A method for spinal etiology analysis and risk prediction based on artificial intelligence modeling

Abstract

The present invention discloses a spinal etiology analysis and risk prediction method based on artificial intelligence modeling, which relates to the field of medical artificial intelligence technology, including: building a data fusion platform to collect multimodal data of imaging, genetics, biomarkers and environmental factors; analyzing the imaging data, genetics data, biomarker data and environmental factor data to obtain imaging feature vectors, genetics feature vectors, biological feature vectors and environmental feature vectors; fusing the four features to obtain overall features, and forming a patient health information library based on the overall features; using Q-learning to build a treatment path recommendation system, extracting individual patient information from the patient health information library and inputting it into the system, and outputting the best treatment plan; collecting patient treatment effect data and dynamically adjusting the treatment strategy; building a risk indicator comprehensive evaluation function to analyze real-time data, and once an abnormal risk indicator is detected, immediately starting an early warning mechanism and taking intervention measures.

Description

Technical Field

[0001] The present invention relates to the field of medical artificial intelligence technology, and in particular to a spinal etiology analysis and risk prediction method based on artificial intelligence modeling. Background Art

[0002] With the rapid development of medical research in recent years, artificial intelligence (AI) technology has gradually become a key force in promoting scientific and personalized clinical decision-making, especially in the field of spinal disease diagnosis and treatment. The application of AI has moved from the initial exploration stage to more mature and extensive practical applications. Early AI models mainly relied on a single type of medical data, including imaging and clinical symptom records, to assist doctors in disease identification and prognosis assessment. However, with the advancement of big data and machine learning algorithms, especially the rise of deep learning, AI has begun to demonstrate the ability to integrate multiple data types, thereby being able to more comprehensively reflect the patient's health status.

[0003] Despite this, current AI-assisted spinal disease management still faces many challenges. On the one hand, data fusion and processing are the primary problems. Traditional methods cannot effectively integrate multimodal data from imaging, genetics, biomarkers and environmental factors, which limits the model's predictive ability. On the other hand, the black box nature of AI models reduces their acceptance in clinical practice. Clinicians often require transparent and explainable results to support their decision-making. In addition, existing technologies are slightly insufficient in recommending personalized treatment pathways and lack the ability to dynamically adjust treatment plans based on individual patient differences, which limits the realization of precision medicine. Finally, the lack of continuous monitoring and instant feedback mechanisms makes the AI ​​system limited in risk warning and grasping the timing of intervention, and it is unable to respond to changes in the disease in a timely manner, missing the best treatment window. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a spinal etiology analysis and risk prediction method based on artificial intelligence modeling to solve the problem that traditional methods are difficult to effectively integrate multimodal data of imaging, genetics, biomarkers and environmental factors, resulting in limited predictive ability of the model. On the other hand, the black box characteristics of the AI ​​model reduce its acceptance in clinical practice. Clinicians often need transparent and explainable results to support their decision-making. In addition, the existing technology is slightly insufficient in recommending personalized treatment pathways and lacks the ability to dynamically adjust treatment plans based on individual differences among patients, which limits the realization of precision medicine. Finally, the lack of continuous monitoring and instant feedback mechanisms makes the AI ​​system limited in risk warning and grasping the timing of intervention, and it is unable to respond to changes in the disease in a timely manner, missing the optimal treatment window.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for spinal etiology analysis and risk prediction based on artificial intelligence modeling, which includes building a data fusion platform to collect multimodal data of imaging, genetics, biomarkers, and environmental factors;

[0008] Analyze imaging data, genetic data, biomarker data, and environmental factor data to obtain imaging feature vectors, genetic feature vectors, biological feature vectors, and environmental feature vectors;

[0009] The four features are integrated to obtain the overall features, and a patient health information database is formed based on the overall features;

[0010] Use Q-learning to build a treatment path recommendation system, extract individual patient information from the patient health information database and input it into the system to output the best treatment plan;

[0011] Collect patient treatment effect data and dynamically adjust treatment strategies;

[0012] Construct a comprehensive risk indicator assessment function to analyze real-time data. Once an abnormal risk indicator is detected, immediately activate the early warning mechanism and take intervention measures.

[0013] As a preferred solution of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling described in the present invention, the data fusion platform is constructed to collect multimodal data of imaging, genetics, biomarkers and environmental factors, and the specific steps are as follows:

[0014] Collect multimodal data on imaging, genetics, biomarkers, and environmental factors;

[0015] Perform data cleaning, denoising, format unification and normalization on multimodal data from different sources;

[0016] For imaging data, convolutional neural network CNN is used to extract features and obtain the image feature vector V img ;

[0017] For genetic data, principal component analysis PCA is used to reduce the dimension and obtain the genetic feature vector V gen ;

[0018] For biomarker data, sparse coding technology is used to convert it into a biological feature vector V bio ;

[0019] Wavelet transform WT is used to extract key features from environmental factor data to form an environmental feature vector V env .

[0020] As a preferred embodiment of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling described in the present invention, wherein: the imaging data, genetic data, biomarker data and environmental factor data are analyzed to obtain imaging feature vectors, genetic feature vectors, biological feature vectors and environmental feature vectors, and the specific steps are:

[0021] Set the input image data to X img , the expression is:

[0022] ;

[0023] Among them, O(i, j) represents the element of the feature map located in the i-th row and j-th column, f represents the activation function ReLU, M represents the width of the convolution kernel, and N represents the height of the convolution kernel. (i+m, j+n) represents the pixel value in the i+m row and j+n column, K represents the convolution kernel, K(m,n) represents the weight value of the convolution kernel in the mth row and nth column, and b represents the bias term in the neural network layer;

[0024] Image feature vector V img It is composed of multiple feature maps O, and the expression is:

[0025] ;

[0026] Among them, vec(O C ) represents the feature map O C The operation of expanding into a vector by column, C represents the number of feature maps;

[0027] Assume the input genetic data is X gen , X gen is an m by n matrix, where m represents the number of samples and n represents the number of features;

[0028] Genetic data were processed by principal component analysis (PCA). The PCA process included centering the data, calculating the covariance matrix, and solving the eigenvalues ​​and eigenvectors of the covariance matrix.

[0029] The centralized data expression is:

[0030] ;

[0031] Among them, X gen_c represents the centralized genetic data matrix, μ represents the genetic data matrix X gen The mean vector of ;

[0032] Calculate the covariance matrix, the expression is:

[0033] ;

[0034] Where C represents the sample covariance matrix, and m represents the number of samples;

[0035] Solve for eigenvalues ​​and eigenvectors, the expressions are:

[0036] ;

[0037] Among them, v i represents the i-th eigenvector, λ i Represents and eigenvector v i The corresponding eigenvalues;

[0038] Select the eigenvectors corresponding to the first k largest eigenvalues ​​to form the matrix P, and the value of k can be set according to the needs;

[0039] The genetic characteristic vector is calculated and expressed as:

[0040] ;

[0041] Among them, V gen represents the genetic eigenvector matrix after dimensionality reduction;

[0042] Biometric vector, the specific steps to obtain it are:

[0043] Assume the input biomarker data is X bio , the expression is:

[0044] ;

[0045] Among them, V bio represents the biometric feature vector, D dictionary matrix, i.e., the basis vector set, α represents the coefficient matrix, λ represents the sparsity penalty parameter, To minimize the objective function, it means finding the optimal dictionary matrix and coefficient matrix to satisfy the constraints in the formula;

[0046] Set the input environmental factor data to X env , the expression is:

[0047] ;

[0048] Among them, W X (a, b) represents the coefficients of the wavelet transform at scale a and position b, is the integral symbol, which represents the integral over the entire real number domain (−∞,∞), t is the time variable, which represents any point on the time axis, is the conjugate and scaled translation version of the wavelet function w, which controls the width of the wavelet function w, and b is the position parameter, which controls the position of the wavelet function w on the time axis and is used to detect the local characteristics of the signal at different time points;

[0049] The environmental feature vector is the set of all these key wavelet coefficients, expressed as:

[0050] ;

[0051] Among them, V env represents the environmental feature vector, m represents the number of scale parameters, and n represents the number of location parameters.

[0052] As a preferred solution of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling described in the present invention, wherein: the four features are fused to obtain an overall feature, and a patient health information database is formed based on the overall feature, and the specific steps are as follows:

[0053] The weighted fusion strategy is used to fuse the four extracted features, and the expression is:

[0054] ;

[0055] Among them, F represents the overall feature after fusion, w img 、w gen 、w bio 、w env are the weight coefficients for imaging, genetics, biomarkers, and environmental factors data, respectively;

[0056] The KS-test method was used to calculate the cumulative distribution functions of imaging, genetics, biomarkers, and environmental factors for each modality of data;

[0057] Randomly combine two groups of modal data into a pair, use KS-test to compare and analyze the means of the two groups of data, and set the dynamic threshold;

[0058] When the probability value of the modal data is lower than the set dynamic threshold, it means that there are data quality problems in the distribution of the two sets of data. In this case, the isolation forest method is used to eliminate the outliers in the two sets of data.

[0059] The fused overall characteristics and the patient's medical history, age, and gender are collected and stored in a cloud database to form a patient health information database.

[0060] As a preferred solution of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling described in the present invention, wherein: the treatment path recommendation system is constructed using Q-learning, individual patient information is extracted from the patient health information database and input into the system, and the optimal treatment plan is output. The specific steps are:

[0061] Select imaging features, genetic features, biomarker features, environmental factor features, and the patient's medical history, age, and gender from the patient health information database to form a state space S. The state space contains several states, each of which is set to s, where each state represents the patient's comprehensive health status at a certain point in time;

[0062] The action space consisting of four treatment strategies, namely, drug therapy, physical therapy, surgical advice, and lifestyle adjustment, is set as A. The action space contains several actions, each of which is set as a, where each action corresponds to a treatment plan;

[0063] The reward function is constructed based on the state space and action space, and the expression is:

[0064] ;

[0065] Among them, R(s, a) represents the output of the reward function, w eff 、w cost 、w side They represent the weight coefficients of the three factors of treatment effect, cost and side effects, E eff (s, a) represents the therapeutic effect of executing action a in state s, C(s, a) represents the cost of executing action a in state s, and S(s, a) represents the side effect of executing action a in state s;

[0066] Using the Q-learning update rule with decay, the expression is:

[0067] ;

[0068] Where Q(s, a) represents the expected cumulative reward for taking action a in state s, α represents the learning rate, and γ represents the discount factor. represents the new state reached after taking action a, are all the actions in the new state, Indicates the next state All possible actions The maximum value in ;

[0069] Create a Q-table based on the state space and action space, and set the initial Q value to 0;

[0070] At each time step t, perform action at according to the current policy and observe the state transition st→st+1 and reward r t ;

[0071] Update the Q value in the Q-table according to the update rule expression;

[0072] Use the ε-greedy strategy to select an action with the highest Q value, iterate continuously, and then gradually reduce random exploration to obtain the optimal action under the current strategy;

[0073] When the update amplitude of the Q-table reaches the set number of iterations, the iteration is stopped and the current strategy is output;

[0074] Set the patient's current multimodal data encoding as state s and input it into the trained Q-table;

[0075] Based on the output of the Q-table, action a with the highest Q value is selected as the recommended treatment plan.

[0076] As a preferred solution of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling described in the present invention, wherein: the patient treatment effect data is collected and the treatment strategy is dynamically adjusted, and the specific steps are:

[0077] Collect real-time data;

[0078] Use Apache Flink to process real-time data;

[0079] storing the processed data in the patient health information database;

[0080] Use stochastic gradient descent SGD to continuously update model parameters, and the update expression is:

[0081] ;

[0082] in, represents the updated model parameters, w t represents the current model parameters, η t represents the learning rate that changes over time, ▽w t L(w t ;D t ) indicates that based on the current data D t In the parameter w t The gradient of the loss function on ;

[0083] The treatment effect quantification function is introduced to combine the patient's physical recovery and patient satisfaction to obtain the treatment effect. The expression is:

[0084] ;

[0085] Among them, E eff (t) is the quantitative function of treatment effect, N is the total number of evaluation indicators included in the evaluation, and w i represents the weight of the i-th evaluation index, e i (t) represents the specific value of the i-th evaluation index at time point t;

[0086] Using the adaptive learning rate adjustment mechanism, the treatment strategy is dynamically adjusted according to the treatment effect quantification function, and the expression is:

[0087]

[0088] Among them, A new Indicates new treatment strategies, represents the element that finds the maximum value in the treatment strategy of the action space, Q(s t , a) represents based on the current state s t and the Q value of action a, △Q represents the change in Q value;

[0089] According to the latest treatment strategy A new and multimodal data of patients to recommend treatment options;

[0090] When recommending a plan, the patient's preferences, financial ability, and treatment history are comprehensively considered and evaluated through a multi-objective optimization function, which is expressed as:

[0091] ;

[0092] Among them, MOP represents the optimization target value, C(t) is the cost function, S(t) is the side effect function, P(t) is the patient satisfaction function, α, β, γ, and δ are E eff The weights of (t), C(t), S(t), and P(t);

[0093] The plan with the largest MOP value is selected as the best treatment plan.

[0094] As a preferred solution of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling described in the present invention, wherein: the risk index comprehensive evaluation function is constructed to analyze the real-time data, and the specific steps are:

[0095] Obtain processed real-time data from patient health information repositories;

[0096] The anomaly detection algorithm is used to process real-time data. The expression is:

[0097] ;

[0098] Where AS(t) represents the anomaly score calculated at time point t, x(t) represents the actual observation value at time point t, μ(t) represents the moving average of the observation value at time point t, and σ(t) represents the moving standard deviation of the observation value at time point t;

[0099] Set dynamic thresholds based on historical data for adaptive adjustment. The expression is:

[0100] ;

[0101] Where DT(t) represents the dynamic threshold at time point t, and k is a constant used to control the sensitivity of the dynamic threshold;

[0102] Construct a comprehensive evaluation function of risk indicators, the expression is:

[0103] ;

[0104] Where RL(t) represents the risk level calculated at time point t, N represents the total number of risk indicators included in the assessment, and w i represents the weight of the i-th risk indicator, AS i (t) represents the anomaly score calculated for the i-th risk indicator at time point t.

[0105] As a preferred solution of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling described in the present invention, once an abnormal risk indicator is detected, the early warning mechanism is immediately activated and intervention measures are taken. The specific steps are:

[0106] Set the warning threshold. When the risk indicator comprehensive evaluation function RL(t) is greater than or equal to the warning threshold, the warning mechanism is immediately activated.

[0107] The setting of warning thresholds is based on the level of risk indicators associated with adverse events in historical data;

[0108] Once the warning mechanism is triggered, the system automatically sends a warning notification to the designated medical team members;

[0109] After receiving the early warning notification, medical staff will develop and implement intervention measures based on the patient's specific situation and guideline recommendations;

[0110] After the intervention is performed, the system continues to monitor the patient’s condition;

[0111] Dynamically adjust warning thresholds and risk indicator weights based on treatment effect feedback;

[0112] Through a continuous closed-loop feedback mechanism, the treatment effect data is collected, and the expression is:

[0113] ;

[0114] in, represents the model parameter vector at time step t+1, L(w t ;D t ) means that at time step t, given parameter w t and dataset D t The loss function under t represents the learning rate that varies over time, Denotes the loss function L at time step t with respect to the model parameter w t gradient.

[0115] In a second aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling as described in the first aspect of the present invention.

[0116] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for spinal etiology analysis and risk prediction based on artificial intelligence modeling as described in the first aspect of the present invention.

[0117] The beneficial effects of the present invention are as follows: by establishing a data fusion platform for comprehensive multi-source data, the present invention not only improves the prediction accuracy of the model, but also enhances the interpretability of the model by integrating SHAP and LIME, so that clinicians can better understand and trust the output of the model. At the same time, a treatment pathway recommendation system is constructed to realize dynamic treatment pathway recommendations based on individual differences, which greatly improves the effect of precision medicine. Finally, the deployment of real-time data analysis and early warning mechanisms ensures active management and early intervention of spinal disease risks, effectively avoiding the deterioration of the disease. The present invention belongs to the field of medical artificial intelligence, and particularly focuses on risk prediction and personalized treatment of spinal diseases, aiming to overcome the limitations of existing technologies and provide patients with more accurate and timely medical services. BRIEF DESCRIPTION OF THE DRAWINGS

[0118] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0119] Figure 1 This is a flow chart of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling in Example 1.

[0120] Figure 2 This is an interval diagram of risk levels in Example 1. DETAILED DESCRIPTION

[0121] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0122] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0123] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0124] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for spinal etiology analysis and risk prediction based on artificial intelligence modeling, comprising the following steps:

[0125] S1 builds a data fusion platform to collect multimodal data on imaging, genetics, biomarkers, and environmental factors;

[0126] Collect multimodal data on imaging, genetics, biomarkers, and environmental factors;

[0127] Perform data cleaning, denoising and format unification on multimodal data from different sources;

[0128] Z-score standardization is used to normalize the data to ensure that data of different scales can be compared and integrated on the same platform. The expression is:

[0129] ;

[0130] Among them, Z i represents the standardized value of the ith data point, X i Represents the original data point value, μ represents the average value of all data points, and σ represents the standard deviation. This formula ensures that all data points fall within the normal distribution range with 0 as the center and a standard deviation of 1, thereby eliminating the dimension effect and improving the comparability of data;

[0131] For imaging data, convolutional neural network CNN is used to extract features and obtain the image feature vector V img ;

[0132] For genetic data, principal component analysis PCA is used to reduce the dimension and obtain the genetic feature vector V gen ;

[0133] For biomarker data, sparse coding technology is used to convert it into a biological feature vector Vbio ;

[0134] Wavelet transform WT is used to extract key features from environmental factor data to form an environmental feature vector V env ;

[0135] Imaging data is obtained by CT medical imaging equipment to obtain imaging data of the spine and related parts. Genetics obtains patient genome sequence data by using whole genome sequencing (WGS). Biomarker data is obtained by blood sample analysis. Environmental factor data is obtained by home environment monitors.

[0136] S2 analyzes imaging data, genetic data, biomarker data, and environmental factor data to obtain imaging feature vectors, genetic feature vectors, biological feature vectors, and environmental feature vectors;

[0137] Image feature vector, the specific steps to obtain it are:

[0138] Set the input image data to X img , the expression is:

[0139] ;

[0140] Among them, O(i, j) represents the element of the feature map located in the i-th row and j-th column, f represents the activation function ReLU, M represents the width of the convolution kernel, and N represents the height of the convolution kernel. (i+m, j+n) represents the pixel value in the i+m row and j+n column, K represents the convolution kernel, K(m,n) represents the weight value of the convolution kernel in the mth row and nth column, and b represents the bias term in the neural network layer;

[0141] Image feature vector V img It is composed of multiple feature maps O, and the expression is:

[0142] ;

[0143] Among them, vec(O C ) represents the feature map O C The operation of expanding into a vector by column, C represents the number of feature maps;

[0144] Genetic feature vector, the specific steps to obtain it are:

[0145] Assume the input genetic data is X gen , X gen is an m by n matrix, where m represents the number of samples and n represents the number of features;

[0146] Genetic data were processed by principal component analysis (PCA). The PCA process included centering the data, calculating the covariance matrix, and solving the eigenvalues ​​and eigenvectors of the covariance matrix.

[0147] The centralized data expression is:

[0148] ;

[0149] Among them, X gen_c represents the centralized genetic data matrix, μ represents the genetic data matrix X gen The mean vector of ;

[0150] Calculate the covariance matrix, the expression is:

[0151] ;

[0152] Where C represents the sample covariance matrix, and m represents the number of samples;

[0153] Solve for eigenvalues ​​and eigenvectors, the expressions are:

[0154] ;

[0155] Among them, v i represents the i-th eigenvector, λ i Represents and eigenvector v i The corresponding eigenvalues;

[0156] Select the eigenvectors corresponding to the first k largest eigenvalues ​​to form the matrix P, and the value of k can be set according to the needs;

[0157] The genetic characteristic vector is calculated and expressed as:

[0158] ;

[0159] Among them, V gen represents the genetic eigenvector matrix after dimensionality reduction;

[0160] Biometric vector, the specific steps to obtain it are:

[0161] Assume the input biomarker data is X bio , the expression is:

[0162] ;

[0163] Among them, V bio represents the biometric feature vector, D dictionary matrix, i.e., the basis vector set, α represents the coefficient matrix, λ represents the sparsity penalty parameter, To minimize the objective function, it means finding the optimal dictionary matrix and coefficient matrix to satisfy the constraints in the formula;

[0164] Environmental feature vector, the specific steps to obtain it are:

[0165] Set the input environmental factor data to X env , the expression is:

[0166] ;

[0167] Among them, W X (a, b) represents the coefficients of the wavelet transform at scale a and position b, is the integral symbol, which represents the integral over the entire real number domain (−∞,∞), t is the time variable, which represents any point on the time axis, is the conjugate and scaled translation version of the wavelet function w, which controls the width of the wavelet function w, and b is the position parameter, which controls the position of the wavelet function w on the time axis and is used to detect the local characteristics of the signal at different time points;

[0168] The environmental feature vector is the set of all these key wavelet coefficients, expressed as:

[0169] ;

[0170] Among them, V env represents the environmental feature vector, m represents the number of scale parameters, and n represents the number of location parameters;

[0171] The closer the value of each element of the eigenvector F is to 0, the closer the feature is to the average level of the overall data; and the larger the value deviates from 0, the more prominent the feature is in the individual and is an indicator of disease risk.

[0172] S3 fuses the four features to obtain the overall feature, and forms a patient health information database based on the overall feature;

[0173] The weighted fusion strategy is used to fuse the four extracted features, and the contribution of different data sources is dynamically adjusted through the adaptive weight allocation algorithm. The expression is:

[0174] ;

[0175] Among them, F represents the overall feature after fusion, w img 、w gen 、w bio 、w env are the weight coefficients for imaging, genetics, biomarkers, and environmental factors data, respectively, which are dynamically adjusted by the adaptive weight assignment algorithm based on the data quality and feedback from the prediction model;

[0176] The KS-test method was used to calculate the cumulative distribution functions of imaging, genetics, biomarkers, and environmental factors for each modality of data;

[0177] Randomly combine two sets of modal data into a pair, use KS-test to compare and analyze the means of the two sets of data, set a dynamic threshold, and test whether their distributions come from the same distribution;

[0178] When the probability value of the modal data is lower than the set dynamic threshold, it means that there are data quality problems in the distribution of the two sets of data;

[0179] When there are data quality issues in the distribution of two sets of data, choose Isolation Forest to remove outliers in the two sets of data;

[0180] The fused overall features are input into the isolation forest model for training, and the outlier labels and anomaly scores are output;

[0181] Based on the output of the model, mark and remove those data points identified as outliers;

[0182] After removing outliers, the data is checked for consistency again to ensure the purity and consistency of the data set;

[0183] The fused overall characteristics and the patient's medical history, age, and gender are combined to form a patient health information database. The patient health information database not only contains static data, but also supports dynamic updates to facilitate real-time reflection of the patient's health status.

[0184] S4 uses Q-learning to build a treatment path recommendation system, extracts individual patient information from the patient health information database and inputs it into the system to output the best treatment plan;

[0185] Select imaging features, genetic features, biomarker features, environmental factor features, and the patient's medical history, age, and gender from the patient health information database to form a state space S. The state space contains several states, each of which is set to s, where each state represents the patient's comprehensive health status at a certain point in time;

[0186] The action space consisting of four treatment strategies, namely, drug therapy, physical therapy, surgical advice, and lifestyle adjustment, is set as A. The action space contains several actions, each of which is set as a, where each action corresponds to a treatment plan;

[0187] The reward function is constructed based on the state space and action space to quantify the effect of taking an action a in state s. The expression is:

[0188] ;

[0189] Among them, R(s, a) represents the output of the reward function, w eff 、w cost 、w side Represent the weight coefficients of treatment effect, cost and side effect, E eff (s, a) represents the treatment effect of executing action a in state s, E eff The value range of (s, a) is between 0 and 1. The higher the value, the better the treatment effect. C(s, a) represents the cost of executing action a in state s. The higher the cost, the higher the value of C(s, a). S(s, a) represents the side effect of executing action a in state s. The side effect is a non-negative value. The larger the side effect, the larger the value of S(s, a).

[0190] Using the Q-learning update rule with decay, the expression is:

[0191] ;

[0192] Among them, Q(s, a) represents the expected cumulative reward for taking action a in state s, α represents the learning rate, and γ represents the discount factor, which is a value between 0 and 1 and is used to measure the current value of future rewards. represents the new state reached after taking action a, are all the actions in the new state, Indicates the next state All possible actions The maximum value in ;

[0193] Create a Q-table based on the state space and action space to store the Q value corresponding to each state and action, and set the initial value of the Q value to 0;

[0194] At each time step t, perform action at according to the current policy and observe the state transition st→st+1 and reward r t ;

[0195] Update the Q value in the Q-table according to the update rule expression;

[0196] Use the ε-greedy strategy to select an action with the highest Q value, iterate continuously, and then gradually reduce random exploration to obtain the optimal action under the current strategy;

[0197] When the update amplitude of the Q-table reaches the set number of iterations, the iteration is stopped and the current strategy is output;

[0198] Set the patient's current multimodal data encoding as state s and input it into the trained Q-table;

[0199] According to the output of the Q-table, select action a with the highest Q value as the recommended treatment plan;

[0200] Based on the feedback on treatment effects, strategies are adjusted dynamically to achieve personalization and optimization of treatment plans.

[0201] S5 collects patient treatment effect data and dynamically adjusts treatment strategies;

[0202] Collect real-time data, including but not limited to imaging changes, new genetic findings, fluctuations in biomarker levels, changes in environmental factors, and patient self-reported lifestyle adjustments and treatment feedback;

[0203] Use Apache Flink to process real-time data to ensure instant analysis and response of data;

[0204] storing the processed data in the patient health information database;

[0205] Use stochastic gradient descent SGD to continuously update model parameters, and the update expression is:

[0206] ;

[0207] in, represents the updated model parameters, w t represents the current model parameters, η t represents the learning rate that changes over time, ▽w t L(w t ;D t ) indicates that based on the current data D t In the parameter w t The gradient of the loss function on ;

[0208] The treatment effect quantification function is introduced to combine the patient's physical recovery and patient satisfaction to obtain the treatment effect. The expression is:

[0209] ;

[0210] Among them, E eff (t) is the quantitative function of treatment effect. The higher the value, the better the treatment effect. N represents the total number of evaluation indicators included in the evaluation, including the degree of physical recovery, improvement in quality of life, degree of pain relief, and incidence of complications. i represents the weight of the i-th evaluation index, e i (t) represents the specific value of the i-th evaluation index at time point t;

[0211] Using the adaptive learning rate adjustment mechanism, the treatment strategy is dynamically adjusted according to the treatment effect quantification function, and the expression is:

[0212]

[0213] Among them, A new Indicates new treatment strategies, represents the element that finds the maximum value in the treatment strategy of the action space, Q(s t , a) represents based on the current state s t and the Q value of action a, △Q represents the change in Q value;

[0214] According to the latest treatment strategy A new and multimodal data of patients to recommend treatment options;

[0215] When recommending a plan, the patient's preferences, financial ability, and treatment history are comprehensively considered and evaluated through a multi-objective optimization function, which is expressed as:

[0216] ;

[0217] Among them, MOP represents the optimization target value. The larger the value, the better the performance of the scheme under comprehensive evaluation. C(t) is the cost function, S(t) is the side effect function, P(t) is the patient satisfaction function, α, β, γ, and δ are E eff The weights of (t), C(t), S(t), and P(t);

[0218] The plan with the largest MOP value is selected as the best treatment plan;

[0219] Continuously monitor the patient's response to recommended treatment, including physiological response, psychological status, and quality of life;

[0220] Feedback data is input into the data stream processing framework in real time, triggering immediate updating of the model and dynamic adjustment of treatment strategies.

[0221] S6 builds a comprehensive risk indicator evaluation function to analyze real-time data. Once an abnormal risk indicator is detected, the early warning mechanism is immediately activated and intervention measures are taken;

[0222] Obtain processed real-time data from patient health information repositories;

[0223] The anomaly detection algorithm is used to process real-time data. The expression is:

[0224] ;

[0225] Where AS(t) represents the anomaly score calculated at time point t, x(t) represents the actual observation value at time point t, μ(t) represents the moving average of the observation value at time point t, and σ(t) represents the moving standard deviation of the observation value at time point t;

[0226] Set dynamic thresholds based on historical data for adaptive adjustment, used to distinguish normal fluctuations from abnormal signals, with the expression:

[0227] ;

[0228] Where DT(t) represents the dynamic threshold at time point t, which is used to determine whether the data point at this time point exceeds the normal range, thereby identifying potential abnormalities, and k is a constant used to control the sensitivity of the dynamic threshold;

[0229] Construct a comprehensive risk index evaluation function to quantify the risk level of spinal diseases, with the expression:

[0230] ;

[0231] Where RL(t) represents the risk level value calculated at time point t, and the value range of RL(t) is from 0 to 1. The higher the value, the greater the risk. N represents the total number of risk indicators included in the evaluation, w i represents the weight of the i-th risk indicator, and AS i (t) represents the abnormal score of the i-th risk indicator calculated at time point t;

[0232] Risk levels include low risk level, medium risk level, and high risk level;

[0233] When RL(t) ≤ 0.3, it is determined as a low risk level;

[0234] When 0.3 < RL(t) ≤ 0.6, it is determined as a medium risk level;

[0235] When 0.6 < RL(t) ≤ 1, it is determined as a high risk level;

[0236] Set a warning threshold. When the comprehensive risk index evaluation function RL(t) is greater than or equal to the warning threshold, the warning mechanism is immediately activated;

[0237] The setting of the warning threshold is based on the risk indicator levels related to adverse events in historical data;

[0238] Once the warning mechanism is triggered, the system automatically sends a warning notice to the designated medical team members;

[0239] After receiving the warning notice, medical staff formulate and implement intervention measures based on the specific situation of the patient and the guidelines;

[0240] After implementing the intervention measures, the system continues to monitor the patient's condition;

[0241] Based on the feedback of treatment effects, dynamically adjust the warning threshold and the weights of risk indicators;

[0242] Through a continuous closed-loop feedback mechanism, the treatment effect data is collected, and the expression is:

[0243] ;

[0244] in, represents the model parameter vector at time step t+1, L(w t ;D t ) means that at time step t, given parameter w t and dataset D t The loss function under t represents the learning rate that varies over time, Denotes the loss function L at time step t with respect to the model parameter w t The gradient of , time step t represents the iteration round in the training process;

[0245] The closed-loop feedback mechanism enables continuous optimization of the model, ensuring that the early warning mechanism and intervention measures can adapt to changes in patient conditions and updates in medical knowledge;

[0246] Intervention measures include:

[0247] When the risk level is low, patients are followed up regularly, daily self-monitoring is strengthened, any possible symptom changes are recorded and reported, and moderate physical activity and physical therapy are recommended to enhance muscle and bone health;

[0248] When the risk level is medium, the medical team will be notified immediately and further examinations including imaging examinations and biomarker testing will be arranged. At the same time, targeted drug treatment and physical therapy will be taken to prevent the disease from worsening;

[0249] When the risk level is high, the emergency response procedure will be immediately initiated, including emergency medical treatment, hospitalization and surgical preparation. At the same time, the patient and family members will be contacted urgently to inform them of the severity of the illness and provide emergency guidance.

[0250] This embodiment also provides a computer device, which is suitable for the case of spinal etiology analysis and risk prediction method based on artificial intelligence modeling, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the spinal etiology analysis and risk prediction method based on artificial intelligence modeling as proposed in the above embodiment.

[0251] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in wired and wireless ways, and the wireless way can be realized through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad and mouse.

[0252] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by the processor, it implements the spinal etiology analysis and risk prediction method based on artificial intelligence modeling as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination of them, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.

[0253] In summary, the present invention not only improves the prediction accuracy of the model by establishing a data fusion platform for comprehensive multi-source data, but also enhances the interpretability of the model by integrating SHAP and LIME, so that clinicians can better understand and trust the output of the model. At the same time, a treatment pathway recommendation system is constructed to achieve dynamic treatment pathway recommendations based on individual differences, which greatly improves the effect of precision medicine. Finally, the deployment of real-time data analysis and early warning mechanisms ensures active management and early intervention of spinal disease risks, effectively avoiding the deterioration of the disease. The present invention belongs to the field of medical artificial intelligence, and particularly focuses on risk prediction and personalized treatment of spinal diseases, aiming to overcome the limitations of existing technologies and provide patients with more accurate and timely medical services.

[0254] Example 2

[0255] Referring to Table 1, which is the second embodiment of the present invention, in order to further verify the advancement of the present invention, experimental simulation data of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling are provided.

[0256] The experiment selected 100 patients at risk of spinal diseases as research subjects. The data of these patients were divided into training set and test set with a ratio of 7:3. The experiment adopted the following steps:

[0257] Imaging data are obtained through CT medical imaging equipment; genetic data are obtained using whole genome sequencing; biomarker data are obtained through blood sample analysis; and environmental factor data are obtained using home environmental monitors.

[0258] Data cleaning, denoising and format unification; Z-score standardization is used for data normalization.

[0259] Imaging data are used to extract features through CNN; genetic data are reduced in dimension through PCA; biomarker data are transformed through sparse coding technology; and environmental factor data are used to extract features through wavelet transform.

[0260] A weighted fusion strategy is used to dynamically adjust the contribution of different data sources through an adaptive weight allocation algorithm.

[0261] The isolation forest model was used to identify and remove outliers.

[0262] Integrate characteristics and basic patient information to establish a patient health information database.

[0263] Based on the Q-learning algorithm, a treatment pathway recommendation system is constructed.

[0264] Collect patient treatment effect data, use Apache Flink to process real-time data, and continuously update model parameters through stochastic gradient descent.

[0265] Construct a comprehensive evaluation function for risk indicators to monitor patient risk levels.

[0266] The details are shown in Table 1:

[0267] Table 1 Experimental record table

[0268]

[0269] By analyzing the data in the above table, we can observe the following points:

[0270] Radiological characteristics: The values ​​of P001 and P004 deviate greatly from 0, indicating that their radiological characteristics are more prominent and may be indicator factors of spinal diseases.

[0271] Genetic characteristics: The values ​​of P001 and P002 deviate more from 0, indicating that the genetic characteristics are more significant in individuals, suggesting that genetic factors play an important role in disease risk.

[0272] Biomarker characteristics: The values ​​of P001 and P003 deviate more from 0, which may be related to inflammation or metabolism and are potential markers of disease progression.

[0273] Characteristics of environmental factors: The values ​​of P001 and P004 deviate greatly from 0, which may reflect the impact of environmental factors such as living conditions and working environment on disease risk.

[0274] Through this embodiment, we can see that the invention can effectively identify and quantify the contribution of different factors to the risk of spinal diseases, thereby providing a scientific basis for the recommendation of personalized treatment paths. For example, for patients with P001, because their imaging, genetic and biomarker characteristics deviate greatly from the average, closer monitoring and targeted treatment plans may be required.

[0275] In addition, the innovation and advantage of this invention lies in its ability to integrate multiple data sources and provide more comprehensive and accurate risk assessments and treatment recommendations through advanced data processing and machine learning techniques. Compared with traditional single data source analysis, this method can more accurately reflect the patient's true health status, thereby improving treatment outcomes and the patient's quality of life.

[0276] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for spinal etiology analysis and risk prediction based on artificial intelligence modeling, characterized in that: include, Build a data fusion platform to collect multimodal data from imaging, genetics, biomarkers, and environmental factors; Analyze imaging data, genetic data, biomarker data, and environmental factor data to obtain imaging feature vectors, genetic feature vectors, biological feature vectors, and environmental feature vectors; The four features are integrated to obtain the overall features, and a patient health information database is formed based on the overall features; Use Q-learning to build a treatment path recommendation system, extract individual patient information from the patient health information database and input it into the system to output the best treatment plan; Collect patient treatment effect data and dynamically adjust treatment strategies; Construct a comprehensive risk indicator evaluation function to analyze real-time data. Once an abnormal risk indicator is detected, the early warning mechanism is immediately activated and intervention measures are taken; The data fusion platform is constructed to collect multimodal data of imaging, genetics, biomarkers and environmental factors, and the specific steps are as follows: Collect multimodal data on imaging, genetics, biomarkers, and environmental factors; Perform data cleaning, denoising, format unification and normalization on multimodal data from different sources; For imaging data, convolutional neural network CNN is used to extract features and obtain image feature vector Vimg; For genetic data, principal component analysis PCA is used to reduce the dimension and obtain the genetic feature vector Vgen; For biomarker data, sparse coding technology is used to convert it into a biological feature vector Vbio; Wavelet transform WT is used to extract key features from environmental factor data to form environmental feature vector Venv; The imaging data, genetic data, biomarker data and environmental factor data are analyzed to obtain imaging feature vectors, genetic feature vectors, biological feature vectors and environmental feature vectors. The specific steps are: Assume the input image data is Ximg, the expression is: ； Among them, O(i, j) represents the element of the feature map located in the i-th row and j-th column, f represents the activation function ReLU, M represents the width of the convolution kernel, and N represents the height of the convolution kernel. (i+m, j+n) represents the pixel value in the i+m row and j+n column, K represents the convolution kernel, K(m,n) represents the weight value of the convolution kernel in the mth row and nth column, and b represents the bias term in the neural network layer; The image feature vector Vimg is composed of multiple feature maps O, and the expression is: ； Among them, vec(OC) represents the operation of expanding the feature map OC into a vector by column, and C represents the number of feature maps; Set the input genetic data to Xgen, which is an m-by-n matrix, where m represents the number of samples and n represents the number of features; Genetic data were processed by principal component analysis (PCA). The PCA process included centering the data, calculating the covariance matrix, and solving the eigenvalues ​​and eigenvectors of the covariance matrix. The centralized data expression is: ； Among them, Xgen_c represents the centralized genetic data matrix, μ represents the mean vector of the genetic data matrix Xgen; Calculate the covariance matrix, the expression is: ； Where C represents the sample covariance matrix, and m represents the number of samples; Solve for eigenvalues ​​and eigenvectors, the expressions are: ； Among them, vi represents the i-th eigenvector, λi represents the eigenvalue corresponding to the eigenvector vi; Select the eigenvectors corresponding to the first k largest eigenvalues ​​to form the matrix P, and the value of k can be set according to the needs; The genetic characteristic vector is calculated and expressed as: ； Among them, Vgen represents the genetic feature vector matrix after dimension reduction; Biometric vector, the specific steps to obtain it are: Assume that the input biomarker data is Xbio, and the expression is: ； Among them, Vbio represents the biometric vector, D dictionary matrix, i.e., the basis vector set, α represents the coefficient matrix, λ represents the sparsity penalty parameter, To minimize the objective function, it means finding the optimal dictionary matrix and coefficient matrix to satisfy the constraints in the formula; Set the input environment factor data to Xenv, and the expression is: ； Among them, WX (a, b) represents the coefficients of wavelet transform at scale a and position b. is the integral symbol, which represents the integral over the entire real number domain (−∞,∞), t is the time variable, which represents any point on the time axis, is the conjugate and scaled translation version of the wavelet function w, which controls the width of the wavelet function w, and b is the position parameter, which controls the position of the wavelet function w on the time axis and is used to detect the local characteristics of the signal at different time points; The environmental feature vector is the set of all these key wavelet coefficients, expressed as: ； Where Venv represents the environment feature vector, m represents the number of scale parameters, and n represents the number of location parameters; The four features are integrated to obtain an overall feature, and a patient health information database is formed based on the overall feature. The specific steps are as follows: The weighted fusion strategy is used to fuse the four extracted features, and the expression is: ； Among them, F represents the overall characteristics after fusion, wimg, wgen, wbio, and wenn are the weight coefficients of imaging, genetics, biomarkers, and environmental factors data, respectively; The KS-test method was used to calculate the cumulative distribution functions of imaging, genetics, biomarkers, and environmental factors for each modality of data; Randomly combine two groups of modal data into a pair, use KS-test to compare and analyze the means of the two groups of data, and set the dynamic threshold; When the probability value of the modal data is lower than the set dynamic threshold, it means that there are data quality problems in the distribution of the two sets of data. In this case, the isolation forest method is used to eliminate the outliers in the two sets of data. The integrated overall characteristics and the patient's medical history, age and gender are collected and stored in a cloud database to form a patient health information database; The Q-learning is used to construct a treatment path recommendation system, extract individual patient information from the patient health information database and input it into the system, and output the best treatment plan. The specific steps are as follows: Select imaging features, genetic features, biomarker features, environmental factor features, and the patient's medical history, age, and gender from the patient health information database to form a state space S. The state space contains several states, each of which is set to s, where each state represents the patient's comprehensive health status at a certain point in time; The action space consisting of four treatment strategies, namely, drug therapy, physical therapy, surgical advice, and lifestyle adjustment, is set as A. The action space contains several actions, each of which is set as a, where each action corresponds to a treatment plan; The reward function is constructed based on the state space and action space, and the expression is: ； Among them, R(s, a) represents the output of the reward function, weff, wcost, and wside represent the weight coefficients of the three factors of treatment effect, cost, and side effect, respectively, Eeff(s, a) represents the treatment effect of performing action a in state s, C(s, a) represents the cost of performing action a in state s, and S(s, a) represents the side effect of performing action a in state s; Using the Q-learning update rule with decay, the expression is: ； Where Q(s, a) represents the expected cumulative reward for taking action a in state s, α represents the learning rate, and γ represents the discount factor. represents the new state reached after taking action a, are all the actions in the new state, Indicates the next state All possible actions The maximum value in ; Create a Q-table based on the state space and action space, and set the initial Q value to 0; At each time step t, perform action at according to the current strategy, observe the state transition st→st+1 and reward rt; Update the Q value in the Q-table according to the update rule expression; Use the ε-greedy strategy to select an action with the highest Q value, iterate continuously, and then gradually reduce random exploration to obtain the optimal action under the current strategy; When the update amplitude of the Q-table reaches the set number of iterations, the iteration is stopped and the current strategy is output; Set the patient's current multimodal data encoding as state s and input it into the trained Q-table; According to the output of the Q-table, action a with the highest Q value is selected as the recommended treatment plan.

2. The method for spinal etiology analysis and risk prediction based on artificial intelligence modeling as claimed in claim 1, characterized in that: The specific steps of collecting patient treatment effect data and dynamically adjusting treatment strategies are as follows: Collect real-time data; Use Apache Flink to process real-time data; storing the processed data in the patient health information database; Use stochastic gradient descent SGD to continuously update model parameters, and the update expression is: ； in, represents the updated model parameters, wt represents the current model parameters, ηt represents the learning rate that changes over time, and ▽wtL(wt; Dt) represents the gradient of the loss function on the parameter wt based on the current data Dt; The treatment effect quantification function is introduced to combine the patient's physical recovery and patient satisfaction to obtain the treatment effect. The expression is: ； Among them, Eeff(t) is the quantitative function of treatment effect, N represents the total number of evaluation indicators included in the evaluation, wi represents the weight of the i-th evaluation indicator, and ei(t) represents the specific value of the i-th evaluation indicator at time point t; Using the adaptive learning rate adjustment mechanism, the treatment strategy is dynamically adjusted according to the treatment effect quantification function, and the expression is: Among them, Anew represents a new treatment strategy. It represents the element that finds the maximum value in the treatment strategy of the action space, Q(st, a) represents the Q value based on the current state st and action a, and △Q represents the change in the Q value; Recommend treatment options based on the latest treatment strategies Anew and the patient's multimodal data; When recommending a plan, the patient's preferences, financial ability, and treatment history are comprehensively considered and evaluated through a multi-objective optimization function, which is expressed as: ； Where MOP represents the optimization target value, C(t) is the cost function, S(t) is the side effect function, P(t) is the patient satisfaction function, α, β, γ, δ are the weights of Eeff(t), C(t), S(t), and P(t), respectively; The plan with the largest MOP value is selected as the best treatment plan.

3. The method for spinal etiology analysis and risk prediction based on artificial intelligence modeling as claimed in claim 2, characterized in that: The risk index comprehensive evaluation function is constructed to analyze the real-time data, and the specific steps are as follows: Obtain processed real-time data from patient health information repositories; The anomaly detection algorithm is used to process real-time data. The expression is: ； Where AS(t) represents the anomaly score calculated at time point t, x(t) represents the actual observation value at time point t, μ(t) represents the moving average of the observation value at time point t, and σ(t) represents the moving standard deviation of the observation value at time point t; Set dynamic thresholds based on historical data for adaptive adjustment. The expression is: ； Where DT(t) represents the dynamic threshold at time point t, and k is a constant used to control the sensitivity of the dynamic threshold; Construct a comprehensive evaluation function of risk indicators, the expression is: ； Among them, RL(t) represents the risk level calculated at time point t, N represents the total number of risk indicators included in the assessment, wi represents the weight of the i-th risk indicator, and ASi(t) represents the abnormal score calculated for the i-th risk indicator at time point t.

4. The method for spinal etiology analysis and risk prediction based on artificial intelligence modeling as claimed in claim 3, characterized in that: Once an abnormal risk indicator is detected, the early warning mechanism is immediately activated and intervention measures are taken. The specific steps are as follows: Set the warning threshold. When the risk indicator comprehensive evaluation function RL(t) is greater than or equal to the warning threshold, the warning mechanism is immediately activated. The setting of warning thresholds is based on the level of risk indicators associated with adverse events in historical data; Once the warning mechanism is triggered, the system automatically sends a warning notification to the designated medical team members; After receiving the early warning notification, medical staff will develop and implement intervention measures based on the patient's specific situation and guideline recommendations; After the intervention is performed, the system continues to monitor the patient’s condition; Dynamically adjust warning thresholds and risk indicator weights based on treatment effect feedback; Through a continuous closed-loop feedback mechanism, the treatment effect data is collected, and the expression is: ； in, represents the model parameter vector at time step t+1, L(wt;Dt) represents the loss function given the parameters wt and the dataset Dt at time step t, ηt represents the learning rate that changes over time, represents the gradient of the loss function L with respect to the model parameters wt at time step t.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling as described in any one of claims 1 to 4 are implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the spinal etiology analysis and risk prediction method based on artificial intelligence modeling as described in any one of claims 1 to 4 are implemented.

Images