: Perform a bidirectional coupled search of the Gray Wolf Algorithm. Bidirectional coupled search means that in each iteration of the Gray Wolf Algorithm, the algorithm is divided into two directions for search, including leader wolf search and tail wolf push; leader wolf search selects the leader wolf, which is the optimal solution currently found, and updates the positions of other wolves according to the position of the leader wolf to explore possible better areas in the solution space; tail wolf push pushes the search from the edge of the solution space, which is near the worst solution, to avoid local optimality and increase the coverage of the solution space.

[0043] Based on the bidirectional coupling search strategy, the position update method of individuals in the gray wolf optimization algorithm is as follows: (5) (6) (7) (8) (9) In the formula, It is In the iteration The position of the gray wolf, that is, the position of the The location of the gray wolf; It is In the iteration The position of the gray wolf, that is, the position of the The location of the gray wolf; is the position of the leader wolf, representing the optimal solution currently searched; The search capability parameters for the leader wolf; is the position of the tail wolf, representing the worst solution currently searched; is the search ability parameter of the tail wolf; is the first coefficient that controls the search range; is the second coefficient that controls the search range; is the first random number in the range [0, 1]; is the second random number in the range [0, 1]; is the linear attenuation coefficient, which decreases linearly from 2 to 0 during the iteration process to adjust the balance between exploration and development; is the first dynamic angle parameter; is the second dynamic angle parameter.

[0044] In one embodiment, the first dynamic angle parameter and the second dynamic angle parameter dynamically adjust their specific values ​​according to the change of the number of iterations to control the dynamic change of the search behavior and improve the adaptability of the algorithm at different stages. The calculation method is expressed as: (10) (11) in, is the first time dependent regulation function, is the second time-dependent adjustment function; For the The iteration based on A factor that dynamically calculates the importance of each feature.

[0045] Furthermore, the first time-dependent adjustment function and the second time-dependent adjustment function are dynamically calculated according to the change of the number of iterations, so that the search behavior gradually shifts from global search to local search as the iteration process proceeds, achieving more accurate optimization, which is expressed as: (12) (13) in, is the current training iteration number of the feedforward network; is the parameter decay coefficient of the feedforward network, is the total number of training iterations of the feedforward network.

[0046] Step 13: In order to improve the performance of the bidirectional coupled gray wolf optimization algorithm in feedforward network training, a dynamic influence factor is used to adaptively adjust the influence of each feature on the final optimization target according to the importance of the feature, so as to solve the shortcomings of the traditional gray wolf algorithm in processing large-scale orthognathic medical data with features of different importance. It can not only adjust the search strategy according to the dynamic changes of each feature, but also strengthen the local search intensity of key features while maintaining the global search capability. Specifically, the calculation method of the factor for dynamic calculation of feature importance is expressed as: (14) In the formula, The output of the feedforward network is The feature is in the feed-forward network The variance in iterations; The output of the feedforward network is The feature is in the feed-forward network The variance in iterations; It is a preset feature importance adjustment parameter used to control the speed of exponential decay to ensure that the importance of the feature can be appropriately amplified or reduced; is the number of eigenvalues ​​of the sample.

[0047] Step 14: Adjust the distance between the leading wolf and the trailing wolf to enhance the flexibility of the algorithm and its ability to cope with complex orthognathic data.

[0048] Specifically, by adjusting the reduction speed of the linear attenuation coefficient, and adjusting the first coefficient controlling the search range and the second coefficient controlling the search range, the search range and parameters are automatically adjusted, and the search strategy is dynamically adjusted according to the quality of the current solution to improve the convergence speed and the quality of the solution. The adjustment method is expressed as: (15) (16) (17) (18) In the formula, For the The linear decay coefficient of the iteration; is the fitness difference between the leader and the last wolf. The fitness difference is obtained by calculating the difference between the fitness values ​​of the feedforward network corresponding to the leader and the last wolf. The fitness value is calculated by the fitness function. is the fitness function of the feedforward network, specifically the inverse of the cross entropy loss function; is the attenuation adjustment factor; It is The first coefficient of the control search range for the iteration; It is The first coefficient of the control search range for the iteration; It is A second coefficient for controlling the search range for the iteration; It is A second coefficient for controlling the search range for the iteration; is the fitness difference between the leader and the last wolf in the previous iteration; is a smoothing factor used to balance the weight of past information and current changes. Preferably, Set to 2.

[0049] Step 15: Combined with the parameters of the feedforward network corresponding to the gray wolf position of the gray wolf algorithm, the parameters of the feedforward network are updated using the gradient descent method. The update method is expressed as: (19) (20) In the formula, is the learning rate of the feedforward network gradient descent, It is a feed-forward network. The weight of the iteration; It is a feed-forward network. The weight of the iteration; It is a feed-forward network. The bias of the iteration; It is a feed-forward network. The bias of the iteration; is the loss function of the feedforward network, specifically the cross entropy loss function; is the symbol of partial derivative; For the The weight function of the feedforward network corresponding to the gray wolf of the iteration is obtained The average value of the weight parameters of the feedforward network corresponding to all gray wolf individuals in the iteration; For the The bias function of the feedforward network corresponding to the gray wolf of the iteration is obtained The average value of the bias parameters of the feed-forward network corresponding to all gray wolf individuals in the iteration.

[0050] Step 16: Repeat the above steps 11 to 15 until the preset stop iteration condition is met, indicating that the model training is completed.

[0051] In one embodiment, the preset condition for stopping iteration is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0052] Because the collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model; for the sequence composed of word vectors after word segmentation received by the Transformer network, this embodiment performs sample expansion, specifically using a generative adversarial network algorithm based on latent space decoupling to generate samples, thereby achieving data expansion.

[0053] The generative adversarial network is a deep learning algorithm that realizes data generation through the game process of the generator and the discriminator. Based on the traditional generative adversarial network, this embodiment adopts a latent space decoupling strategy to separate the latent space, and adopts a reverse adversarial strategy to optimize the data generation process, thereby enhancing the diversity and personalized characteristics of the generated data.

[0054] Specifically, the training process of the generative adversarial network algorithm based on latent space decoupling is as follows: Step 21: Initialize the generator and discriminator, assuming the generator is , the discriminator is , initialized using the standard neural network initialization strategy.

[0055] Generator Receive random noise from the latent space , and generate pseudo data similar to the real data distribution , is the dimension of random noise.

[0056] Discriminator The task is to distinguish the generated data from the real data and output the probability of discriminating the real data and the probability of discriminating the generated data , is the discriminator function, is the real data input to the generative adversarial network.

[0057] Step 22: A decoupling mechanism is used in the latent space of the generator to decompose the latent space into multiple subspaces, each of which independently learns a specific latent feature such as anatomical features, treatment plan parameters, etc.

[0058] Specifically, the generator processes the potential features of each dimension through multiple decoupled sub-networks. These sub-networks independently generate specific sub-data from different feature spaces. The decoupled features in the latent space are expressed as: (twenty one) In the formula, is the potential space subspace features; is the potential space subspace features; is the potential space subspace features; is the potential space subspace features; is the number of features in the latent space.

[0059] Furthermore, the generator Receive the decoupled latent vector and output the corresponding pseudo data, expressed as: (twenty two) In the formula, It is the mapping function of the generator, responsible for mapping the potential features to the data space; is a generator function.

[0060] In one embodiment, in order to ensure that the contribution of each subspace to the final generated data is independent and complex, the mapping function of the generator is defined as: (twenty three) In the formula, It is The weight matrix corresponding to the subspace; It is a function that nonlinearly maps the subspace features of the latent space; It is the mapping function bias term, which is used to adjust the benchmark of the generated result.

[0061] Furthermore, for each subspace, the mapping function is defined as a neural network model with nonlinear activation, expressed as: (twenty four) In the formula, is the ReLU activation function, It is The transpose of the weight matrix corresponding to the subspace, It is The offset corresponding to the subspace.

[0062] Step 23: In the adversarial training between the generator and the discriminator, the reverse adversarial strategy is adopted. The generator not only needs to generate sufficiently realistic data to deceive the discriminator, but also needs to optimize the diversity and personalized features of the generated data through the reverse loss. The reverse loss function adjusts the gradient of the generator to encourage more diverse feature generation during the generation process and avoid mode collapse during the generation process. This allows the generator to not only generate high-quality samples but also ensure data diversity. The calculation method of the reverse adversarial loss function is expressed as: (25) In the formula, To control the hyperparameters of the diversity constraint, Represents the variance of the generated samples in the latent space; is the inverse adversarial loss function, which is also the loss function of the generator; express expectations; Indicates compliance with a specific distribution; represents the underlying distribution of random noise; is the discriminator function.

[0063] Furthermore, the variance of the generated samples in the latent space is used to improve the diversity of the generated data, and the calculation method is expressed as: (26) In the formula, For the Generate samples; is the first input to the generator A random noise; is the mean of the generated samples; is the number of samples input into the generative adversarial network in the current batch, is a generator function.

[0064] Furthermore, the calculation method of the mean of the generated samples is expressed as: (27) By adopting variance constraints, the diversity of generated data can be controlled more accurately, ensuring that the generated data is not only real but also has sufficient personalized features.

[0065] Step 24: The discriminator receives real data and generated data as input in each round of training. Its task is to distinguish the difference between the two and update the weights through back propagation. In order to enhance the discriminator's discriminative ability and promote the generation quality of the generator, the discriminator loss adopts an entropy-based regularization term to improve the discriminator's ability to recognize samples. The discriminator's loss function is expressed as: (28) In the formula, is the loss function of the discriminator; is the distribution of real data; is a hyperparameter of entropy regularization; For real data.

[0066] Step 25: The training process of the generator is relatively independent of the training process of the discriminator. During the adversarial training process, the training update method for the generator and the discriminator is expressed as: (29) (30) In the formula, For the The generator of iterations, For the The generator of iterations; For the The discriminator of the iteration, For the The discriminator of the iteration; is the learning rate of the generator; is the learning rate of the discriminator; is the gradient of the generator loss function with respect to the generator parameters; is the gradient of the discriminator loss function with respect to the discriminator parameters.

[0067] In one embodiment, in order to improve the convergence speed and stability of training, an adaptive learning rate update mechanism is adopted in the training process, and the learning rate update method of the generator and the discriminator is expressed as: (31) (32) In the formula, For the The learning rate of the generator for this iteration; For the The learning rate of the generator for each iteration; for No. The learning rate of the discriminator at this iteration; For the The learning rate of the discriminator at this iteration; represents the mean squared error of the gradient of the generator loss function with respect to the generator parameters; represents the mean squared error of the gradient of the discriminator loss function with respect to the generator parameters; is a small constant that prevents division by zero errors. Preferably, Set to 0.001.

[0068] Step 26: Repeat the above steps 21 to 25 until the preset stop iteration condition is met, which means that the training of the generative adversarial network model is completed.

[0069] In one embodiment, the preset condition for stopping iteration is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0070] In this embodiment, after the prediction model training is completed, when generating personalized treatment recommendations, the main steps are as follows: 1. Input patient information. Organize the patient's personal information (age, gender, medical history, symptoms), diagnosis results, treatment plans and other data into text that conforms to the model input format, such as: "The patient was a 25-year-old male with Class III malocclusion. CT scan showed mandibular overgrowth and X-rays showed malocclusion."

[0071] 2. Use Chinese word segmentation tools (such as Jieba, BERT word segmenter, etc.) to split the input text into words or subwords, such as: Original text: "The patient is 25 years old, male, and suffers from Class III malocclusion." After word segmentation: ["patient", "age", "25", "years old", "gender", "male", "suffering from", "Class III", "malocortiocytosis"].

[0072] Furthermore, each word or subword is mapped into a corresponding vector through pre-trained word embedding (such as Word2Vec, BERT, etc.) so as to be input into the Transformer model.

[0073] 3. Since the Transformer model does not have the ability to process sequences, it is necessary to add position encoding to each word vector to mark the position of each word in the sequence to help the model understand the order of words. The input data is a matrix containing word vectors and position encodings, and the Transformer model will process through this matrix.

[0074] 4. The encoder part of the trained Transformer model will learn the relationship and semantic information between each word in the text and other words through modules such as the multi-head self-attention mechanism. In the decoder part of the Transformer model, the output of the encoder will be used to generate natural language text of treatment recommendations. The decoder will generate personalized recommendation text based on the patient information, such as: "Based on your diagnosis, double jaw surgery is recommended, such as maxillary LeFort I + mandibular double-segment osteotomy. Six months of orthodontic preparation is required before surgery." This text can be displayed directly to doctors or patients as auxiliary information for treatment decisions.

[0075] This embodiment proposes a prediction model based on the Transformer network to generate personalized treatment recommendations for orthognathic surgery. Different from the traditional expert decision-making system, the prediction model is trained using a large amount of structured orthognathic medical data. The prediction model can automatically generate treatment plans and provide customized recommendations based on individual differences of patients.

[0076] This embodiment provides a more accurate and standardized data input method by structuring orthognathic medical data such as patient information, diagnosis results, treatment plans, etc. and converting them into natural language format. It can be easily input into the Transformer model for training, and the word segmentation and word embedding processing of the data enhances the data representation ability.

[0077] This embodiment uses the multi-head self-attention mechanism of Transformer to capture the relationship between various fields in the input data and learn the complex semantic information in the text. The prediction model can better understand the various medical relationships in orthognathic medical data and generate personalized treatment recommendations.

[0078] In order to improve the performance of the model when processing high-dimensional, nonlinear and dynamically changing data, this embodiment optimizes the feedforward network based on a bidirectionally coupled gray wolf optimization algorithm. By dynamically adjusting the search range and intensity, the convergence speed of the model is accelerated, and the stability and adaptability of the feedforward network are improved.

[0079] This embodiment expands the training samples by generating adversarial networks based on latent space decoupling, thereby solving the problem of insufficient samples in traditional training. By generating diverse and personalized samples, the diversity of the data set is enhanced, and the generalization ability and accuracy of the prediction model are improved.

[0080] Embodiment 2 The purpose of this embodiment is to provide a method for generating orthognathic surgery treatment suggestions, comprising: Obtain the patient's orthognathic medical data and perform preprocessing; Inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment recommendation for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

[0081] In further embodiments, there is also provided: An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, no further description is given here.

[0082] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0083] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0084] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method described in embodiment 1 is completed.

[0085] The method in the first embodiment can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0086] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in the first embodiment is implemented.

[0087] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0088] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0089] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, etc. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0090] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0091] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A system for generating orthognathic surgery treatment recommendations, characterized in that: The system comprises: An acquisition module is used to acquire the orthognathic medical data of the patient and perform preprocessing; A suggestion generation module, used for inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment suggestion for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

2. The orthognathic surgery treatment suggestion generating system according to claim 1, characterized in that: In the suggestion generation module, during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm. In the optimization of the feedforward network parameters, the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features, specifically including: Initialization unit, used to initialize the parameters of the feedforward network and the positions of each individual in the gray wolf optimization algorithm; A position updating unit, used to dynamically update the position of the gray wolf according to the current iteration number of the gray wolf algorithm, the importance of the orthognathic feature, and the fitness difference between the leading wolf and the last wolf; The parameter updating unit is used to update the parameters of the feedforward network using a gradient descent method based on the updated position of the gray wolf.

3. The orthognathic surgery treatment suggestion generating system according to claim 2, characterized in that: The position updating unit dynamically updates the position of the gray wolf according to the current iteration number of the gray wolf algorithm, the importance of the orthognathic feature, and the fitness difference between the leading wolf and the last wolf, specifically including: A first updating subunit, configured to adjust a first dynamic angle parameter and a second dynamic angle parameter according to a current training iteration number of the feedforward network and a total training iteration number of the feedforward network; a second updating subunit for calculating a dynamic impact factor of a feature according to the importance of the orthognathic feature during the iteration process; The third updating subunit is used to calculate the first coefficient and the second coefficient of the control search range according to the fitness difference between the leading wolf and the last wolf corresponding to the previous iteration of the gray wolf algorithm and the fitness difference between the leading wolf and the last wolf corresponding to the current iteration; The fourth updating subunit is used to update the position of the gray wolf according to the adjusted dynamic angle parameter, dynamic impact factor, and the first coefficient and the second coefficient for controlling the search range.

4. The orthognathic surgery treatment suggestion generating system according to claim 1, characterized in that: The suggestion generation module further includes: The training sample expansion module is used to construct a training set, expand the training set using a trained generative adversarial network model based on latent space decoupling to obtain an expanded training set, and use the expanded training set to train the Transformer network model.

5. The orthognathic surgery treatment suggestion generating system according to claim 4, characterized in that: In the training sample expansion module, the training of the generative adversarial network model based on latent space decoupling specifically includes: Initialization unit, used to initialize the generator and discriminator; The decomposition unit is used to decompose the latent space of the generator into multiple subspaces using a decoupling mechanism, and each subspace independently learns a specific latent feature; The adversarial training unit is used to perform adversarial training on the generator and the discriminator using a reverse adversarial strategy; A discriminator training unit, used for updating weights by back propagation in each round of training of the discriminator, and adopting a function based on an entropy regularization term as a loss function of the discriminator; A learning rate updating unit is used to update the learning rates of the generator and the discriminator using an adaptive learning rate updating mechanism.

6. The orthognathic surgery treatment suggestion generating system according to claim 5, characterized in that: In the decomposition unit, a decoupling mechanism is used to decompose the potential space of the generator into multiple subspaces, each of which independently learns a specific potential feature, including: The first decomposition sub-unit is used for the generator to process the potential features of each dimension respectively through multiple decoupled sub-networks to generate specific sub-data from different feature spaces; The second decomposition sub-unit is used for the generator to generate corresponding pseudo data based on the generated specific sub-data.

7. The orthognathic surgery treatment suggestion generating system according to claim 1, characterized in that: In the acquisition module, the preprocessing specifically includes: performing word segmentation operation on the acquired basic information of the patient and orthognathic medical data.

8. A method for generating orthognathic surgery treatment recommendations, characterized in that: The method comprises: Obtain the patient's orthognathic medical data and perform preprocessing; Inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment recommendation for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

9. An electronic device, characterized in that: The invention comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein the computer instructions are executed by the processor to complete the following steps: Obtain the patient's orthognathic medical data and perform preprocessing; Inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment recommendation for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the following steps: Obtain the patient's orthognathic medical data and perform preprocessing; Inputting the pre-processed orthognathic medical data of the patient into the prediction model to obtain auxiliary information of orthognathic surgery treatment recommendation for the patient; Wherein, the prediction model is obtained by training the Transformer network model; during the training process of the Transformer network model, the feedforward network parameters of the Transformer network model are adjusted based on the bidirectionally coupled gray wolf optimization algorithm, and the bidirectionally coupled gray wolf optimization algorithm dynamically adjusts the search strategy based on the importance of orthognathic features when optimizing the feedforward network parameters.

Images