Comorbid network and efficacy co-decision making lung cancer treatment regimen evaluation method and system

Abstract

This invention provides a method and system for evaluating lung cancer treatment plans based on comorbidity networks and collaborative decision-making on treatment efficacy. The method includes: inputting preprocessed clinical data into a comorbidity risk prediction model, including a multimodal feature extraction module, a multimodal feature fusion module, a comorbidity network model based on a graph convolutional neural network, and a classification head; the comorbidity network model based on the graph convolutional neural network determines comorbidity associations based on fused features, using these associations as edges and each disease as a node, and employing a graph convolutional neural network to learn deep-level features from each node; the classification head performs comorbidity risk prediction; MLP prediction is used to obtain efficacy prediction values ​​for various treatment plans to be evaluated; according to comorbidity-efficacy collaborative decision-making rules, the efficacy prediction values ​​of various treatment plans to be evaluated are adjusted and ranked, and the evaluation results of various treatment plans to be evaluated are output. This invention can evaluate lung cancer treatment plans.

Description

Technical Field

[0001] This invention belongs to the field of intelligent medical assessment technology, and specifically refers to a method and system for evaluating lung cancer treatment plans based on comorbidity networks and efficacy co-decision. Background Technology

[0002] Lung cancer, as one of the most common malignant tumors worldwide, has complex pathological features and a high incidence of complex comorbidities, making diagnosis and treatment of lung cancer patients extremely difficult. With the development of medical technology, more and more studies have shown that comorbidities have a significant impact on the treatment and prognosis of lung cancer.

[0003] However, existing lung cancer treatment options often overlook the complex comorbidities, resulting in varying efficacy and side effects (comorbidities) among different treatment options. Therefore, an effective method for evaluating lung cancer treatment options is needed. Summary of the Invention

[0004] To address the technical problems existing in the prior art, the present invention provides a method and system for evaluating lung cancer treatment plans based on comorbidity networks and efficacy synergistic decision-making, the technical solution of which is as follows: On the one hand, a method for evaluating lung cancer treatment plans based on comorbidity networks and synergistic decision-making regarding treatment efficacy is provided. This method includes: S1. Acquire and preprocess clinical diagnosis and treatment data of lung cancer patients; S2. Input the preprocessed clinical diagnosis and treatment data into the trained comorbidity risk prediction model. The comorbidity risk prediction model includes: a multimodal feature extraction module, a multimodal feature fusion module, a comorbidity network model based on graph convolutional neural network, and a classification head. The multimodal feature extraction module extracts features from the preprocessed clinical diagnosis and treatment data; The multimodal feature fusion module fuses the extracted multimodal features to obtain fused features; The co-disease network model based on graph convolutional neural network determines co-disease associations according to the fusion features, uses the co-disease associations as edges, uses each disease as a node, and uses graph convolutional neural network to learn features for each node to obtain deep features. The classification head predicts comorbidity risk based on the deep-level features; S3. Based on the patient's clinical diagnosis and treatment data and various treatment options to be evaluated, combined with the prediction results of the comorbidity risk, a trained lightweight fully connected neural network (MLP) is used as the efficacy prediction model for the treatment options to predict the efficacy prediction values ​​of the various treatment options to be evaluated. S4. Based on the comorbidity risk prediction results and efficacy prediction values, adjust the efficacy prediction values ​​of various treatment options to be evaluated according to the comorbidity-efficacy collaborative decision-making rules. S5. Sort the predicted efficacy values ​​of the various treatment options to be evaluated after adjustment, and output the evaluation results of the various treatment options to be evaluated.

[0005] On the other hand, a lung cancer treatment plan evaluation system that integrates comorbidity networks and efficacy decision-making is provided, the system comprising: The preprocessing module is used to acquire and preprocess clinical diagnosis and treatment data of lung cancer patients; A comorbidity risk prediction model is used to input the preprocessed clinical diagnosis and treatment data into the trained comorbidity risk prediction model. The comorbidity risk prediction model includes: a multimodal feature extraction module, a multimodal feature fusion module, a comorbidity network model based on graph convolutional neural network, and a classification head. The multimodal feature extraction module extracts features from the preprocessed clinical diagnosis and treatment data; The multimodal feature fusion module fuses the extracted multimodal features to obtain fused features; The co-disease network model based on graph convolutional neural network determines co-disease associations according to the fusion features, uses the co-disease associations as edges, uses each disease as a node, and uses graph convolutional neural network to learn features for each node to obtain deep features. The classification head predicts comorbidity risk based on the deep-level features; A lightweight fully connected neural network MLP is used to predict the efficacy of various treatment options to be evaluated based on the patient's clinical diagnosis and treatment data and the prediction results of the comorbidity risk. The trained lightweight fully connected neural network MLP is used as a model to predict the efficacy of the treatment options. The adjustment module is used to adjust the efficacy prediction values ​​of various treatment options to be evaluated based on the comorbidity risk prediction results and efficacy prediction values, according to the comorbidity-efficacy collaborative decision-making rules. The output module is used to sort the adjusted efficacy prediction values ​​of various treatment options to be evaluated and output the evaluation results of various treatment options to be evaluated.

[0006] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to realize the above-mentioned lung cancer treatment plan evaluation method of synergistic decision-making between comorbidity network and efficacy.

[0007] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to realize the above-mentioned method for evaluating lung cancer treatment plans based on synergistic decision-making between comorbidity networks and efficacy.

[0008] The beneficial effects of the technical solution provided by this invention include at least the following: This invention utilizes comorbidity network modeling to represent lung cancer and its related comorbidities in a networked form. This allows for in-depth exploration of complex correlation patterns among multiple diseases, effectively reflecting multidimensional and dynamic relationships, enhancing the understanding of the comorbidity status of lung cancer patients, and improving the accuracy and flexibility of comorbidity prediction. By employing graph convolutional neural networks for deep feature learning, it extracts potential correlations between diseases, improving the accuracy of comorbidity risk prediction. Furthermore, through the synergistic optimization of treatment plan prediction results and comorbidity risk prediction results, it addresses the technical deficiency of "single comorbidity prediction ignoring the impact of treatment." Based on the synergistic decision-making between the comorbidity network and efficacy, it scientifically and accurately evaluates various lung cancer treatment plans to be assessed. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used 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 those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart of a lung cancer treatment plan evaluation method based on comorbidity network and efficacy synergistic decision-making, provided by an embodiment of the present invention. Figure 2 This is a block diagram of a lung cancer treatment plan evaluation system based on comorbidity network and efficacy synergistic decision-making, provided by an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0011] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0012] This invention provides a method for evaluating lung cancer treatment plans based on comorbidity networks and collaborative decision-making on treatment efficacy. This method can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The diagram shown is a flowchart of the method. The processing flow may include the following steps:

[0013] S1. Acquire and preprocess clinical diagnosis and treatment data of lung cancer patients; In one possible implementation, clinical diagnosis and treatment data specifically include: medical orders, treatment information, medical record texts, DICOM images, surgical information, and laboratory results.

[0014] Among them, medical orders refer to the medical instructions issued by doctors to patients, including medication, examination, and treatment arrangements; treatment information refers to the treatment records involving patients, including chemotherapy, radiotherapy, surgery, immunotherapy, and other plans and their implementation; medical record text records are the records written by doctors during the diagnosis and treatment process, including initial medical records, follow-up records, and diagnostic opinions; DICOM imaging refers to the Digital Imaging and Communications in Medicine standard, used to store and transmit medical imaging data; surgical information records detailed information about the patient's surgery, including the type of surgery, procedure, intraoperative findings, and postoperative recovery; and laboratory results are the test results from the patient's biological samples (such as blood, urine, and tissue), used for diagnosis and treatment guidance.

[0015] The specific details of obtaining clinical diagnosis and treatment data for lung cancer patients in S1 are as follows: Clinical diagnosis and treatment data of lung cancer patients were extracted from hospital information systems and electronic medical records.

[0016] Among them, the Hospital Information System (HIS) refers to the comprehensive system within a hospital used to manage patient medical records, finances, resource allocation, and other information, while the Electronic Medical Record (EMR) refers to patient medical information stored in a digital manner, including medical record texts, test results, imaging reports, etc.

[0017] It should be noted that extracting clinical diagnosis and treatment data of lung cancer patients from the Hospital Information System (HIS) and Electronic Medical Record (EMR) can efficiently and comprehensively obtain multimodal information of patients (such as medical orders, images, test results, etc.). This automated data extraction method improves the efficiency of information collection, avoids errors caused by manual entry, and ensures the integrity and real-time nature of the data.

[0018] The specific clinical data for pre-treated lung cancer patients in S1 are as follows: Preprocessing of clinical diagnosis and treatment data includes data cleaning and data standardization.

[0019] Data cleaning refers to processing missing, outlier, and duplicate values ​​in clinical data to ensure data integrity and consistency. Data standardization is the process of converting data into a form with the same distribution and scale for unified analysis.

[0020] It should be noted that preprocessing clinical diagnosis and treatment data through data cleaning and standardization can effectively improve the quality and usability of the data. The cleaned data eliminates errors and noise, ensuring the reliability of the analysis results. Standardization processing allows data with different features to be compared on the same scale, avoiding interference from feature differences on model learning.

[0021] Data cleaning specifically includes: Interpolation algorithms or mean substitution methods are used to fill in missing clinical diagnosis and treatment data; Abnormal clinical diagnosis and treatment data were removed using a filtering method; Data standardization specifically refers to:

[0022] in, This represents standardized clinical diagnosis and treatment data. x i This represents clinical diagnosis and treatment data. μ This represents the mean of clinical diagnosis and treatment data. σ This represents the standard deviation of clinical diagnosis and treatment data.

[0023] S2. Input the preprocessed clinical diagnosis and treatment data into the trained comorbidity risk prediction model. The comorbidity risk prediction model includes: a multimodal feature extraction module, a multimodal feature fusion module, a comorbidity network model based on graph convolutional neural network, and a classification head. The multimodal feature extraction module extracts features from the preprocessed clinical diagnosis and treatment data; Optionally, the processing procedure of the multimodal feature extraction module includes: Using the BERT model to extract text features (The dimension of this embodiment is 768), and the ResNet50 model is used to extract image features. (The dimensions of this invention are 2048), and Z-score standardization is used to extract numerical features of laboratory results and treatment information. (In this embodiment of the invention, the dimension is K, where K is the number of numerical indicators).

[0024] The multimodal feature fusion module fuses the extracted multimodal features to obtain fused features; The processing procedure of the multimodal feature fusion module includes: The fused feature is obtained by concatenating the text features, image features, and numerical features column-wise according to their feature dimensions.

[0025] Where X represents the fused feature, and concat represents the feature column concatenation function, which concatenates text features, image features, and numerical features according to their dimensions. If the dimensions of the features are inconsistent, the text features and image features are first mapped to the same dimensions as the numerical features through a fully connected layer before concatenation. f text ( T ) represents text features, f image ( I ) represents image features, f num ( N ) represents numerical characteristics.

[0026] The co-disease network model based on graph convolutional neural network determines co-disease associations according to the fusion features, uses the co-disease associations as edges, uses each disease as a node, and uses graph convolutional neural network to learn features for each node to obtain deep features. Optionally, the processing procedure of the co-morbidity network model based on graph convolutional neural networks includes: S21. Based on the fusion features, the co-morbidity associations are determined using the support formula and confidence formula, and the initial edge weights of the co-morbidity associations are obtained. :

[0027]

[0028] Where Support represents the support function, Confidence represents the confidence function, Count represents the calculation function, A and B represent disease combinations, and Total represents the total number of patients. Support threshold (based on industry standards, default) =0.05), filter The disease associations are weighted and fused, with β being the weighting coefficient of confidence to make high-confidence associations stand out (usually β=2). S22. Treat the aforementioned co-disease associations as edges. Represents the initial edge weights; each disease is used as a node to establish a preliminary comorbidity network; S23. Perform cluster analysis on the preliminary co-disease network using the K-means clustering algorithm:

[0029] in, J This represents the cluster analysis results. x i Indicates the first i The feature vector of each patient consists of comorbidity-related features extracted through feature fusion.c i Indicates the first i The cluster to which the feature vectors of each patient belong. Indicates the first c i The center point of each cluster, , n This represents the total number of patients participating in the cluster; The clustering analysis result J is used to supplement the comorbidity patterns of diseases, enrich the node attributes, and make the constructed comorbidity network more closely resemble the dynamic association network in clinical practice. Based on the clustering analysis result J, comorbidity pattern labels are assigned to patients (such as label 1 = "lung cancer + hypertension + coronary heart disease", label 2 = "lung cancer + diabetes + chronic kidney disease", etc.). The most frequent disease combinations among all the comorbidity pattern labels of patients are counted, and a node importance attribute is added to each disease in the combination to represent the importance of the disease node. S24. Based on the comorbidity association and clustering analysis results, construct a complete comorbidity network; S25. Perform feature weighting on the complete co-disease network based on the initial edge weights. The final edge weights are calculated using the feature weighting formula to determine the comorbidity network model:

[0030]

[0031] in, S ij Indicates disease i and disease j The comorbidity score represents the strength of the association. σ This represents the activation function. For disease i and disease j The initial edge weights, The clustering results are normalized to [0,1]. and These are the maximum and minimum values ​​of the clustering results, respectively. The weighting coefficients for the clustering results are determined through cross-validation. For bias terms; S26. Based on the comorbidity score between the two diseases. S ij Valid disease associations are screened using a threshold τ, and the selected valid disease associations are used to construct a comorbidity network adjacency matrix. :

[0032] in Indicates disease With disease There is a valid comorbidity association. This indicates that there is no valid association. The screening threshold is set based on the average association strength of the comorbidity network; Apply the adjacency matrix Obtain the set of neighboring nodes of the target lung cancer node:

[0033] in Targeting lung cancer nodes, For nodes The set of neighboring nodes, For all association strengths with the target lung cancer nodes comorbidity nodes; S27. A graph convolutional neural network (GCN) is used to learn features from each node to obtain deep features F. These deep features F refer to the node embedding vectors output by the comorbidity network model, including potential association information between lung cancer and various comorbidities, and fusion information of the patient's multimodal clinical features. Specifically, they include: S27-1. Learn deep features through GCN to obtain individual nodes. The feature representation is calculated using the following formula:

[0034] in, Represents a node In the convolutional neural network k Feature representation in layers, σ Represents the ReLU activation function. W (k) Indicates the first k The learnable weight matrix of the layer, AGG The mean aggregation function representing the features of neighboring nodes. N ( v ) represents a node The set of neighboring nodes, Representing neighboring nodes u In the k Feature representation in the -1 layer; S27-2, Optimizing a single node through a multi-head attention mechanism. The feature representation yields a single node. The final embedding representation is calculated using the following formula:

[0035] in, Represents a node The final embedding representation, where softmax represents the activation function. Represents nodes in the multi-head attention hierarchy The query vector, This represents the key vector of neighbor node u in multi-head attention. This represents the value vector of neighbor node u in multi-head attention. , , It is based on the node features of the last layer of GCN. and They are obtained by mapping different learnable weight matrices. The dimension of the key vector. , T This represents the total number of heads receiving multi-head attention. S27-3. Fuse the final embedded representations of all disease nodes by dimension to generate the deep feature. The calculation formula is as follows:

[0036]

[0037] Where D is the number of feature dimensions of F. d∈[1,D], representing the first... The comorbidity association feature dimension is essentially the th dimension of all disease nodes. The fusion results of dimensional features directly reflect the patient's comorbidity patterns and risk propensity. This is a feature fusion operator, where M is the total number of disease nodes. For nodes v Global importance weights, For nodes v The final embedded representation of the first d Dimensional features, It is a D-dimensional vector, corresponding to deep-level features. The number of feature dimensions D.

[0038] The classification head predicts comorbidity risk based on the deep-level features; Optionally, the processing of the classification head includes: Predicting comorbidity risk Essentially, it refers to the risk of treatment-related complications arising from the patient's comorbidity pattern, i.e., the risk of side effects. Through the deep features The mapping yields, i.e.:

[0039] in, For activation function, The feature weight matrix, For deeper features, This is a bias term.

[0040] Optionally, the classification head is a multi-classification head, which outputs the probability of various comorbidity risks respectively.

[0041] S3. Based on the patient's clinical diagnosis and treatment data and various treatment options to be evaluated, combined with the prediction results of the comorbidity risk, a trained lightweight fully connected neural network (MLP) is used as the efficacy prediction model for the treatment options to predict the efficacy prediction values ​​of the various treatment options to be evaluated. Optionally, the processing procedure of the lightweight fully connected neural network MLP includes: S31. Combining the patient's clinical diagnosis and treatment data with the proposed various treatment plans to be evaluated, obtain the initial efficacy prediction values ​​for each treatment plan to be evaluated. :

[0042] in These are the feature vectors representing the patient's lung cancer characteristics, clinical diagnosis and treatment characteristics, and the initial proposed treatment plan. This is the weight matrix corresponding to the features of lung cancer. This is the weight matrix corresponding to the clinical features. The weight matrix corresponding to the features of the initially proposed treatment plan. This is a bias term used to adjust the linear transformation reference. This is the activation function for the hidden layer, used to filter the negative results after the linear transformation; This is a bias term used to fine-tune the final prediction baseline. The weight vector is a linear transformation used to determine the impact of feature combinations on the final therapeutic effect; Sigmoid is the output layer activation function, which normalizes the calculation result to [0,1]. S32. Based on the prediction results of the aforementioned comorbidity risk, Fine-tuning is performed to obtain the final predicted therapeutic effect. :

[0043] in The comorbidity-related efficacy impact coefficient, based on clinical guidelines, reflects the degree of influence of comorbidity on efficacy, combined with comorbidity risk prediction results. Calculate the final predicted efficacy value. ( The value ranges from [0,1], with values ​​closer to 1 indicating better predictive efficacy.

[0044] S4. Based on the comorbidity risk prediction results and efficacy prediction values, adjust the efficacy prediction values ​​of various treatment options to be evaluated according to the comorbidity-efficacy collaborative decision-making rules. Optionally, the comorbidity-treatment co-decision-making rule includes:

[0045] in This is the adjusted predictive value for therapeutic efficacy. This is a predictor of comorbidity risk. This is a predictive value for therapeutic efficacy.

[0046] S5. Sort the predicted efficacy values ​​of the various treatment options to be evaluated after adjustment, and output the evaluation results of the various treatment options to be evaluated.

[0047] Optionally, the training process of the comorbidity risk prediction model includes: S01. Collect historical clinical diagnosis and treatment data of lung cancer patients and corresponding comorbidity risk labels; The labeling of multiple comorbidity risks is as follows: Comorbidity types: Predefined set of common comorbidities in lung cancer (like =High blood pressure, =diabetes, =Coronary heart disease, =COPD, category H (determined based on historical clinical data) Multi-label vector construction: For each patient, a multi-label vector is generated based on whether they have the comorbidity.

[0048] in: : No. The patient had the first Comorbidity (risk of having this comorbidity); : No. The patient did not have the first Comorbidity (no risk of this comorbidity); Example: The patient also suffers from hypertension ( ) and diabetes ( ),but .

[0049] S02. Divide the historical data into training, validation, and test sets in a 7:2:1 ratio. Use the overall loss function below and the Adam optimization algorithm (learning rate 1e-4~1e-3) to train the model using the training set data to obtain the comorbidity risk prediction model. S03. Use the AUC (area under the curve) of the validation set as the model performance metric. When AUC ≥ 0.85 and there is no improvement after 5 consecutive training rounds, stop training and determine the final model parameters. S04. Validate the model performance on the test set, ensuring that the test set AUC ≥ 0.8, and complete the training of the comorbidity risk prediction model.

[0050] The training process for the efficacy prediction model is similar and will not be described in detail here.

[0051] Optionally, the overall loss function for training the comorbidity risk prediction model and the efficacy prediction model is:

[0052] Where min represents minimization. This indicates the error between the actual and predicted efficacy of a treatment plan. y Labels indicating actual therapeutic effects Indicates the prediction of therapeutic effect. μ This indicates the weighting that balances the efficacy of treatment and the risk of comorbidities. This represents the error between the actual comorbidity risk and the predicted comorbidity risk. s Indicates the actual comorbidity risk label. Indicates the predicted risk of comorbidities; By applying the overall loss function and continuously minimizing the overall loss value, the learnable parameters of the comorbidity risk prediction model and the efficacy prediction model are adjusted in reverse to ensure co-optimization of the two models and constrain the precise and coordinated output of the two models. and This provides reliable input for collaborative decision-making regarding comorbidity and treatment efficacy.

[0053] Optionally, loss of efficacy prediction ,in For patients Actual therapeutic efficacy label, The efficacy value predicted by the model. The total number of samples; Optionally, the multi-comorbidity risk prediction loss adopts the multi-label cross-entropy loss, as shown in the following formula: in, , is the output of the first Multiple comorbidity risk prediction vector for each patient Indicates the patient The first The probability of comorbidity (values ​​[0,1]).

[0054] like Figure 2 As shown in the figure, this invention also provides a lung cancer treatment plan evaluation system based on comorbidity networks and efficacy synergistic decision-making, the system comprising: The preprocessing module 210 is used to acquire and preprocess clinical diagnosis and treatment data of lung cancer patients. Comorbidity risk prediction model 220 is used to input the preprocessed clinical diagnosis and treatment data into the trained comorbidity risk prediction model. The comorbidity risk prediction model includes: a multimodal feature extraction module, a multimodal feature fusion module, a comorbidity network model based on graph convolutional neural network, and a classification head. The multimodal feature extraction module extracts features from the preprocessed clinical diagnosis and treatment data; The multimodal feature fusion module fuses the extracted multimodal features to obtain fused features; The co-disease network model based on graph convolutional neural network determines co-disease associations according to the fusion features, uses the co-disease associations as edges, uses each disease as a node, and uses graph convolutional neural network to learn features for each node to obtain deep features. The classification head predicts comorbidity risk based on the deep-level features; The lightweight fully connected neural network MLP230 is used to predict the efficacy of various treatment options to be evaluated based on the patient's clinical diagnosis and treatment data and various treatment options to be evaluated, combined with the prediction results of the comorbidity risk. The trained lightweight fully connected neural network MLP is used as the efficacy prediction model for the treatment options to predict the efficacy prediction values ​​of the various treatment options to be evaluated. The adjustment module 240 is used to adjust the efficacy prediction values ​​of various treatment options to be evaluated based on the comorbidity risk prediction results and efficacy prediction values, according to the comorbidity-efficacy collaborative decision-making rules. Output module 250 is used to sort the efficacy prediction values ​​of various treatment options to be evaluated after adjustment and output the evaluation results of various treatment options to be evaluated.

[0055] The lung cancer treatment plan evaluation system based on comorbidity network and efficacy co-decision-making provided in this embodiment of the invention has a functional structure that corresponds to the lung cancer treatment plan evaluation method based on comorbidity network and efficacy co-decision-making provided in this embodiment of the invention, and will not be described again here.

[0056] Figure 3This is a schematic diagram of the structure of an electronic device 300 provided in an embodiment of the present invention. The electronic device 300 may vary considerably due to different configurations or performance. It may include one or more central processing units (CPUs) 301 and one or more memories 302. The memory 302 stores at least one instruction, which is loaded and executed by the processor 301 to implement the steps of the lung cancer treatment plan evaluation method of comorbidity network and efficacy synergistic decision-making.

[0057] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions that can be executed by a processor in a terminal to complete the aforementioned method for evaluating lung cancer treatment plans based on comorbidity networks and efficacy co-decision. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device.

[0058] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for evaluating lung cancer treatment plans based on comorbidity networks and synergistic decision-making regarding treatment efficacy, characterized in that, The method includes: S1. Acquire and preprocess clinical diagnosis and treatment data of lung cancer patients; S2. Input the preprocessed clinical diagnosis and treatment data into the trained comorbidity risk prediction model. The comorbidity risk prediction model includes: a multimodal feature extraction module, a multimodal feature fusion module, a comorbidity network model based on graph convolutional neural network, and a classification head. The multimodal feature extraction module extracts features from the preprocessed clinical diagnosis and treatment data; The multimodal feature fusion module fuses the extracted multimodal features to obtain fused features; The co-disease network model based on graph convolutional neural network determines co-disease associations according to the fusion features, uses the co-disease associations as edges, uses each disease as a node, and uses graph convolutional neural network to learn features for each node to obtain deep features. The classification head predicts comorbidity risk based on the deep-level features; S3. Based on the patient's clinical diagnosis and treatment data and various treatment options to be evaluated, combined with the prediction results of the comorbidity risk, a trained lightweight fully connected neural network (MLP) is used as the efficacy prediction model for the treatment options to predict the efficacy prediction values ​​of the various treatment options to be evaluated. S4. Based on the comorbidity risk prediction results and efficacy prediction values, adjust the efficacy prediction values ​​of various treatment options to be evaluated according to the comorbidity-efficacy collaborative decision-making rules. S5. Sort the predicted efficacy values ​​of the various treatment options to be evaluated after adjustment, and output the evaluation results of the various treatment options to be evaluated.

2. The method according to claim 1, characterized in that, The processing steps of the multimodal feature extraction module include: Extracting text features using the BERT model Image features were extracted using the ResNet50 model. Z-score standardization was used to extract numerical features from laboratory results and treatment information. ; The processing procedure of the multimodal feature fusion module includes: The fused feature is obtained by concatenating the text features, image features, and numerical features column-wise according to their feature dimensions. Where X represents the fused feature, and concat represents the feature column concatenation function, which concatenates text features, image features, and numerical features according to their dimensions. If the dimensions of the features are inconsistent, the text features and image features are first mapped to the same dimensions as the numerical features through a fully connected layer before concatenation. f text ( T ) represents text features, f image ( I ) represents image features, f num ( N ) represents numerical characteristics.

3. The method according to claim 1, characterized in that, The processing procedure of the co-morbidity network model based on graph convolutional neural networks includes: S21. Based on the fusion features, the co-morbidity associations are determined using the support formula and confidence formula, and the initial edge weights of the co-morbidity associations are obtained. : Where Support represents the support function, Confidence represents the confidence function, Count represents the calculation function, A and B represent disease combinations, and Total represents the total number of patients. Filtering based on support threshold The disease associations are weighted and fused, with β being the weighting coefficient of confidence to make high-confidence associations stand out more; S22. Treat the aforementioned co-disease associations as edges. Represents the initial edge weights; each disease is treated as a node to establish a preliminary co-disease network; S23. Perform cluster analysis on the preliminary co-disease network using the K-means clustering algorithm: in, J This represents the cluster analysis results. x i Indicates the first i The feature vector of each patient consists of comorbidity-related features extracted through feature fusion. c i Indicates the first i The cluster to which the feature vector of each patient belongs. Indicates the first c i The center point of each cluster, , n This represents the total number of patients participating in the cluster; The clustering analysis result J is used to supplement the comorbidity pattern of the disease, enrich the node attributes, and make the constructed comorbidity network more in line with the dynamic association network in clinical practice. Based on the clustering analysis result J, comorbidity pattern labels are assigned to patients, and the most frequent disease combinations among all comorbidity pattern labels of patients are counted. For each disease in the combination, a node importance attribute is added to represent the importance of the disease node. S24. Based on the comorbidity association and clustering analysis results, construct a complete comorbidity network; S25. Perform feature weighting on the complete co-disease network based on the initial edge weights. The final edge weights are calculated using the feature weighting formula to determine the comorbidity network model: in, S ij Indicates disease i and disease j The comorbidity score represents the strength of the association. σ This represents the activation function. For disease i and disease j The initial edge weights, The clustering results are normalized to [0,1]. and These are the maximum and minimum values ​​of the clustering results, respectively. The weighting coefficients for the clustering results are determined through cross-validation. For bias terms; S26. Based on the comorbidity score between the two diseases. S ij Valid disease associations are screened using a threshold τ, and the selected valid disease associations are used to construct a comorbidity network adjacency matrix. : in Indicates disease With disease There is a valid comorbidity association. This indicates that there is no valid association. The screening threshold is set based on the average association strength of the comorbidity network; Apply the adjacency matrix Obtain the set of neighboring nodes of the target lung cancer node: in Targeting lung cancer nodes, For nodes The set of neighboring nodes, For all association strengths with the target lung cancer nodes comorbidity nodes; S27. A graph convolutional neural network (GCN) is used to learn features from each node to obtain deep features F. These deep features F refer to the node embedding vectors output by the comorbidity network model, including potential association information between lung cancer and various comorbidities, and fusion information of the patient's multimodal clinical features. Specifically, they include: S27-1. Learn deep features through GCN to obtain individual nodes. The feature representation is calculated using the following formula: in, Represents a node In the convolutional neural network k Feature representation in the layer, σ Represents the ReLU activation function. W (k) Indicates the first k The learnable weight matrix of the layer, AGG The mean aggregation function representing the features of neighboring nodes. N ( v ) represents a node The set of neighboring nodes, Representing neighboring nodes u In the k Feature representation in the -1 layer; S27-2, Optimizing a single node through a multi-head attention mechanism. The feature representation yields a single node. The final embedding representation is calculated using the following formula: in, Represents a node The final embedding representation, where softmax represents the activation function. Represents nodes in the multi-head attention hierarchy The query vector, This represents the key vector of neighbor node u in multi-head attention. This represents the value vector of neighbor node u in multi-head attention. , , It is based on the node features of the last layer of GCN. and They are obtained by mapping different learnable weight matrices. The dimension of the key vector. , T This represents the total number of heads receiving multi-head attention. S27-3. Fuse the final embedded representations of all disease nodes by dimension to generate the deep feature. The calculation formula is as follows: Where D is the number of feature dimensions of F. d∈[1,D], representing the first... The comorbidity association feature dimension is essentially the th dimension of all disease nodes. The fusion results of dimensional features directly reflect the patient's comorbidity patterns and risk propensity. This is a feature fusion operator, where M is the total number of disease nodes. For nodes v Global importance weights, For nodes v The final embedded representation of the first d Dimensional features, It is a D-dimensional vector, corresponding to deep-level features. The number of feature dimensions D.

4. The method according to claim 1, characterized in that, The processing of the classification head includes: Predicting comorbidity risk Essentially, it refers to the risk of treatment-related complications arising from the patient's comorbidity pattern, i.e., the risk of side effects. Through the deep features The mapping yields, i.e.: in, For activation function, The feature weight matrix, For deeper features, This is a bias term.

5. The method according to claim 1, characterized in that, The processing steps of the lightweight fully connected neural network (MLP) include: S31. Combining the patient's clinical diagnosis and treatment data with the proposed various treatment plans to be evaluated, obtain the initial efficacy prediction values ​​for each treatment plan to be evaluated. : in These are the feature vectors representing the patient's lung cancer characteristics, clinical diagnosis and treatment characteristics, and the initial proposed treatment plan. This is the weight matrix corresponding to the features of lung cancer. This is the weight matrix corresponding to the clinical features. The weight matrix corresponding to the features of the initially proposed treatment plan. This is a bias term used to adjust the linear transformation reference. This is the activation function for the hidden layer, used to filter the negative results after the linear transformation; This is a bias term used to fine-tune the final prediction baseline. The weight vector is a linear transformation used to determine the impact of feature combinations on the final therapeutic effect; Sigmoid is the output layer activation function, which normalizes the calculation result to [0,1]. S32. Based on the prediction results of the aforementioned comorbidity risk, Fine-tuning is performed to obtain the final predicted therapeutic effect. : in The comorbidity-related efficacy impact coefficient, based on clinical guidelines, reflects the degree of influence of comorbidity on efficacy, combined with comorbidity risk prediction results. Calculate the final predicted efficacy value. .

6. The method according to claim 5, characterized in that, The comorbidity-treatment synergistic decision-making rule includes: in This is the adjusted predictive value for therapeutic efficacy. This is a predictor of comorbidity risk. This is a predictive value for therapeutic efficacy.

7. The method according to claim 6, characterized in that, The overall loss function for training the comorbidity risk prediction model and the efficacy prediction model is: Where min represents minimization. This indicates the error between the actual and predicted efficacy of a treatment plan. y Labels indicating actual therapeutic effects Indicates the prediction of therapeutic effect. μ This indicates the weighting of balancing treatment efficacy and comorbidity risk. This represents the error between the actual comorbidity risk and the predicted comorbidity risk. s Indicates the actual comorbidity risk label. Indicates the predicted risk of comorbidities; By applying the overall loss function and continuously minimizing the overall loss value, the learnable parameters of the comorbidity risk prediction model and the efficacy prediction model are adjusted in reverse to ensure co-optimization of the two models and constrain the precise and coordinated output of the two models. and This provides reliable input for collaborative decision-making regarding comorbidity and treatment efficacy.

8. A lung cancer treatment plan evaluation system based on comorbidity network and efficacy co-decision, characterized in that, The system includes: The preprocessing module is used to acquire and preprocess clinical diagnosis and treatment data of lung cancer patients; A comorbidity risk prediction model is used to input the preprocessed clinical diagnosis and treatment data into the trained comorbidity risk prediction model. The comorbidity risk prediction model includes: a multimodal feature extraction module, a multimodal feature fusion module, a comorbidity network model based on graph convolutional neural network, and a classification head. The multimodal feature extraction module extracts features from the preprocessed clinical diagnosis and treatment data; The multimodal feature fusion module fuses the extracted multimodal features to obtain fused features; The co-disease network model based on graph convolutional neural network determines co-disease associations according to the fusion features, uses the co-disease associations as edges, uses each disease as a node, and uses graph convolutional neural network to learn features for each node to obtain deep features. The classification head predicts comorbidity risk based on the deep-level features; A lightweight fully connected neural network MLP is used to predict the efficacy of various treatment options to be evaluated based on the patient's clinical diagnosis and treatment data and the prediction results of the comorbidity risk. The trained lightweight fully connected neural network MLP is used as a model to predict the efficacy of the treatment options. The adjustment module is used to adjust the efficacy prediction values ​​of various treatment options to be evaluated based on the comorbidity risk prediction results and efficacy prediction values, according to the comorbidity-efficacy collaborative decision-making rules. The output module is used to sort the adjusted efficacy prediction values ​​of various treatment options to be evaluated and output the evaluation results of various treatment options to be evaluated.

9. An electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, characterized in that, The processor loads and executes at least one instruction to implement the lung cancer treatment plan evaluation method for synergistic decision-making on comorbidity networks and efficacy, as described in any one of claims 1-7.

10. A computer-readable storage medium storing at least one instruction, characterized in that, The at least one instruction is loaded and executed by the processor to implement the lung cancer treatment plan evaluation method for synergistic decision-making on comorbidity networks and efficacy as described in any one of claims 1-7.

Images