A trauma level assessment method, device and program product

Abstract

The present application relates to the field of intelligent medical care, and specifically to a method, device, and program product for evaluating trauma levels. The method includes obtaining data from trauma patients; inputting the data into an evaluation model to obtain an evaluation result; wherein the training process of the evaluation model is: obtaining a dataset and labels from trauma patients; inputting the dataset and labels into a neural network for training to obtain an evaluation model, wherein the parameter update of the neural network includes a first update rule and a second update rule, and the neural network updates parameters by alternately executing the first update rule and the second update rule. When the model performance of the neural network is lower than the preset performance, the iteration of the neural network is paused through intermittent control and the ratio of the execution times of the first update rule and the second update parameter rule is adjusted before it is iterated again until the model converges or the preset number of iterations is completed. The method of the present application can efficiently and accurately give the degree of trauma of the patient, and has great clinical value.

Description

Technical Field

[0001] The present application relates to the field of intelligent medical care, and specifically to a trauma level assessment method, device, program product, and computer-readable storage medium. Background Art

[0002] With the rapid development of the automobile industry and increased infrastructure, the incidence of traumatic accidents has been increasing annually. Trauma is the leading cause of death in people under 45 and the leading cause of disability in those under 65. Once trauma occurs, the injuries are complex, with high rates of disability and mortality, placing a heavy burden on society and families. Epidemiological studies of emergency trauma patients show that falls and traffic accidents are the main causes of injury and mortality. Following trauma, the mortality risk varies from patient to patient, and a high proportion of patients suffer from multiple injuries, making it difficult to quickly and accurately assess the mortality risk. Failure to quickly predict a patient's mortality risk can lead to overtreatment and missed treatment opportunities for potentially curable patients, resulting in serious consequences. Currently, clinical estimates of mortality risk for trauma patients rely primarily on physician experience, based on the effectiveness of medications or initial first aid. This approach is not only time-consuming, subjective, and delayed, but also lacks consistency and accuracy due to individual differences. Summary of the Invention

[0003] In response to the above problems, the present invention proposes a method for assessing trauma level, which specifically includes:

[0004] Obtain data on trauma patients;

[0005] The data is input into the evaluation model to obtain mild, moderate and severe evaluation results; wherein the training process of the evaluation model is:

[0006] Obtain a dataset and labels of trauma patients;

[0007] The data set and labels are input into a neural network for training to obtain an evaluation model, wherein the parameter update of the neural network includes a first update rule and a second update rule. The neural network updates the parameters by alternately executing the first update rule and the second update rule. When the model performance of the neural network is lower than the preset performance, the iteration of the neural network is paused through intermittent control and the ratio of the execution times of the first update rule and the second update parameter rule is adjusted before iterating again until the model converges or the preset number of iterations is executed.

[0008] The first update rule and the second update rule include one or more of the following: batch gradient descent, stochastic gradient descent, momentum update;

[0009] The first update rule and the second update rule also include rule update and chaos update; the rule update is to adjust parameters based on Fourier transform through non-gradient parameter update rules; the chaos update is to generate a chaotic sequence through chaotic mapping and map it to the parameters to be updated to obtain updated parameters.

[0010] The chaotic mapping method is expressed as:

[0011]

[0012] in, For the The chaotic variable of the iteration, For the The chaotic variable of the iteration, is the perturbation intensity; is the order of the chaotic sequence; is the chaos control parameter.

[0013] The training of the neural network also includes parameter optimization, and the parameter optimization algorithms include one or more of the following: ant colony optimization algorithm, genetic annealing algorithm, particle swarm optimization algorithm, whale optimization algorithm, gray wolf optimization algorithm, and Riemann geometry optimization algorithm; wherein, the Riemann geometry optimization algorithm is a Riemannian metric obtained by the gradient flow of the Riemann manifold in the parameter space, the Riemannian gradient is calculated based on the Riemannian metric, and then the optimized parameters are obtained by optimizing the parameters through the Riemannian gradient.

[0014] The performance of the model is judged by the performance improvement rate. When the performance improvement rate meets the preset performance improvement rate, the model performance is judged to be excellent. When the performance improvement rate meets the preset performance improvement rate, the model performance is judged to be poor. When the model performance is poor, the iteration of the neural network is adjusted through intermittent control. The performance improvement rate is expressed as:

[0015]

[0016] in, is the performance improvement rate, For the The loss function value of the iteration; For the The performance improvement rate of the iteration, is the performance change smoothing coefficient.

[0017] The method also includes data expansion, inputting the data set into a generative adversarial neural network for data expansion to obtain an expanded data set, and then inputting the expanded data set into the neural network for training; wherein, the generator in the generative adversarial neural network adjusts the data by calculating the information entropy of the input data to obtain adjusted data, and the generator generates data based on the adjusted data.

[0018] The method also includes recommending treatment plans, and providing corresponding treatment plans and treatment costs based on the evaluation results of mild, moderate and severe injuries; the mild, moderate and severe levels are determined by encoding the data of the degree of trauma and determining the level based on the size of the encoded value.

[0019] An object of the present invention is to provide a computer program product comprising a computer program or instructions, wherein the computer program or instructions are executed by a processor to implement the above-mentioned method for assessing the injury level.

[0020] An object of the present invention is to provide a computer device comprising a memory, a processor, and a computer program or instructions stored in the memory, wherein the computer program or instructions are executed by the processor to implement the above-mentioned trauma level assessment method.

[0021] An object of the present invention is to provide a computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions are executed by a processor to implement the above-mentioned method for assessing the level of injury.

[0022] Advantages of the present invention:

[0023] 1. This invention adaptively adjusts the execution ratio of multiple parameter-specific rules based on changes in neural network model performance during neural network training. This dynamic adjustment mechanism enables the model to better adapt to various data conditions during training and optimize network performance.

[0024] 2. The intermittent characteristics of chaotic systems are utilized in the neural network parameter update rules to optimize the weight and bias updates of the neural network. By alternating between rule updates and chaotic updates, the exploration capability of the parameter space is enhanced, helping the neural network avoid local optimality and improving global optimization performance.

[0025] 3. Define the Riemannian metric tensor in the parameter space of the neural network and use the concepts of Riemannian geometry to optimize the parameters, ensuring that the update process follows the steepest descent path, thereby improving optimization efficiency and model convergence.

[0026] 4. Generative adversarial networks are used for data augmentation. During the initialization phase of the generative adversarial network, the network parameters of the generator and discriminator adopt specific distributions to increase diversity at network startup. This helps explore a wider data distribution and improves the diversity and quality of generated samples. By calculating the information entropy of the data, the generative adversarial network's ability to capture the complex characteristics of trauma patient data is enhanced. By adjusting the information entropy contribution, the model can better adjust the information flow in the trauma patient data, thereby improving the detail and quality of the generated data. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 A schematic flow chart of a method for assessing trauma grade provided by an embodiment of the present invention;

[0029] Figure 2 A schematic diagram of a trauma grade assessment system provided by an embodiment of the present invention;

[0030] Figure 3 A schematic diagram of a trauma grade assessment device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0032] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as S101, S102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, nor do they limit "first" and "second" to be different types.

[0033] Figure 1 A schematic diagram of a method for assessing trauma grade provided by an embodiment of the present invention specifically includes:

[0034] S101: Obtaining data on trauma patients;

[0035] In one embodiment, this embodiment only describes one data modality and data type. For relevant trauma patient data, the sample expansion model and deep learning model training of this application are also applicable. For example, the patient data may also include at least one of patient demographic data, admission diagnosis data, trauma severity data, multiple injury data, past medical history data, clinical treatment data, and surgical condition data. The details are as follows:

[0036] Patient demographic characteristics, including age and sex;

[0037] The characteristic quantities of admission diagnosis data include: whether the patient was comatose before admission and the mechanism of injury;

[0038] The characteristic quantities of trauma degree data include: the severity of injury of the patient's fixed part;

[0039] The characteristic quantities of polytrauma data include: whether the patient has polytrauma;

[0040] The characteristic quantities of past medical history data include: hypertension, coronary heart disease, diabetes, myocardial infarction, peptic ulcer, anemia, and dyspnea;

[0041] The characteristic quantities of clinical treatment data include: whether the patient is comatose, whether he uses a ventilator, whether he needs a blood transfusion, whether he needs to be admitted to the intensive care unit for treatment, and whether he needs to use protein products;

[0042] The characteristic quantities of surgical data include: surgical type, highest surgical level, number of surgeries, and surgical time.

[0043] Among them, the characteristic quantities of patient demographic data include: patient age, gender, etc.;

[0044] The characteristic quantities of the patient's demographic data basically include age and gender. If the patient is a pregnant woman, the characteristic quantities of this data may also include the pregnant woman's gestational status.

[0045] The characteristic quantities of the admission diagnosis data include: whether the patient was comatose before admission (1 = yes, 0 = no);

[0046] Mechanism of injury (1 = traffic accident, 2 = fall, 3 = fall from height, 4 = assault by others, 5 = assault with sharp or blunt objects);

[0047] Trauma severity data features: injury location and severity, where the larger the coding value, the higher the severity of the injury;

[0048] head injury (0 = no, 1 = first degree, 2 = second degree, 3 = third degree, 4 = fourth degree, 5 = fifth degree);

[0049] facial injury (0=no, 1=first degree, 2=second degree);

[0050] Neck injury (0=no, 1=first degree, 2=second degree);

[0051] chest injury (0 = no, 1 = first degree, 2 = second degree, 3 = third degree);

[0052] abdominal injury (0 = no, 1 = first degree, 2 = second degree, 3 = third degree);

[0053] spinal injury (0=no, 1=grade 1, 2=grade 2);

[0054] limb and pelvic ring injuries (0 = no, 1 = first degree, 2 = second degree, 3 = third degree);

[0055] Feature quantity of polytrauma data: whether polytrauma exists (1 = yes, 0 = no);

[0056] Feature quantities of past medical history data: hypertension (1=yes, 0=no), coronary heart disease (1=yes, 0=no), diabetes (1=yes, 0=no), myocardial infarction (1=yes, 0=no), peptic ulcer (1=yes, 0=no), anemia (1=yes, 0=no), dyspnea (1=yes, 0=no), etc.

[0057] Characteristic quantities of clinical treatment data: whether the patient was in a coma (1=yes, 0=no), whether a ventilator was used (1=yes, 0=no), whether a blood transfusion was given (1=yes, 0=no), whether the patient was admitted to the intensive care unit for treatment (1=yes, 0=no), whether protein products were used (1=yes, 0=no), etc.

[0058] Feature quantities of surgical data: type of surgery, highest surgical level (1=level 1, 2=level 2, 3=level 3, 4=level 4), number of surgeries, surgical time, etc.

[0059] In a specific embodiment, if the collected data contains missing values, outliers, etc., it is necessary to implement at least one method of missing value filling, normalization processing or feature transformation.

[0060] Missing value filling: The extracted treatment characteristic records of the patient data may be incomplete. This method is used to interpolate missing characteristic values, where continuous characteristics are interpolated using the median value and categorical characteristics are interpolated using the highest frequency level.

[0061] Normalization processing: The extracted feature quantities are normalized and used as feature vectors for predicting the patients to be processed.

[0062] Feature transformation: Use transformation functions such as sin and tan to perform nonlinear transformation on the extracted feature quantities.

[0063] In a specific embodiment, as an optional implementation of the present application, sorting the treatment data related to the trauma treatment decision in chronological order further includes:

[0064] According to the time sequence, the treatment data related to the trauma treatment decision are divided into pre-treatment data and treatment data;

[0065] The treatment data related to the trauma treatment decision are mainly divided into pre-treatment data and in-treatment data to facilitate observation of the corresponding patient data before and after the development of the disease.

[0066] Sorting and marking the pre-treatment data in chronological order to obtain the patient's pre-treatment data;

[0067] When a patient suffers a trauma and has not yet received treatment, the patient's data before treatment can be processed for predictive analysis to obtain the patient's trauma condition progression in the untreated state, such as whether the wound will aggravate infection and cause the patient to remain in a coma. Medical staff can formulate a treatment strategy suitable for the patient based on the patient's pre-treatment condition risk assessment report to minimize the deterioration of the condition.

[0068] The data during treatment are sorted and marked in chronological order to obtain the data during treatment of the patient.

[0069] After a trauma patient receives treatment, the patient data of the trauma patient will be accumulated and recorded, and the risk assessment report of the trauma patient's in-hospital treatment can be predicted according to a fixed time period. However, it should be noted that if new patient data is added during the treatment process, for example, no surgical treatment was performed before a certain treatment, and surgery is required due to the progression of the disease, then the newly added patient data is surgical status data.

[0070] For surgical data, feature quantities related to the surgical situation are recorded. If new patient data is added within the preset data extraction period, it will also be extracted. The extracted data is used for disease risk probability prediction analysis to better reflect the patient's disease progression.

[0071] In a specific embodiment, as an optional implementation scheme of the present application, it also includes: the disease risk assessment report includes at least one of blood creatinine, urine volume, clinical test report, complications, vital signs, and medical scores.

[0072] In one embodiment, any one of the above-mentioned results can largely reflect the condition of the trauma patient. Any one of the above-mentioned data can be input into a trained evaluation model to roughly determine the severity of the trauma patient's condition and provide targeted recommendations for treatment plans for the trauma patient.

[0073] S102: Inputting the data into an assessment model to obtain mild, moderate, and severe assessment results;

[0074] The training process of the evaluation model is as follows:

[0075] Obtain a dataset and labels of trauma patients;

[0076] The data set and labels are input into a neural network for training to obtain an evaluation model, wherein the parameter update of the neural network includes a first update rule and a second update rule. The neural network updates the parameters by alternately executing the first update rule and the second update rule. When the model performance of the neural network is lower than the preset performance, the iteration of the neural network is paused through intermittent control and the ratio of the execution times of the first update rule and the second update parameter rule is adjusted before iterating again until the model converges or the preset number of iterations is executed.

[0077] In one embodiment, the preset performance is the performance in an existing evaluation model or a performance parameter set based on existing experience.

[0078] In one embodiment, the first update rule and the second update rule include one or more of the following: batch gradient descent, stochastic gradient descent, and momentum update.

[0079] In one embodiment, the first update rule and the second update rule further include rule update and chaos update; the rule update is to adjust parameters based on Fourier transform through non-gradient parameter update rules; the chaos update is to generate a chaotic sequence through chaotic mapping and map it to the parameters to be updated to obtain updated parameters.

[0080] The chaotic mapping method is expressed as:

[0081]

[0082] in, For the The chaotic variable of the iteration, For the The chaotic variable of the iteration, is the perturbation intensity; is the order of the chaotic sequence; is the chaos control parameter.

[0083] In one embodiment, the training of the neural network also includes parameter optimization, and the parameter optimization algorithms include one or more of the following: ant colony optimization algorithm, genetic annealing algorithm, particle swarm optimization algorithm, whale optimization algorithm, gray wolf optimization algorithm, and Riemann geometry optimization algorithm; wherein, the Riemann geometry optimization algorithm is a Riemannian metric obtained by the gradient flow of the Riemann manifold in the parameter space, the Riemannian gradient is calculated based on the Riemannian metric, and then the optimized parameters are obtained by optimizing the parameters through the Riemannian gradient.

[0084] In one embodiment, the performance of the model is judged by the performance improvement rate. When the performance improvement rate meets a preset performance improvement rate, the model performance is judged to be excellent. When the performance improvement rate meets the preset performance improvement rate, the model performance is judged to be poor. When the model performance is poor, the iteration of the neural network is adjusted through intermittent control. The performance improvement rate is expressed as:

[0085]

[0086] in, is the performance improvement rate, For the The loss function value of the iteration; For the The performance improvement rate of the iteration, is the performance change smoothing coefficient.

[0087] In one embodiment, the method further includes data expansion, inputting the data set into a generative adversarial neural network for data expansion to obtain an expanded data set, and then inputting the expanded data set into the neural network for training; wherein, the generator in the generative adversarial neural network adjusts the data by calculating the information entropy of the input data to obtain adjusted data, and the generator generates data based on the adjusted data.

[0088] In one embodiment, the method further includes recommending a treatment plan, providing corresponding treatment plans and treatment costs based on the mild, moderate, and severe assessment results; the mild, moderate, and severe levels are determined by encoding the data on the degree of trauma and determining the level based on the size of the encoded value.

[0089] In a specific embodiment, the training data of the present invention comes from the information systems related to the hospital's emergency center and general wards, including data from medical examination equipment such as X-ray machines, CT scanners, and vital signs monitoring equipment. The data collection method adopts real-time digital transmission. The collected data is first transmitted to the data processing center through the internal network, and then converted into a unified JSON format and stored in a highly available distributed file system.

[0090] In a specific embodiment, the collected data is vector data, which may include the following attributes:

[0091] Ra represents the patient's age, Da represents the patient's gender, Pa represents blood pressure, Ta represents body temperature, Ha represents heart rate, Ba represents respiratory rate, Sa represents blood oxygen saturation, Ia represents injury type, La represents injury location, and Oa represents surgical necessity.

[0092] It should be noted that this embodiment is only intended to illustrate one data format and type of the present application. In actual applications, the attributes of the data are usually more than 10 attributes, and the number of attributes of the data may reach dozens or even hundreds.

[0093] Furthermore, the collected data is annotated. The annotating method of this application is manual annotating. In one embodiment, the annotated categories include different disease levels, such as mild, moderate, and severe. Each level corresponds to a corresponding treatment plan, and each treatment plan includes a corresponding treatment cost assessment range, such as:

[0094] Mild: basic treatment, cost range is 1000-3000 yuan;

[0095] Moderate: Intensive treatment, cost range is 3000-15000 yuan;

[0096] Severe: Emergency surgery and intensive care, with costs ranging from 15,000 yuan or more.

[0097] In a specific embodiment, in some cases, the acquisition, labeling, and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor model generalization ability and affect the accuracy of the model. This application uses a generative adversarial network algorithm to generate samples and thus achieve data expansion. The training process of the generative adversarial network algorithm is as follows:

[0098] 1. Initialize the network parameters of the generator and discriminator of the generative adversarial network. The generator is responsible for generating realistic trauma patient data, and the discriminator attempts to distinguish the generated trauma patient data from the real trauma patient data. In the initialization stage, the generator and the discriminator The network parameters are randomly set and expressed as:

[0099]

[0100]

[0101]

[0102]

[0103] Where, is the weight of the generator; is the weight of the discriminator; is the bias of the generator; is the bias of the discriminator; Indicates that it obeys a specific distribution; is the output dimension of the generator; is the input dimension of the discriminator; Represents drawing values from a uniform distribution.

[0104] 2. In the adversarial training phase, the generator and discriminator conduct adversarial training. The generator attempts to generate trauma patient data that can deceive the discriminator, while the discriminator strives to distinguish between the generated trauma patient data and real trauma patient data. In this way, the generator continuously learns how to improve the trauma patient data it generates to make it closer to the distribution of real trauma patient datasets. The adversarial training phase constrains the process through the adversarial loss function, which is expressed as:

[0105]

[0106] Where, It is a sample of real trauma patient data; The distribution of the real trauma patient dataset; is a random noise distribution; is a random noise distribution Random noise extracted from is used to generate trauma patient data; express expectations; represents the discriminator function; represents a generator function; represents the discriminator; Represents a generator; It means that the loss of the constraint generator is the smallest and the loss of the discriminator is the largest.

[0107] In one embodiment, the decision-making method of the discriminator function is expressed as:

[0108]

[0109] Where, is the Sigmoid activation function; for The transpose of is the feature extraction function.

[0110] In this embodiment, the feature extraction function processes complex trauma patient data features through nonlinear transformation, and the implementation is expressed as follows:

[0111]

[0112] Where, is the hyperbolic tangent function; is the first hyperbolic tangent parameter matrix, is the second hyperbolic tangent parameter matrix, and the first hyperbolic tangent parameter matrix and the second hyperbolic tangent parameter matrix are learnable parameter matrices, and parameter learning is performed by gradient descent; yes No. characteristic components; is the number of components of the feature vector of the real trauma patient data sample, corresponding to the projection dimension of the feature vector of the real trauma patient data sample in the feature space. Preferably, Set to 3.

[0113] In this embodiment, the calculation method of the generator function is expressed as:

[0114]

[0115] Where, It is a stacked nonlinear activation function; is the number of eigenvector components of random noise, corresponding to the projection dimension of the eigenvector of random noise in the feature space; The generator corresponding to The weights of the eigenvector components of the random noise; The generator corresponding to The bias of the eigenvector component of the random noise. Preferably, Set to 3.

[0116] In this embodiment, the stacked nonlinear activation function averages the outputs of all activation functions to generate a more complex and delicate trauma patient data feature simulation. Assuming its input is y_cr, the calculation method is expressed as:

[0117]

[0118] Where, It is the input of the stacked nonlinear activation function; The number of feature vector components that are input to the stacked nonlinear activation function; is the first feature vector of the input of the stacked nonlinear activation function Preferably, Set to 3.

[0119] 3. Calculate the information entropy of the data to enhance the ability of the generative adversarial network to capture the characteristics of complex trauma patient data. The calculation method is expressed as:

[0120]

[0121]

[0122] Where, represents the information entropy function based on the second-order entropy contribution; is the composite information entropy function; represents a trauma patient dataset; The data are from the trauma patient dataset; is the probability distribution of trauma patient data in the trauma patient dataset; is the scaling factor for adjusting information entropy; is the set of noise variables input to the generator; is a coefficient that adjusts the contribution of the second-order entropy and is used to increase the complexity and sensitivity of the calculation to better adjust the information flow in trauma patient data; is the weight matrix for trauma patient data.

[0123] In one embodiment, for trauma patient data, For The feature set extracted from is the weight matrix of trauma patient data, is the weight matrix of trauma patient data. elements, corresponding to The Features , the adjustment of the weight matrix for trauma patient data is governed by the following optimization problem:

[0124]

[0125] Where, Indicates that minimize; is the regularization parameter of the weight matrix of trauma patient data, which controls the sparsity of the weights; are the labels for the training trauma patient data; and They are L1 and L2 norms, which are used to increase the generalization ability of the model and reduce overfitting.

[0126] Furthermore, to solve this optimization problem, the gradient descent method is used to update the weight matrix of trauma patient data, which is expressed as:

[0127]

[0128] Where, For the The weight matrix of the trauma patient data for the iteration; For the The weight matrix of the trauma patient data for the iteration; is the adjusted learning rate of the weight matrix for trauma patient data; is a sign function. Preferably, Set to 0.01, Set to 0.3.

[0129] 4. Loss function optimization: The training process of the generator and discriminator is constrained by a composite loss function to guide the generator to more accurately simulate the intrinsic characteristics and statistical properties of trauma patient data from various modalities. The composite loss function uses a square term in a linear form for the adversarial loss term to increase the stability and noise resistance of adversarial learning. The calculation method is expressed as:

[0130]

[0131] Where, is the weight coefficient that controls the adversarial loss; is the weight coefficient that controls the information entropy loss.

[0132] 5. The adversarial training process is iterative until the model converges, that is, the difference between the generated trauma patient data and the real trauma patient data is minimized. During this process, the parameters of the generator and discriminator are continuously adjusted. The adjustment method is expressed as:

[0133]

[0134]

[0135] Where, Indicates parameter update operation; Represents the parameters of the generator; Represents the parameters of the discriminator; is the learning rate of the generative adversarial network; represents the gradient of the generator parameters; represents the gradient of the discriminator parameters.

[0136] In one embodiment, since traditional adversarial training usually uses a fixed learning rate for the generative adversarial network, which may lead to slow or unstable convergence during training, this application uses a dynamic learning rate adjustment mechanism to dynamically adjust the learning rate according to the loss changes during model training to accelerate convergence and improve model stability. The adjustment method is expressed as follows:

[0137]

[0138]

[0139] Where, is the initial learning rate of the generative adversarial network, For the The learning rate of the generative adversarial network at this iteration; To control the sensitivity parameter of learning rate changes; is the difference between the loss of the previous iteration and the loss of the current iteration, For the The loss of the generative adversarial network at this iteration; For the The loss of the generative adversarial network at iterations. Preferably, Set to 3.

[0140] 6. 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.

[0141] After the data augmentation model is trained, the trained data augmentation model is used to increase the number of trauma patient samples. In one embodiment, assuming that the original number of trauma patient samples collected is 800 and the data augmentation model generates 200 trauma patient samples, the expanded trauma patient dataset will contain 1000 samples.

[0142] In a specific embodiment, the expanded set of trauma patients is input into a deep learning model for training. The deep learning model is a neural network model structure, specifically a 6-layer fully connected neural network, and the last layer of the neural network is a Softmax function, which is used to classify the features extracted from the trauma patient data to obtain different levels of illness.

[0143] The training process of the fully connected neural network algorithm is as follows:

[0144] 1. Randomly initialize the weights of the neural network and the bias of the neural network , set the initial number of rule update iterations and the number of chaotic update iterations In one embodiment, in order to increase the diversity of neural network parameter initialization, the combination of Sobol sequence and Halton sequence is used to generate low-discrepancy, high-dimensional initial parameters. Specifically, the Sobol sequence is set to , the Halton sequence is , then the initial weights and biases can be expressed as:

[0145]

[0146]

[0147] Where, is the initial value of the neural network weight parameter; is the initial value of the neural network bias parameter; is pi.

[0148] Furthermore, the process of Sobol sequence generation is expressed as:

[0149]

[0150] Where, For the The binary digit of the bit, is the Sobol sequence length, For the The decimal value of the bit.

[0151] Furthermore, the calculation method of binary bits is expressed as:

[0152]

[0153] Where, Represents the exclusive OR operation, For the Direction coefficient in each direction; is the number of directions of the binary bit, which is the same as the number of rows in the neural network weight matrix; For the The binary digit of the bit.

[0154] Furthermore, the process of generating the Halton sequence is expressed as:

[0155]

[0156] Where, For the the binary digits of prime numbers; For the prime numbers; is the length of the Halton sequence, which is the same as the number of columns of the neural network weight matrix.

[0157] 2. Update the rules and execute the preset number of times The parameter update rule adopts non-gradient parameter update rule and utilizes parameter adjustment based on Fourier transform to capture the frequency characteristics of parameter space. The weights and biases of the neural network for the iteration are and , then the rule update is expressed as:

[0158]

[0159]

[0160] Where, and represent Fourier transform and inverse Fourier transform respectively; is the frequency component; For the The weights of the neural network in the intermediate feature domain of the iteration; For the The bias of the neural network in the intermediate feature domain of the iteration; is the first Fourier hyperparameter, is the second Fourier hyperparameter, is the third Fourier hyperparameter, is the fourth Fourier hyperparameter. Preferably, set ,

[0161] Furthermore, the parameters of the neural network in the intermediate feature domain are used to obtain new weights and biases through nonlinear mapping, which can be expressed as:

[0162]

[0163]

[0164] Where, is the hyperbolic tangent function; is the nonlinearity control parameter, preferably, Set to 2.

[0165] 3. After the rules are updated, the chaos update phase begins and the execution The secondary chaotic parameter update uses chaotic mapping to generate a chaotic sequence, maps it to the parameter adjustment amount, and updates the weights and biases of the neural network nonlinearly. The chaotic mapping method is expressed as:

[0166]

[0167] Where, For the The chaotic variable of the iteration, the initial value of the chaotic variable Pick The normalized value of For the The chaotic variable of the iteration; is the perturbation intensity; is the order of the chaotic sequence; is the chaos control parameter. Preferably, Set to , Set to 0.01, Set to 3.

[0168] Furthermore, chaotic variables are used to update the weights and biases of the neural network, which can be expressed as:

[0169]

[0170]

[0171] Where, is the chaotic update step, preferably, Set to 0.05.

[0172] In one embodiment, the calculation method of the initial value of the chaotic variable is expressed as:

[0173]

[0174] Where, Take the maximum value function.

[0175] Furthermore, chaotic variables are used to update the weights and biases of the neural network. The update method is expressed as:

[0176]

[0177]

[0178]

[0179] Where, For the The adaptive weight coefficient of the iteration, For the Performance improvement rate per iteration; For the The weights of the neural network at the iteration, For the The bias of the neural network at this iteration; is the Riemann gradient learning rate; is the Riemann gradient of the neural network weights; is the Riemann gradient of the bias of the neural network; For the The loss function value of the iteration. Preferably, Set to 0.01.

[0180] 4. Using the parameter optimization method based on Riemannian geometry, the weights and biases of the neural network are optimized according to the gradient flow on the Riemannian manifold and the geometric characteristics of the parameter space. By defining a suitable Riemannian metric in the parameter space, the parameter update process follows the steepest descent path, thereby improving the optimization efficiency and convergence. Specifically, the Riemannian metric tensor of the weights and biases of the neural network is defined in the parameter space and is expressed as:

[0181]

[0182] Where, is the Riemannian metric tensor of the neural network weights and biases, is the first component of the Riemannian metric tensor of the neural network weights and biases, is the second component of the Riemannian metric tensor of the neural network weights and biases, is the third component of the Riemannian metric tensor of the neural network weights and biases, The fourth component of the Riemannian metric tensor for the weights and biases of the neural network.

[0183] Furthermore, the calculation method of each component is expressed as:

[0184]

[0185]

[0186]

[0187] Where, is the Riemann regularization parameter; is the identity matrix, whose dimensions are or Preferably, Set to 0.01.

[0188] Furthermore, the Riemannian gradient is calculated on the Riemannian manifold, and the gradient calculation method is expressed as:

[0189]

[0190]

[0191] Where, is the inverse matrix of the metric tensor; is the Riemann gradient of the neural network's loss function with respect to the weights; is the Riemann gradient of the loss function of the neural network with respect to the bias.

[0192] 5. Perform intermittent control and adaptively adjust the network according to the performance changes in the previous iterations and If the network performance is found to be improving slowly, the proportion of chaotic updates is increased to strengthen the exploration of the parameter space; otherwise, the number of rule updates is increased to consolidate the existing optimization results. The performance change is measured by the performance improvement rate, which is calculated as follows:

[0193]

[0194] Where, For the The loss function value of the iteration; For the The performance improvement rate of each iteration. is the performance change smoothing coefficient, which plays a role in smoothing to avoid the influence of excessive fluctuations. Set to 0.9.

[0195] Furthermore, the calculation method of the loss function value is expressed as:

[0196]

[0197] Where, The number of samples entered for the current batch; is a neural network function that obtains classification labels through the Softmax function; is the first input to the neural network samples; is the loss parameter adjustment factor. Preferably, Set to 0.1.

[0198] Furthermore, according to the performance improvement rate, the number of rule update iterations and chaos update iterations are adjusted, which can be expressed as:

[0199]

[0200] Where, Improve thresholds for performance; is the first adjustment step; is the second adjustment step. Preferably, Set to 0.01, Set to 1, Set to 2.

[0201] 6. 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.

[0202] In one specific embodiment, after a deep learning model for trauma patient condition assessment is trained, it is used to process new samples to assess the patient's condition and recommend appropriate treatment plans. Newly collected trauma patient data is input into the deep learning model for trauma patient condition assessment. This model has been trained and optimized based on a large amount of historical data and is capable of identifying and extracting the most critical information for condition assessment. The extracted features are then classified using the Softmax function in the model's final layer to determine the condition level.

[0203] In a specific embodiment, the obtained disease level includes mild, moderate, and severe. Each level corresponds to a corresponding treatment plan, and each treatment plan includes a corresponding treatment cost assessment range, such as:

[0204] Mild: basic treatment, cost range is 1000-3000 yuan;

[0205] Moderate: Intensive treatment, cost range is 3000-15000 yuan;

[0206] Severe: Emergency surgery and intensive care, with costs ranging from 15,000 yuan or more.

[0207] In one embodiment, the trauma grade assessment process of the present invention involves obtaining trauma data, clinical data, treatment data, and other data from a trauma patient, inputting one or more of these data into a trained assessment model to perform a trauma grade assessment and obtain an assessment result. Different assessment models (and data augmentation models) are trained for different data types. The training process for the assessment model may include any of the following:

[0208] 1 The iteration of the neural network is controlled by an intermittent control mechanism, and the parameters of the neural network are updated using arbitrary update rules.

[0209] 2 The iteration of the neural network is controlled by an intermittent control mechanism, and the parameters of the neural network are updated using rule update + chaos update.

[0210] 3. The iteration of the neural network is controlled by an intermittent control mechanism. The parameters of the neural network are updated using arbitrary update rules, and the parameter selection is optimized through Riemannian geometry.

[0211] 4 The iteration of the neural network is controlled by an intermittent control mechanism. The parameter update of the neural network adopts rule update + chaotic update, and the parameter selection is optimized through Riemannian geometry.

[0212] 5 Conventional generative adversarial networks perform data expansion, control the iteration of the neural network through an intermittent control mechanism, and update the parameters of the neural network using arbitrary update rules.

[0213] 6. Data expansion is performed by optimizing the generative adversarial network generated by calculating the information entropy of the data. The iteration of the neural network is controlled by an intermittent control mechanism, and the parameters of the neural network are updated using arbitrary update rules.

[0214] 7. Data expansion is performed by optimizing the generative adversarial network generated by calculating the information entropy of the data. The iteration of the neural network is controlled by an intermittent control mechanism, and the parameters of the neural network are updated using rule update + chaos update.

[0215] 8. Data expansion is performed by optimizing the generative adversarial network generated by calculating the information entropy of the data. The iteration of the neural network is controlled by an intermittent control mechanism. The parameters of the neural network are updated using rule update + chaotic update, and the parameter selection is optimized through Riemannian geometry.

[0216] The disclosed embodiments of the present invention further provide a computer program product or system, including a computer program, which implements the steps of the above-mentioned trauma level assessment method when executed by a processor.

[0217] Figure 2 A schematic diagram of a trauma grade assessment system provided by an embodiment of the present invention specifically includes:

[0218] Acquisition unit: acquires data of trauma patients;

[0219] Evaluation unit: inputs the data into the evaluation model to obtain mild, moderate, and severe evaluation results; wherein the training process of the evaluation model is:

[0220] Obtain a dataset and labels of trauma patients;

[0221] The data set and labels are input into a neural network for training to obtain an evaluation model, wherein the parameter update of the neural network includes a first update rule and a second update rule. The neural network updates the parameters by alternately executing the first update rule and the second update rule. When the model performance of the neural network is lower than the preset performance, the iteration of the neural network is paused through intermittent control and the ratio of the execution times of the first update rule and the second update parameter rule is adjusted before iterating again until the model converges or the preset number of iterations is executed.

[0222] Figure 3 A schematic diagram of a trauma grade assessment device provided by an embodiment of the present invention specifically includes:

[0223] A memory and a processor; the memory is used to store program instructions; the processor is used to call the program instructions, and when the program instructions are executed, any one of the above-mentioned trauma level assessment methods is performed.

[0224] The disclosed embodiment of the present invention further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs any of the above-mentioned methods for assessing trauma levels.

[0225] The validation results of this validation example demonstrate that assigning inherent weights to indications can improve the performance of the present method compared to the default settings. Those skilled in the art will readily appreciate that, for ease of description and brevity, the specific operating processes of the systems, devices, and units described above can be referenced to the corresponding processes in the aforementioned method embodiments and will not be further elaborated upon here. It should be understood that the disclosed systems, devices, and methods can be implemented in other ways within the several embodiments provided herein. For example, the device embodiments described above are merely illustrative. For example, the division of units described is merely a logical functional division. In actual implementation, other divisions may be employed, such as combining or integrating multiple units or components into another system, or omitting or disabling certain features. Furthermore, the coupling, direct coupling, or communication connection shown or discussed may be through interfaces, indirect coupling, or communication connection between devices or units, and may be electrical, mechanical, or other forms. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the units may be selected to achieve the objectives of the present embodiment as needed. In addition, the functional units in the various embodiments of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware or in the form of software functional units. Those skilled in the art will understand that all or part of the steps in the various methods of the above-mentioned embodiments may be completed by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, which may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0226] Those skilled in the art will understand that all or part of the steps in the above-mentioned embodiment method can be implemented by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The above-mentioned medium storage can be a read-only memory, a disk or an optical disk, etc.

[0227] The above is a detailed introduction to a computer device provided by the present invention. For those skilled in the art, according to the concept of the embodiments of the present invention, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for assessing trauma level, characterized in that: include: Obtain data on trauma patients; The data is input into the evaluation model to obtain mild, moderate and severe evaluation results; wherein the training process of the evaluation model is: Obtain a dataset and labels of trauma patients; Inputting the data set and the label into a neural network for training to obtain an evaluation model, wherein the parameter update of the neural network includes a first update rule and a second update rule, and the neural network updates the parameters by alternately executing the first update rule and the second update rule. When the model performance of the neural network is lower than a preset performance, the iteration of the neural network is paused through intermittent control and the ratio of the execution times of the first update rule and the second update rule is adjusted before iterating again until the model converges or the preset number of iterations is completed; The performance of the model is judged by the performance improvement rate. When the performance improvement rate meets the preset performance improvement rate, the model performance is judged to be excellent. When the performance improvement rate does not meet the preset performance improvement rate, the model performance is judged to be poor. When the model performance is poor, the iteration of the neural network is adjusted through intermittent control, wherein the performance improvement rate is expressed as: in, is the performance improvement rate, For the The loss function value of the iteration; For the The performance improvement rate of the iteration, is the performance change smoothing coefficient.

2. The method for assessing the level of injury according to claim 1, wherein: The first update rule and the second update rule include one or more of the following: batch gradient descent, stochastic gradient descent, momentum update; The first update rule and the second update rule also include rule update and chaos update; the rule update is to adjust parameters based on Fourier transform through non-gradient parameter update rules; the chaos update is to generate a chaotic sequence through chaotic mapping and map it to the parameters to be updated to obtain updated parameters.

3. The method for assessing the level of injury according to claim 2, wherein: The chaotic mapping method is expressed as: in, For the The chaotic variable of the iteration, For the The chaotic variable of the iteration, is the perturbation intensity; is the order of the chaotic sequence; is the chaos control parameter.

4. The method for assessing the level of injury according to claim 1, wherein: The training of the neural network also includes parameter optimization, and the parameter optimization algorithms include one or more of the following: ant colony optimization algorithm, genetic annealing algorithm, particle swarm optimization algorithm, whale optimization algorithm, gray wolf optimization algorithm, and Riemann geometry optimization algorithm; wherein, the Riemann geometry optimization algorithm is a Riemannian metric obtained by the gradient flow of the Riemann manifold in the parameter space, the Riemannian gradient is calculated based on the Riemannian metric, and then the optimized parameters are obtained by optimizing the parameters through the Riemannian gradient.

5. The method for assessing the level of injury according to claim 1, wherein: The method also includes data expansion, inputting the data set into a generative adversarial neural network for data expansion to obtain an expanded data set, and then inputting the expanded data set into the neural network for training; wherein, the generator in the generative adversarial neural network adjusts the data by calculating the information entropy of the input data to obtain adjusted data, and the generator generates data based on the adjusted data.

6. The method for assessing the level of injury according to claim 1, wherein: The method also includes recommending treatment plans, and providing corresponding treatment plans and treatment costs based on the evaluation results of mild, moderate and severe injuries; the mild, moderate and severe levels are determined by encoding the data of the degree of trauma and determining the level based on the size of the encoded value.

7. A computer program product comprising a computer program or instructions, characterized in that The computer program or instructions are executed by a processor to implement the method for assessing the injury level according to any one of claims 1 to 6.

8. A computer device comprising a memory, a processor, and a computer program or instruction stored in the memory, wherein: The computer program or instructions are executed by a processor to implement the method for assessing the injury level according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: The computer program or instructions are executed by a processor to implement the method for assessing the injury level according to any one of claims 1 to 6.

Images