Temporomandibular joint disease automatic diagnosis method and system based on artificial intelligence

Abstract

The invention belongs to the technical field of medical data processing, and discloses a temporal-mandibular joint disease automatic diagnosis method and system based on artificial intelligence, and the method comprises the steps: collecting and preprocessing data; expanding the patient data, dynamically adjusting the structure and parameters of a data generator, and generating the patient data; performing feature extraction on the patient data, and automatically adjusting an evolution rule of a feature extraction network to obtain patient data features; performing dimensionality reduction on the data features to obtain low-dimensional patient data features; inputting the low-dimensional patient data features into a classifier, and dynamically adjusting the configuration and weight of hidden layer nodes of the classifier according to the low-dimensional patient data features to obtain a classification result; and defining a loss function, and training the model. Parameters can be automatically adjusted in the data feature extraction process, the data dimension reduction process is optimized by adopting quantum state mapping and an adaptive adjustment mechanism, the classifier can dynamically adjust the number of hidden layer nodes and an activation function, and the accuracy and robustness of a diagnosis result are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to an automatic diagnosis method and system for temporomandibular joint disease based on artificial intelligence. 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] Temporomandibular joint disease (TMJD) is a common oral and maxillofacial disease, usually manifested by facial pain, discomfort in the TMJ area, difficulty chewing, limited mouth opening, or joint clicking. Because its symptoms are highly individual and have similar clinical manifestations to many other diseases, the diagnosis of TMJD is often complex and prone to misdiagnosis and missed diagnosis. With the advancement of data acquisition technology, modern medical fields have gradually introduced more auxiliary data sources for disease diagnosis, such as self-reported symptom data, physiological signal data, motor function data, etc. However, there are still many challenges in the acquisition and analysis of these data. Due to the heterogeneity of various data sources, how to effectively extract features with diagnostic value from these non-imaging data and avoid noise interference has become a key issue in improving diagnostic accuracy. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes an automatic diagnosis method and system for temporomandibular joint disease based on artificial intelligence, which adopts generative adversarial networks to expand data, and the data feature extraction process can automatically adjust parameters. Quantum state mapping and adaptive adjustment mechanism are used to optimize the data dimension reduction process. The classifier can dynamically adjust the number of hidden layer nodes and activation functions, thereby improving the accuracy and robustness of the diagnosis results.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides an automatic diagnosis method for temporomandibular joint disease based on artificial intelligence, comprising the following steps: Collect patient data and pre-process the patient data; Expand the patient data, dynamically adjust the structure and parameters of the data generator according to the noise characteristics of the patient data, and generate the patient data; A neural network based on dynamic population evolution optimization is used to extract features from patient data, and a self-correction mechanism is used to automatically adjust the evolution rules according to the characteristics of the current patient data to obtain patient data features; Perform dimensionality reduction operations on patient data features, and use quantum projection strategies to find low-dimensional embeddings in high-dimensional feature space to obtain low-dimensional patient data features; Inputting low-dimensional patient data features into the classifier, dynamically adjusting the configuration and weights of the hidden layer nodes of the classifier according to the low-dimensional patient data features, and obtaining classification results; Define the loss function, optimize the model parameters, and train the automatic diagnosis model.

[0006] As an optional implementation, the structure and parameters of the data generator are dynamically adjusted by using an adaptive gating mechanism in the data generator, wherein the adaptive gating mechanism is specifically:

[0007] in, is the gated output, is the weight matrix of the gating layer, is the bias term of the gating layer, is the ReLU activation function, which is used to control the activation degree of different paths in the generator.

[0008] As an optional implementation, the loss values ​​of the output results of the generator and the discriminator are calculated separately. The loss of the generator is obtained through the judgment feedback of the discriminator on the generated patient data, and the loss of the discriminator is calculated based on the classification results of the real patient data and the fake patient data. The parameters of the generator and the discriminator are optimized according to the loss value to generate patient data.

[0009] As an optional implementation, the loss functions of the output results of the generator and the discriminator are:

[0010]

[0011] in, is the discriminator loss function, is the generator loss function, is the distribution of real patient data, To generate the distribution of patient data, Express expectations, It means that it obeys a specific distribution. is the noise distribution, is the feature embedding function, is the feature embedding weight factor, Represents the similarity calculation function between two feature vectors, is the discriminator, is the entropy calculation function, is the entropy calculation of the discriminator, is a hyperparameter that controls the influence of the entropy term.

[0012] As an optional implementation, a self-correction mechanism is used to automatically adjust the evolution rules according to the characteristics of the current patient data, specifically: According to the changing trend of the loss function of the feature extraction network, the probability of crossover and mutation operations of the population is adjusted, and then the evolution rules are adjusted.

[0013] As an optional implementation, a quantum projection strategy is used to find a low-dimensional embedding in a high-dimensional feature space, specifically: The high-dimensional features are converted into quantum states through quantum projection operations, and the features are represented in the quantum state space. The adaptive adjustment mechanism is adopted so that the quantum state weight coefficient and the projection operator can be dynamically updated according to the current patient data feature distribution. The update method is expressed as:

[0014] in, For the updated The weight coefficient of each quantum state, To dynamically optimize the learning rate, is the gradient of the similarity loss function, is the symbolic function, For the The weight coefficient after projection, is a small positive number.

[0015] In a second aspect, the present invention provides an automatic diagnosis system for temporomandibular joint disease based on artificial intelligence, comprising: The data acquisition module is configured to: collect patient data and pre-process the patient data; The data expansion module is configured to: expand the patient data, dynamically adjust the structure and parameters of the data generator according to the characteristics of the noise of the patient data, and generate the patient data; The feature extraction module is configured to: extract features from patient data using a neural network based on dynamic population evolution optimization, and automatically adjust the evolution rules according to the characteristics of current patient data using a self-correction mechanism to obtain patient data features; The feature dimensionality reduction module is configured to: perform dimensionality reduction operations on patient data features, find low-dimensional embeddings in high-dimensional feature space through quantum projection strategy, and obtain low-dimensional patient data features; The classification module is configured to: input the low-dimensional patient data features into the classifier, dynamically adjust the configuration and weight of the hidden layer nodes of the classifier according to the low-dimensional patient data features, and obtain the classification results; The model training module is configured to: define the loss function, optimize the model parameters, and train the automatic diagnosis model.

[0016] In a third aspect, the present invention provides an electronic device comprising 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 the first aspect is performed.

[0017] 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 method described in the first aspect is performed.

[0018] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention proposes an automatic diagnosis method and system for temporomandibular joint disease based on artificial intelligence. Since insufficient training samples can easily lead to poor generalization ability of the model and affect the accuracy of the model, the present invention adopts a generative adversarial network to generate fake patient data and expand the patient data. Combined with the adaptive gating mechanism and feature space similarity measurement, the generator can dynamically adjust the generation strategy according to the characteristics of the noise of the input patient data, thereby improving the authenticity and diversity of the generated data, thereby increasing the scale of the training data set, overcoming the problem of insufficient samples in traditional data collection, and improving the generalization ability and diagnostic accuracy of the model.

[0020] 2. The present invention proposes an automatic diagnosis method and system for temporomandibular joint disease based on artificial intelligence, which uses a neural network algorithm based on dynamic population evolution optimization to extract patient data features. Through the dynamic population evolution optimization algorithm, the feature extraction process can automatically adjust parameters and automatically adapt to data distribution by adjusting the evolution rules, avoiding problems such as gradient vanishing, gradient explosion, and local optimal solutions in traditional neural network training, improving the training stability of the feature extraction model, and reducing the risk of overfitting.

[0021] 3. The present invention proposes an automatic diagnosis method and system for temporomandibular joint disease based on artificial intelligence. The feature dimension reduction of patient data is performed through the quantum projection autoencoder algorithm. The quantum projection autoencoder optimizes the dimension reduction process through adaptive adjustment of the quantum state, ensuring the retention of feature information, while reducing the dimension of the feature space, thereby improving the training efficiency and classification accuracy of the subsequent classification model.

[0022] 4. The present invention proposes an automatic diagnosis method and system for temporomandibular joint disease based on artificial intelligence, and adopts an extreme learning machine algorithm based on dynamic adjustment of the hidden layer to train the classifier. The extreme learning machine dynamically adjusts the number of hidden layer nodes and the activation function, so that the classifier can automatically optimize its structure according to the complexity of the input data, thereby improving the classifier's adaptability to different data distributions, solving the problem of performance fluctuations in traditional classifiers caused by changes in data complexity, and thereby improving the accuracy and robustness of the diagnosis results.

[0023] 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

[0024] 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.

[0025] Figure 1 A flowchart of an automatic diagnosis method for temporomandibular joint disease based on artificial intelligence provided in Example 1 of the present invention; Figure 2 A training flow chart of a neural network algorithm based on dynamic population evolution optimization provided in Example 1 of the present invention; Figure 3 A flowchart of training an automatic diagnosis model for temporomandibular joint disease based on artificial intelligence provided in Example 1 of the present invention; Figure 4 This is a flow chart of the application of the artificial intelligence-based automatic diagnosis model for temporomandibular joint disease provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0026] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0027] 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.

[0028] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0030] Example 1 like Figure 1 , Figure 3 and Figure 4 As shown, this embodiment provides an automatic diagnosis method for temporomandibular joint disease based on artificial intelligence, comprising the following steps: S1 collects patient data and preprocesses the patient data; S2 expands the patient data and dynamically adjusts the structure and parameters of the data generator according to the noise characteristics of the patient data to generate the patient data; S3 uses a neural network based on dynamic population evolution optimization to extract features from patient data, and uses a self-correction mechanism to automatically adjust the evolution rules according to the characteristics of the current patient data to obtain patient data features; S4 performs dimensionality reduction on patient data features and uses quantum projection strategy to find low-dimensional embedding in high-dimensional feature space to obtain low-dimensional patient data features; S5 inputs the low-dimensional patient data features into the classifier, dynamically adjusts the configuration and weight of the hidden layer nodes of the classifier according to the low-dimensional patient data features, and obtains the classification result; S6 defines the loss function, optimizes the model parameters, and trains the automatic diagnosis model.

[0031] First, S1, collect patient data and pre-process the patient data.

[0032] The training data for training the automatic diagnosis model of temporomandibular joint disease disclosed in the present invention comes from clinical medical records, patient medical records extracted from the hospital's electronic health record system, and patient questionnaire survey results. All collected data are stored in a standardized format, specifically a CSV file.

[0033] The properties of the data include: is the patient ID, For gender, For age, For medical history, Self-described by the patient, For medical diagnosis description, For treatment methods, The diagnosis result of the disease, For family history, For other symptoms.

[0034] It should be noted that this embodiment is only an illustration of one data format and type of the present disclosure. In actual applications, the attributes of data are usually more than 10 attributes, and the number of attributes of data may reach dozens or even hundreds. For the collected text format data, the present disclosure adopts the Word2Vec algorithm to vectorize the text. The Word2Vec algorithm is a commonly used vectorization algorithm in this field. It scans the texts to be vectorized based on a preset large-scale corpus, and represents each word as a one-hot encoded vector. The dimension of the one-hot encoded vector is equal to the size of the vocabulary in the corpus.

[0035] Furthermore, the collected data is labeled. The labeling method disclosed in the present invention is manual labeling. In one embodiment, the labeled categories include: normal, mild disease, moderate disease, and severe disease.

[0036] S2. Expand the patient data, dynamically adjust the structure and parameters of the data generator according to the noise characteristics of the patient data, and generate the patient data.

[0037] Expanding the collected patient data: In the task of this disclosure, 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 and affect the accuracy of the model. Therefore, this disclosure uses a generative adversarial network algorithm to generate samples, thereby achieving patient data expansion.

[0038] The structure of the generative adversarial network includes a generator and a discriminator. The generator is responsible for generating fake patient data similar to real patient data, and the discriminator is responsible for distinguishing between real patient data and generated patient data.

[0039] On the basis of the traditional generative adversarial network, the present invention adopts an adaptive loss function, which not only considers the distance between the generated patient data and the real patient data, but also adopts the similarity measurement of the feature embedding space so that the generator can pay more attention to generating patient data that is consistent with the real patient data in the high-dimensional feature space during the training process, thereby effectively improving the quality of the generated patient data.

[0040] Specifically, the training process of the generative adversarial network algorithm is as follows: S201, the parameters of the generator and discriminator of the generative adversarial network are set by random initialization to ensure that the algorithm can explore a variety of solution spaces. Preferably, the parameters of the generator and the discriminator obey a normal distribution with a mean of 0 and a variance of the unit matrix.

[0041] S202. The generator structure of the traditional generative adversarial network is in a static form, that is, the noise data is processed through fixed network layers and parameters. The present disclosure adopts an adaptive dynamic generator structure in the patient data generation process of the generator, aiming to dynamically adjust the structure and parameters of the generation network according to the characteristics of the input noise. Specifically, it is achieved by adopting an adaptive gating mechanism in the generator to enhance the diversity and quality of the generated patient data. The implementation method of the adaptive gating mechanism is expressed as:

[0042] In the formula, is the gated output, is the weight matrix of the gating layer, is the bias term of the gating layer, is the ReLU activation function, which is used to control the activation degree of different paths in the generator.

[0043] S203, the generator generates fake patient data from random noise. The noise sample is used as input, passes through the generator network, and undergoes multiple layers of nonlinear transformation to output generated patient data. The generator flexibly adjusts the generated patient data according to the input characteristics by adopting an adaptive gating mechanism. The generator converts the input noise into generated patient data as follows:

[0044] In the formula, Patient data generated by the generator, is the generator function, is the input noise of the generator, is the weight matrix of the generator, is a nonlinear activation function with parameters, used as the activation function of the generator, is the bias term of the generator.

[0045] Furthermore, the nonlinear activation function with parameters makes the generator more flexible in processing negative inputs, thereby generating richer features. The calculation method is expressed as:

[0046] In the formula, To control the slope parameter of the negative part, it is a learnable parameter and is learned and updated through the gradient descent strategy.

[0047] S204, the discriminator receives the generated patient data and the real patient data, and outputs the classification results of the two types of patient data. The discriminator extracts the patient data features through 3 convolutional layers and 1 fully connected layer, and finally maps the results to [0, 1] through the Sigmoid activation function, indicating the probability that the patient data is real, expressed as:

[0048]

[0049] In the formula, is the Sigmoid activation function, is the weight matrix of the discriminator, is the feature extraction function of the discriminator, is the bias term of the discriminator, is the discriminator function.

[0050] In one embodiment, the feature extraction function of the discriminator is composed of multiple layers of convolution and pooling operations, which helps to extract more abstract feature representations, thereby improving the performance of the discriminator. The specific form is:

[0051] In the formula, represents the convolution operation, is the convolution kernel, is the bias term of the convolutional layer, This is a pooling operation, specifically using maximum pooling.

[0052] S205. According to the output results of the generator and the discriminator, their loss values ​​are calculated respectively. The loss of the generator is mainly obtained through the judgment feedback of the discriminator on the generated patient data, and the loss of the discriminator is calculated based on the classification results of the real patient data and the fake patient data. The calculation method of the loss function is expressed as:

[0053]

[0054] In the formula, is the discriminator loss function, is the generator loss function, is the distribution of real patient data, To generate the distribution of patient data, Express expectations, It means that it obeys a specific distribution. is the noise distribution, is the feature embedding function, is the feature embedding weight factor, Represents the similarity calculation function between two feature vectors, is the discriminator, is the entropy calculation function, is the entropy calculation of the discriminator, is a hyperparameter for controlling the influence of the entropy term. Preferably, the feature embedding function uses a hyperbolic tangent function for mapping transformation.

[0055] Furthermore, the similarity calculation function between two feature vectors enables the generator to focus not only on the authenticity of the generated data during optimization, but also on the similarity between the generated patient data and the real patient data in the feature space. The implementation method is expressed as:

[0056] In the formula, is the L2 norm.

[0057] Furthermore, the entropy calculation of the discriminator can guide the discriminator to maintain a good classification boundary during training and avoid overfitting. The implementation method is expressed as:

[0058] S206. According to the loss value, the parameters of the generator and the discriminator are updated through the back propagation algorithm. The generator will adjust its generation strategy based on the feedback of the discriminator to improve the authenticity of the generated patient data. The discriminator will enhance its recognition ability through updating. The calculation method of the parameter update amount of the generator and the discriminator is expressed as:

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] In the formula, is the learning rate of the generative adversarial network, Update the intermediate quantities for the generator weights, Update the intermediate amount for the discriminator weights, and Represent the update amount of the generator weight and bias respectively, and Represent the update amount of the discriminator weight and bias respectively, is the symbol of partial derivative, is the momentum coefficient, which is preset manually. It allows the generator and discriminator to maintain a certain inertia during the weight update process, thereby accelerating the speed of training convergence. is the gradient information of the generator to the input noise, indicating the sensitivity to noise changes. By dynamically combining the gradient information, the generator can respond to the input features more intelligently. Preferably, Set to 0.3.

[0065] S207. After the generator generates the patient data, a distribution correction process is used to optimize the quality of the generated patient data. Specifically, the statistical characteristics of the generated patient data are calculated. By calculating the mean and variance of the generated patient data, its distribution characteristics can be obtained. The calculation method is expressed as:

[0066]

[0067] In the formula, To generate the mean of the patient data, To generate the variance of the patient data, is the number of samples to generate patient data, Indicates Generate samples.

[0068] Furthermore, the statistical characteristics of the real patient data are calculated, and the method of calculating the mean and variance of the real patient data is expressed as:

[0069]

[0070] In the formula, The number of samples of real patient data input to the generative adversarial network for the current batch, Indicates A real sample, is the mean of the real patient data, is the variance of the real patient data.

[0071] Furthermore, based on the calculated mean and variance, the generated patient data is corrected, and the generated patient data is recalibrated using the mean and variance of the real patient data, so that the distribution of the generated patient data is closer to the distribution of the real patient data. The correction method is expressed as:

[0072] Here, The corrected generated patient data is used as the generated patient data for the next iteration.

[0073] S208, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0074] After the data expansion model training is completed, the trained patient data expansion model is used to increase the number of samples. In one embodiment, assuming that the original collected samples are 800, and the data expansion model expands and generates 200 samples, the expanded patient data set contains 1000 samples.

[0075] S3. Use a neural network based on dynamic population evolution optimization to extract features from patient data, and use a self-correction mechanism to automatically adjust the evolution rules according to the characteristics of the current patient data to obtain patient data features.

[0076] Patient data feature extraction model training: The expanded patient data is input into the patient data feature extraction model to train the patient data feature extraction model. The present disclosure adopts a 6-layer fully connected neural network to extract patient data features. In the prior art, some solutions use neural networks to extract patient data features.

[0077] In some neural network structures, you may encounter problems such as gradient disappearance, gradient explosion, or falling into a local optimal solution, which will affect the stability of training and the performance of the model.

[0078] The present invention adopts a neural network algorithm based on dynamic population evolution optimization as a patient data feature extraction model. In traditional population evolution algorithms, the evolution of all individuals is based on fixed rules. The present invention uses a self-correction mechanism to automatically adjust the evolution rules according to the characteristics of the current training data, and adjusts the probabilities of crossover and mutation according to the changing trend of the loss function, so that the algorithm can adapt to different patient data distributions more flexibly and improve the generalization ability and training efficiency of the model.

[0079] Specifically, the neural network algorithm based on dynamic population evolution optimization is as follows Figure 2 shown.

[0080] S301. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, and each individual represents a configuration of network weights. Specifically, the population size is , for The individual initialization weights and biases are expressed as:

[0081]

[0082] In the formula, For the The weight of the neural network corresponding to each individual, For the The bias of the neural network corresponding to each individual, Indicates The weight matrix of each individual in the initial state is Indicates The bias of each individual in the initial state, represents the initialization variance, It means the mean is 0 and the variance is The normal distribution of is a normal distribution. Preferably, Set to 0.01.

[0083] S302: For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the first Individuals use their weights and biases to calculate the loss on the training data set, expressed as:

[0084] In the formula, For the The loss of the neural network corresponding to each individual, Indicates the number of patient data samples input in the current batch, represents the composite loss function, represents the neural network model function, Indicates The characteristics of the patient data sample, Indicates The labels of the patient data samples.

[0085] In one embodiment, the composite loss function includes a regularization term, which can increase the generalization ability of the model. The calculation method is expressed as:

[0086] In the formula, is the mean square error function, is the regularization term, is the regularization parameter.

[0087] Furthermore, the calculation method of the regularization term is expressed as:

[0088] In the formula, For the first The neural network corresponding to each individual A weight.

[0089] S303. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. Specifically, based on the fitness of the individual, the excellent individual has a higher probability of being selected. The calculation method of the probability of the individual being selected is expressed as:

[0090] In the formula, Indicates The probability of an individual being selected is is the parameter that controls the selection pressure, For the The loss of the neural network corresponding to each individual. Preferably, Set to 2.

[0091] S304. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0092]

[0093] In the formula, is the crossover rate, is the weight of the neural network corresponding to the first individual selected, is the bias of the neural network corresponding to the first individual selected, is the weight of the neural network corresponding to the second individual selected, is the bias of the neural network corresponding to the second individual selected, is the weight of the neural network corresponding to the individual after the crossover operation, is the bias of the neural network corresponding to the individual after the crossover operation. Preferably, Set to 0.3.

[0094] And the mutation operation performs a small random perturbation on the newly generated individual weights, expressed as:

[0095]

[0096] In the formula, represents the variance of the variation, is the weight of the neural network corresponding to the individual after the mutation operation, is the bias of the neural network corresponding to the individual after the mutation operation. Preferably, Set to 0.04.

[0097] S305, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0098] S4. Perform dimensionality reduction operations on patient data features, and use quantum projection strategy to find low-dimensional embedding in high-dimensional feature space to obtain low-dimensional patient data features.

[0099] Feature dimension reduction model training: The data after feature extraction is input into the feature dimension reduction model for feature dimension reduction model training. The present invention adopts a quantum projection-based autoencoder algorithm as a feature dimension reduction algorithm, and uses a quantum projection strategy to find low-dimensional embedding in a high-dimensional feature space, thereby effectively reducing the feature dimension and maintaining the essential information of the patient data after feature extraction.

[0100] Specifically, the training process of the quantum projection-based autoencoder algorithm is as follows: S401, set the initial quantum state, define the base state of the feature space, and set the quantum state as the mapping of the high-dimensional feature space. The state is expressed as:

[0101] In the formula, For the The weight coefficient of each quantum state, The first The ground state representation of the input features in the quantum state space, is the total number of input features, is a quantum state.

[0102] In one embodiment, the weight coefficient uses Gaussian distribution to assign weights to features so that features closer to the center have greater influence. The calculation method is expressed as:

[0103] In the formula, Features The L2 norm of Features The L2 norm of is the L2 norm, The first input features, is the Gaussian kernel width, The first input features.

[0104] S402, converting the input high-dimensional features into quantum states through quantum projection operations, and performing feature representation in the quantum state space. Specifically, the high-dimensional features are mapped to the quantum state space using a projection operator, which is represented as:

[0105] In the formula, is the quantum projection operator, is the quantum state transformed by the quantum projection operation, For the The weight coefficient after projection, For the The quantum state representation of the projected features, For the The conjugate state represented by the quantum state of the projected feature, is the number of features in the low-dimensional space.

[0106] S403. In the dynamic optimization iteration process, an adaptive adjustment mechanism is adopted so that the quantum state weight coefficient and the projection operator can be dynamically updated according to the feature distribution of the patient data after the current feature extraction. The update method is expressed as:

[0107] In the formula, For the updated The weight coefficient of each quantum state, To dynamically optimize the learning rate, is the gradient of the similarity loss function, is the symbolic function, For the The weight coefficient after projection, is a small positive number to avoid division by zero errors. Preferably, Set to 0.01, Set to 0.0001.

[0108] Furthermore, the calculation method of the weight coefficient after projection is expressed as:

[0109] In the formula, For the A quantum state transformed by quantum projection operation, is the Gaussian kernel width, For the A quantum state transformed by a quantum projection operation.

[0110] S404, the quantum state is mapped by an autoencoder to generate a low-dimensional feature representation. The relationship between the output and input of the autoencoder is expressed as:

[0111] In the formula, is the activation function of the autoencoder, is the weight matrix of the autoencoder, is the bias term of the autoencoder. Represents the feature representation after dimensionality reduction.

[0112] In one embodiment, the activation function of the autoencoder adopts a nonlinear combination form, so that the feature map has a stronger expressive ability and can capture the complex nonlinear relationship of the patient data after the input feature extraction. The calculation method is expressed as:

[0113] In the formula, is a hyperparameter that controls the strength of the nonlinear activation. Preferably, Set to 0.3.

[0114] S405. According to the contrastive learning mechanism, an embedding space contrastive learning strategy is adopted. By defining a similarity loss function, similar features are forced to be close in the low-dimensional space, while dissimilar features are kept at a certain distance, thereby enhancing the separability of the reduced-dimensional features. Specifically, the similarity loss function is used to promote the aggregation of similar features in the low-dimensional space. The calculation method is expressed as follows:

[0115] In the formula, is the similarity loss function, as the loss function of the autoencoder, is the distance metric function, is the regularization coefficient of the autoencoder.

[0116] In one embodiment, the distance metric function can effectively calculate the similarity between features, and the calculation method is expressed as:

[0117] In the formula, for and The dot product of .

[0118] S406. In each iteration, the projection strategy of the quantum state is dynamically adjusted by combining the information of the current quantum state and the feedback of the training data to ensure that the dimensionality reduction process can adapt to the changes in the feature space and finally converge to a stable low-dimensional representation. The update rule is expressed as:

[0119] In the formula, is the learning rate of the autoencoder, is the gradient of the similarity loss function.

[0120] S407, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0121] After the training is completed, the final low-dimensional feature representation is output to provide input for the subsequent classifier model training. The final low-dimensional feature representation is extracted through the output of the autoencoder as follows:

[0122] In the formula, represents the quantum state projection after training, Represents the low-dimensional features of the final output.

[0123] S5. Input the low-dimensional patient data features into the classifier, dynamically adjust the configuration and weight of the hidden layer nodes of the classifier according to the low-dimensional patient data features, and obtain the classification results.

[0124] Classifier model training: The reduced-dimensional data is input into the classifier to train the classifier model. The present disclosure adopts an extreme learning machine algorithm based on dynamic adjustment of the hidden layer as the classifier model.

[0125] The extreme learning machine includes an input layer, a hidden layer and an output layer. The hidden layer is composed of multiple activation functions. The nonlinear mapping of features is achieved by optimizing the configuration and weights of the hidden layer nodes.

[0126] Based on the traditional extreme learning machine, the present invention adopts a dynamic adjustment mechanism to adapt to the changes in the feature space. The number of hidden layer nodes and the connection weights will be dynamically adjusted according to the complexity of the patient data after the input feature dimensionality reduction, ensuring the balance between the complexity and accuracy of the model is automatically optimized during the training process.

[0127] Specifically, the training process of the extreme learning machine algorithm based on dynamic adjustment of the hidden layer is as follows: S501. The feature vector output from the feature dimensionality reduction model is used as the input of the classifier. Suppose the feature vector input to the extreme learning machine is , the label of the training sample input to the extreme learning machine is , Represents the dimensional space of data, represents the number of samples input into the extreme learning machine, represents the feature dimension of the input extreme learning machine, is the number of categories for classification, represents the input to the extreme learning machine The feature vector of the samples, represents the input to the extreme learning machine The label of a sample is expressed as:

[0128]

[0129] In the formula, represents the input to the extreme learning machine The labels of samples, represents the input to the extreme learning machine The labels of samples, represents the input to the extreme learning machine The labels of samples, represents the input to the extreme learning machine The labels of samples, represents the input to the extreme learning machine The labels of samples, represents the input to the extreme learning machine The labels of the samples; the samples input into the extreme learning machine are the patient data after feature dimensionality reduction.

[0130] S502, randomly initialize the weights of the hidden layer nodes of the extreme learning machine, and set the activation function of the hidden layer nodes according to the distribution characteristics of the input features. Each hidden layer node will transform the input features in a nonlinear manner, thereby increasing the expression ability of the model. The hidden layer weight matrix is ​​set to , is the number of hidden layer nodes of the extreme learning machine, and the output of the hidden layer of the extreme learning machine is expressed as:

[0131] In the formula, is the activation function of the extreme learning machine, is the bias term of the hidden layer of the extreme learning machine, is the transpose of the hidden layer weight matrix of the extreme learning machine, is the output of the hidden layer of the extreme learning machine.

[0132] Furthermore, the output of the hidden layer of the extreme learning machine is expressed as:

[0133] In the formula, is the first feature of the output of the hidden layer of the extreme learning machine, is the second feature of the output of the hidden layer of the extreme learning machine, is the output of the hidden layer of the extreme learning machine Features.

[0134] In one embodiment, the activation function of the extreme learning machine adopts a nonlinear function, and a periodic fluctuation calculation method is used to enhance the expression ability of the model. The calculation method is expressed as:

[0135] In the formula, is a parameter that controls the steepness, and They are the parameters for adjusting the amplitude and frequency of the activation function components respectively.

[0136] S503. Through real-time analysis of the input patient data after feature dimensionality reduction, the complexity of the features is dynamically monitored. If the complexity increases, the number of hidden layer nodes is increased; if the complexity decreases, the number of nodes is appropriately reduced to adapt to the changes in patient data and ensure that the classifier maintains the best performance in various environments. The calculation method of complexity is expressed as:

[0137] In the formula, is the complexity of the features input to the extreme learning machine.

[0138] Furthermore, the number of hidden layer nodes is dynamically adjusted by calculating the complexity of the input patient data. The adjustment method is expressed as:

[0139] In the formula, The threshold is adjusted to a preset complexity. Preferably, Set to 0.5.

[0140] S504, by mapping the hidden layer nodes to the input features, the weight of the output layer is calculated, and the weight between the hidden layer and the output layer is adjusted by a randomized algorithm according to the feedback of the loss function so that it continuously approaches the optimal solution. The weight matrix from the hidden layer to the output layer is set to , the output of the extreme learning machine is expressed as:

[0141] Where is the transpose of the output of the hidden layer of the extreme learning machine, is the output classification vector of the extreme learning machine.

[0142] Furthermore, the loss function of the extreme learning machine prevents overfitting by adopting the penalty term of the hidden layer nodes, and the calculation method is expressed as:

[0143] In the formula, is the regularization coefficient of the extreme learning machine, is the regularization coefficient of the hidden layer nodes, For the extreme learning machine The classification vector of samples, is the jth feature of the output of the hidden layer of the extreme learning machine, is the loss function of the extreme learning machine, For the The true label vector of samples. Preferably, Set to 0.3, Set to 0.5.

[0144] S505. In each iteration, the loss function of the extreme learning machine is calculated, and the weight matrix from the hidden layer to the output layer is updated by the gradient descent method to ensure the learning efficiency of the classifier in the feature space. The updating method is expressed as:

[0145] In the formula, is the learning rate of the extreme learning machine, is the updated weight matrix from the hidden layer to the output layer.

[0146] S506, through multiple rounds of iterations, the weights are continuously updated until the loss function converges. The convergence condition is expressed as:

[0147] In the formula, is the preset extreme learning machine convergence threshold, is the loss function value of the extreme learning machine in the previous iteration, is the loss function value of the extreme learning machine in this iteration, is the convergence threshold of the extreme learning machine parameters, ensuring that not only the loss function changes small enough, but also the amplitude of weight updates is small enough to ensure the stability of the model. Preferably, Set to 0.01, Set to 0.05.

[0148] After the extreme learning machine training is completed, the optimized classifier model is output.

[0149] S6. Define the loss function, optimize the model parameters, and train the automatic diagnosis model.

[0150] Automatic diagnosis of temporomandibular joint disease: Automatic diagnosis of temporomandibular joint disease is achieved using a trained model. In one embodiment, the collected original patient data is input into a trained feature extraction and feature dimension reduction model for feature processing. Further, the processed features are input into a classifier model for classification to obtain classification results. In this embodiment, the classification categories include: normal, mild disease, moderate disease, and severe disease.

[0151] Example 2 This embodiment provides an automatic diagnosis system for temporomandibular joint disease based on artificial intelligence, including: The data acquisition module is configured to: collect patient data and pre-process the patient data; The data expansion module is configured to: expand the patient data, dynamically adjust the structure and parameters of the data generator according to the characteristics of the noise of the patient data, and generate the patient data; The feature extraction module is configured to: extract features from patient data using a neural network based on dynamic population evolution optimization, and automatically adjust the evolution rules according to the characteristics of current patient data using a self-correction mechanism to obtain patient data features; The feature dimensionality reduction module is configured to: perform dimensionality reduction operations on patient data features, find low-dimensional embeddings in high-dimensional feature space through quantum projection strategy, and obtain low-dimensional patient data features; The classification module is configured to: input the low-dimensional patient data features into the classifier, dynamically adjust the configuration and weight of the hidden layer nodes of the classifier according to the low-dimensional patient data features, and obtain the classification results; The model training module is configured to: define the loss function, optimize the model parameters, and train the automatic diagnosis model.

[0152] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0153] 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, it will not be described in detail here.

[0154] 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.

[0155] 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.

[0156] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.

[0157] The method in Example 1 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 is not described in detail here.

[0158] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in embodiment 1 is implemented.

[0159] 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.

[0160] 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.

[0161] 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, and the like. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, and the like.

[0162] 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.

[0163] 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. An automatic diagnosis method for temporomandibular joint disease based on artificial intelligence, characterized in that: The following steps are involved: Collect patient data and pre-process the patient data; Expand the patient data, dynamically adjust the structure and parameters of the data generator according to the noise characteristics of the patient data, and generate the patient data; A neural network based on dynamic population evolution optimization is used to extract features from patient data, and a self-correction mechanism is used to automatically adjust the evolution rules according to the characteristics of the current patient data to obtain patient data features; Perform dimensionality reduction operations on patient data features, and use quantum projection strategies to find low-dimensional embeddings in high-dimensional feature space to obtain low-dimensional patient data features; Inputting low-dimensional patient data features into the classifier, dynamically adjusting the configuration and weights of the hidden layer nodes of the classifier according to the low-dimensional patient data features, and obtaining classification results; Define the loss function, optimize the model parameters, and train the automatic diagnosis model.

2. The method for automatic diagnosis of temporomandibular joint disease based on artificial intelligence according to claim 1, characterized in that: The structure and parameters of the data generator are dynamically adjusted by using an adaptive gating mechanism in the data generator, wherein the adaptive gating mechanism is specifically: in, is the gated output, is the weight matrix of the gating layer, is the bias term of the gating layer, is the ReLU activation function, which is used to control the activation degree of different paths in the generator.

3. The method for automatic diagnosis of temporomandibular joint disease based on artificial intelligence according to claim 1, characterized in that: The loss values ​​of the output results of the generator and discriminator are calculated respectively. The loss of the generator is obtained through the judgment feedback of the discriminator on the generated patient data. The loss of the discriminator is calculated based on the classification results of the real patient data and the fake patient data. The parameters of the generator and the discriminator are optimized according to the loss value to generate patient data.

4. The method for automatic diagnosis of temporomandibular joint disease based on artificial intelligence as claimed in claim 3, characterized in that: The loss functions of the output results of the generator and the discriminator are: in, is the discriminator loss function, is the generator loss function, is the distribution of real patient data, To generate the distribution of patient data, Express expectations, It means that it obeys a specific distribution. is the noise distribution, is the feature embedding function, is the feature embedding weight factor, Represents the similarity calculation function between two feature vectors, is the discriminator, is the entropy calculation function, is the entropy calculation of the discriminator, is a hyperparameter that controls the influence of the entropy term.

5. The method for automatic diagnosis of temporomandibular joint disease based on artificial intelligence according to claim 1, characterized in that: The self-correction mechanism is used to automatically adjust the evolution rules according to the characteristics of the current patient data, specifically: According to the changing trend of the loss function of the feature extraction network, the probability of crossover and mutation operations of the population is adjusted, and then the evolution rules are adjusted.

6. The method for automatic diagnosis of temporomandibular joint disease based on artificial intelligence according to claim 1, characterized in that: The quantum projection strategy is used to find low-dimensional embedding in high-dimensional feature space. Specifically: The high-dimensional features are converted into quantum states through quantum projection operations, and the features are represented in the quantum state space. The adaptive adjustment mechanism is adopted so that the quantum state weight coefficient and the projection operator can be dynamically updated according to the current patient data feature distribution. The update method is expressed as: in, For the updated The weight coefficient of each quantum state, To dynamically optimize the learning rate, is the gradient of the similarity loss function, is the symbolic function, For the The weight coefficient after projection, is a small positive number.

7. The automatic diagnosis system of temporomandibular joint disease based on artificial intelligence is characterized by: include: The data acquisition module is configured to: collect patient data and pre-process the patient data; The data expansion module is configured to: expand the patient data, dynamically adjust the structure and parameters of the data generator according to the characteristics of the noise of the patient data, and generate the patient data; The feature extraction module is configured to: extract features from patient data using a neural network based on dynamic population evolution optimization, and automatically adjust the evolution rules according to the characteristics of current patient data using a self-correction mechanism to obtain patient data features; The feature dimensionality reduction module is configured to: perform dimensionality reduction operations on patient data features, find low-dimensional embeddings in high-dimensional feature space through quantum projection strategy, and obtain low-dimensional patient data features; The classification module is configured to: input the low-dimensional patient data features into the classifier, dynamically adjust the configuration and weight of the hidden layer nodes of the classifier according to the low-dimensional patient data features, and obtain the classification results; The model training module is configured to: define the loss function, optimize the model parameters, and train the automatic diagnosis model.

8. An electronic device, characterized in that: The method comprises 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 according to any one of claims 1 to 6 is completed.

9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images