Kidney cancer risk assessment method, equipment and program product based on graph retrieval

Abstract

The invention relates to the field of intelligent medical treatment, in particular to a kidney cancer risk assessment method and equipment based on graph retrieval and a program product. Comprising the steps of obtaining clinical data of a to-be-tested person; transmitting the clinical data to a graph retrieval model for similarity retrieval to obtain similar case data; and the clinical data and the similar case data are transmitted to an evaluation model to obtain evaluation results of high risk, medium risk and low risk of occurrence of kidney cancer, kidney cancer risk evaluation is performed through a graph retrieval model and a large model, and the method has good clinical value.

Description

Technical Field

[0001] The present application relates to the field of intelligent medical care, and specifically to a kidney cancer risk assessment method based on graph retrieval, a device program product, and a computer-readable storage medium. Background Art

[0002] As a common malignant tumor of the urinary system, the incidence of renal cancer is increasing year by year, and the age of onset tends to be younger. At present, the risk assessment of renal cancer mainly relies on clinical pathological indicators, but there are problems such as low prediction accuracy, easy missed diagnosis or misdiagnosis. With the accumulation of medical big data and the development of artificial intelligence technology, the use of advanced algorithms such as deep learning for tumor risk assessment has become a research hotspot. However, the existing risk assessment models are mostly based on a single data source or algorithm, which makes it difficult to fully capture the complex mechanism of the occurrence and development of renal cancer. In addition, in the modeling process of the existing evaluation model, cluster modeling is often performed based on similarity. For example, CN109924965A is a cardiovascular rehabilitation training system designed for cardiovascular rehabilitation. Rehabilitation training system, in which patient data is collected and then processed and feature extracted, and then cluster modeling is performed based on the similarity between features. When the patient is consulted, the patient data is used for similarity calculation to find similar case groups, and the diagnostic scheme in the similar case group is recommended to the consulting patient. However, cosine similarity is the most intuitive and concise similarity calculation method, and is the most commonly used similarity calculation method. However, for different data, each data involves multiple dimensions, which will consume a lot of computing resources during calculation. Summary of the invention

[0003] In response to the above problems, the present invention proposes a kidney cancer risk assessment model based on graph retrieval, which integrates multi-source medical big data (including genomics, imaging, and clinical medical records), uses a deep learning model for feature extraction and pattern recognition, and combines graph retrieval technology to achieve efficient data association and retrieval, thereby building a high-precision kidney cancer risk assessment system, and making improvements in similarity calculation, thereby improving the accuracy of similarity calculation and reducing the amount of calculation. Specifically, it includes:

[0004] Obtain clinical data of the subjects;

[0005] Inputting the clinical data into a graph retrieval model to perform similarity retrieval to obtain similar case data;

[0006] The clinical data and the similar case data are input into the evaluation model to obtain the evaluation results of high risk, medium risk and low risk of renal cancer.

[0007] The similarity search calculates the weighted similarity between data vectors by weighted cosine similarity to obtain similar case data. The weighted cosine similarity is obtained by assigning different weights to data of different dimensions when calculating the similarity between data vectors of different dimensions.

[0008] Optionally, the calculation of the weighted cosine similarity S_T is expressed as:

[0009]

[0010] Among them, w i is the weight of the i-th dimension, p is the embedding vector of the query graph, and p is the embedding vector of the medical record graph;

[0011] Optionally, the weight of the weighted cosine similarity is dynamically adjusted based on characteristics of the data.

[0012] The similarity search is replaced by calculating the nonlinear similarity between data vectors through nonlinear similarity to obtain similar case data, wherein the nonlinear similarity is to first calculate the similarity of the vectors to obtain the similarity, and then perform data conversion on the similarity through a nonlinear function to obtain the nonlinear similarity;

[0013] Optionally, the calculation of the nonlinear similarity S_L is expressed as:

[0014]

[0015] Among them, k and θ are parameters that control the shape and position of nonlinear similarity, and S_C represents cosine similarity;

[0016] Optionally, the nonlinear similarity is obtained by mapping a similarity model, and the similarity model is obtained by training by learning the relationship between the original similarity and the nonlinear similarity.

[0017] The similarity search is replaced by first obtaining weighted similarity through weighted similarity calculation, then performing nonlinear similarity calculation on the weighted similarity to obtain comprehensive similarity, and retrieving similar case data based on the comprehensive similarity;

[0018] Optionally, the calculation formula of the comprehensive similarity is:

[0019]

[0020] Among them, S_T represents the weighted similarity, k and θ are parameters that control the shape and position of the non-linear similarity.

[0021] The graph retrieval model includes a query graph module, an index module, a calculation module, and a graph database module. The clinical data is input into the query graph module and converted into a query graph. The query graph is input into the index module and parsed into a retrieval feature or a retrieval attribute. According to the retrieval feature or retrieval attribute, the graph data is retrieved in the graph database module and similarity is calculated by the calculation module to obtain similar case data.

[0022] Optionally, the similarity is calculated by a graph embedding method, which is to map graph structure data to vectors in a low-dimensional space, and obtain the similarity by calculating the distance between the vectors.

[0023] Optionally, the graph database construction process of the graph retrieval model is: obtaining a case data set and converting the case data set into graph data to obtain a graph data set; the graph data set includes data entities, entity relationships, and data attributes, and a graph database is constructed based on the graph data set.

[0024] The evaluation model is a large language model, including one or more of the following: GPT2.0, GPT-4, ChatGPT, Wenxin Yiyan, Tongyi Qianwen, and Tencent Hunyuan.

[0025] The evaluation model includes an input layer, a coding module, and an output layer. The coding module is composed of N coding layers, where N is a natural number greater than 1. The coding layer includes a multi-head attention layer, a feedforward neural network layer, and a residual connection and a normalization layer between the multi-head attention layer and the feedforward neural network layer. The input data passes through the input layer and is input to the multi-head attention layer to obtain attention data. The attention data and the input data are fused through a residual connection to obtain attention fusion data. The attention fusion data is input to the feedforward neural network after passing through a normalization layer to obtain feedforward data. The feedforward data and the attention fusion data are fused through a residual connection to obtain feedforward fusion data. The feedforward fusion data passes through the normalization layer and is output through the output layer.

[0026] Optionally, the feedforward neural network includes M linear transformation layers and M-1 activation layers, where M is a natural number greater than or equal to 2.

[0027] The object of the present invention is to provide a computer program product, which includes a computer program or instructions, including: the computer program or instructions are executed by a processor to implement the above-mentioned graph retrieval-based kidney cancer risk assessment method.

[0028] The object of the present invention is to provide a computer device, which includes a memory, a processor and a computer program or instructions stored in the memory, including: the computer program or instructions are executed by the processor to implement the above-mentioned graph retrieval-based kidney cancer risk assessment method.

[0029] The object of the present invention is to provide a computer-readable storage medium having a computer program or instruction stored thereon, including: the computer program or instruction is executed by a processor to implement the above-mentioned graph retrieval-based kidney cancer risk assessment method.

[0030] Advantages of the present invention:

[0031] 1. The present invention proposes a kidney cancer risk assessment method based on graph retrieval, wherein the graph retrieval uses a graph retrieval model to perform similarity calculation to obtain similar case data. The similarity calculation takes into account the importance of different dimensions of different data, assigns weights to different dimensions, improves the calculation of more important dimensional data, reduces the calculation of less important dimensional data, saves computing resources, and improves the high similarity of the retrieved text.

[0032] 2. The present invention takes into account that traditional similarities are basically based on linear relationships, while there are other more complex relationships between data. In order to capture more data relationships, the present invention introduces a nonlinear mapping function to perform similarity calculation, which is used to improve the accuracy of data retrieval and avoid missing relevant data.

[0033] 3. The present invention further combines weighted similarity and nonlinear similarity to retrieve similar data, thereby improving the high similarity of the retrieved data from the weight and nonlinear relationship, thereby obtaining highly similar case information and improving the accuracy of kidney cancer risk estimation.

[0034] 4. The present invention combines a graph retrieval model and a large language assessment model to perform kidney cancer risk assessment. Similar texts are obtained through the graph retrieval model and fused with the current clinical information of the tester and input into the large language assessment model for risk prediction. This can assist doctors in disease prediction and early warning, timely detection and treatment, and has great clinical reference value. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 A schematic diagram of a flow chart of a method for assessing kidney cancer risk based on graph retrieval provided by an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of a kidney cancer risk assessment system based on graph retrieval provided by an embodiment of the present invention;

[0038] Figure 3A schematic diagram of a kidney cancer risk assessment device based on graph retrieval provided by an embodiment of the present invention;

[0039] Figure 4 This is a structural diagram of the evaluation model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0041] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The sequence numbers of the operations, such as S101, S102, etc., are only used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0042] Figure 1 A schematic diagram of a kidney cancer risk assessment method based on graph retrieval provided by an embodiment of the present invention specifically includes:

[0043] S101: Obtain clinical data of the subject;

[0044] In one embodiment, the clinical data includes one or more of the following: basic information, clinical data, imaging data, and gene data.

[0045] S102: inputting the clinical data into a graph retrieval model to perform similarity retrieval to obtain similar case data;

[0046] In one embodiment, the graph retrieval model includes a query graph module, an index module, a calculation module, and a graph database module. The clinical data is input into the query graph module and converted into a query graph. The query graph is input into the index module and parsed into retrieval features or retrieval attributes. The graph data is retrieved in the graph database module based on the retrieval features or retrieval attributes, and the similarity is calculated by the calculation module to obtain similar case data.

[0047] In one embodiment, the similarity is calculated by a graph embedding method, which maps graph structure data to vectors in a low-dimensional space and obtains the similarity by calculating the distance between the vectors.

[0048] In one embodiment, the graph database construction process of the graph retrieval model is: obtaining a case data set and converting the case data set into graph data to obtain a graph data set; the graph data set includes data entities, entity relationships, and data attributes, and a graph database is constructed based on the graph data set.

[0049] In one embodiment, the similarity retrieval obtains similar case data by calculating weighted similarity between data vectors using weighted cosine similarity. The weighted cosine similarity is obtained by assigning different weights to data of different dimensions when calculating similarity between data vectors of different dimensions.

[0050] In one embodiment, the calculation of the weighted cosine similarity S_T is expressed as:

[0051]

[0052] Among them, w_i is the weight of the i-th dimension, p is the embedding vector of the query graph, and p is the embedding vector of the medical record graph.

[0053] In one embodiment, the weight of the weighted cosine similarity is dynamically adjusted based on characteristics of the data.

[0054] In another embodiment, the similarity search calculates the nonlinear similarity between data vectors by a nonlinear similarity to obtain similar case data, wherein the nonlinear similarity is obtained by first calculating the similarity of the vectors, and then converting the similarity by a nonlinear function to obtain the nonlinear similarity. In one embodiment, the calculation of the nonlinear similarity S_L is expressed as:

[0055]

[0056] Wherein, k and θ are parameters controlling the shape and position of nonlinear similarity, and S_C represents cosine similarity. In one embodiment, the nonlinear similarity is obtained by mapping the nonlinear similarity through a similarity model, and the similarity model is obtained by training by learning the relationship between the original similarity and the nonlinear similarity.

[0057] In another embodiment, the similarity search is to first obtain weighted similarity by weighted similarity calculation, then perform nonlinear similarity calculation on the weighted similarity to obtain comprehensive similarity, and obtain similar case data based on comprehensive similarity retrieval; the calculation formula of the comprehensive similarity is:

[0058]

[0059] Among them, S_T represents the weighted similarity, k and θ are parameters that control the shape and position of the non-linear similarity.

[0060] In a specific embodiment, the data processing process of the graph retrieval model is:

[0061] 1) Input:

[0062] The input of graph retrieval mainly includes:

[0063] Query graph: This is a graph structure representing the cases to be retrieved, which contains the characteristic information of the cases (such as the patient's clinical data, living habits, genetic information, etc.), which is encoded as nodes and edges in the graph.

[0064] 2) Processing of graph retrieval model:

[0065] Graph database: This is a database that stores a large number of known case graph structures. Each case is represented as a graph that contains all relevant information about the case.

[0066] The calculation steps of the graph retrieval model are:

[0067] ①Graph matching and indexing:

[0068] Index construction: During the graph database construction phase, each case graph is usually indexed for quick retrieval. The index is based on the attribute information of the graph.

[0069] Query parsing: Parsing the query graph into a set of features or attributes that can be used for retrieval.

[0070] ②Similarity calculation:

[0071] This step is the core of graph retrieval. For the query graph and each case graph in the graph database, the similarity between them needs to be calculated. The similarity calculation method adopts the graph embedding method.

[0072] Graph embedding methods: map the graph structure to vectors in a low-dimensional space and then calculate the distance between these vectors.

[0073] Vector calculation formula:

[0074] Traditional cosine similarity does not consider that the importance of each dimension of the vector may be different. This invention introduces a weighted cosine similarity and dynamically adjusts the weight according to the characteristics of the data, expressed as S_T:

[0075]

[0076] Among them, w i is the weight of the i-th dimension, p is the embedding vector of the query graph, and p is the embedding vector of the medical record graph.

[0077] Result sorting and output: Sort the case images in the graph database according to the similarity calculation results. Select the top 3 case images with the highest similarity as the retrieval result output.

[0078] In another specific embodiment, similarity calculation in a graph retrieval model:

[0079] For the query graph and each case graph in the graph database, the similarity between them needs to be calculated. The similarity calculation method adopts the graph embedding method. The graph embedding method: map the graph structure to vectors in a low-dimensional space, and then calculate the distance between these vectors.

[0080] Calculation formula between vectors, traditional similarity calculation is often based on linear relationships. In order to capture more complex data relationships, the present invention introduces a nonlinear mapping function, proposes a nonlinear similarity calculation, and uses a nonlinear function to transform the original similarity. Alternatively, a nonlinear mapping from the original similarity to the final similarity is learned through a neural network.

[0081] In another specific embodiment, the calculation between similarity vectors in the graph retrieval model: first, weighted cosine similarity is calculated to obtain weighted similarity, then nonlinear similarity is calculated on the weighted similarity to obtain comprehensive similarity, and finally, similar case data is retrieved based on the comprehensive similarity, the similar case data is sorted, and the top three similar case data are output.

[0082] Data construction graph structure: Multi-source data is organized in the form of a graph structure. Nodes represent data entities, including patients' genes and imaging features, and edges represent the relationships between entities, such as the relationship between gene mutations and kidney cancer, and the relationship between imaging features and disease stages.

[0083] S103: Inputting the clinical data and the similar case data into an evaluation model to obtain evaluation results of high risk, medium risk, and low risk of developing renal cancer.

[0084] In one embodiment, the evaluation model is a large language model, including one or more of the following: GPT2.0, GPT-4, ChatGPT, Wenxin Yiyan, Tongyi Qianwen, and Tencent Hunyuan.

[0085] In another embodiment, the evaluation model includes an input layer, an encoding module, and an output layer. The encoding module is composed of N encoding layers, where N is a natural number greater than 1. The encoding layer includes a multi-head attention layer, a feedforward neural network layer, and a residual connection and a normalization layer between the multi-head attention layer and the feedforward neural network layer. The input data passes through the input layer and is input to the multi-head attention layer to obtain attention data. The attention data and the input data are fused through a residual connection to obtain attention fusion data. The attention fusion data is input to the feedforward neural network after passing through a normalization layer to obtain feedforward data. The feedforward data and the attention fusion data are fused through a residual connection to obtain feedforward fusion data. The feedforward fusion data passes through the normalization layer and is output through the output layer.

[0086] Optionally, the feedforward neural network includes M linear transformation layers and M-1 activation layers, where M is a natural number greater than or equal to 2.

[0087] In a specific embodiment, the construction process of the evaluation model is:

[0088] Data collection: Collect multi-source data including genomic data of kidney cancer patients and healthy people, medical imaging data, clinical medical records, etc. Then, define a model to train and learn these data. The defined model structure is:

[0089] (1) Input processing;

[0090] Text encoding and position encoding:

[0091] Assume that the input text has been segmented into a series of tokens, and each token is converted into a fixed-dimensional vector x through the word embedding layer. i , where i is the index of the token.

[0092] Positional encoding i is added to the embedding vector of each token to form the final input vector

[0093] z i =x i +p i

[0094] (2) Encoder layer;

[0095] The encoder is composed of multiple identical layers stacked together, each of which contains two sub-layers: Multi-Head Attention and Feed-Forward Network, as well as residual connections and layer normalization between the two sub-layers.

[0096] 1) Multi-Head Attention;

[0097] Split input: First, the input vector z i Split into multiple "heads", each head performs self-attention calculations independently. Assume there are h heads, and the dimension of each head is:

[0098] d k =d model / h

[0099] where d model is the total dimension of the model.

[0100] Linear transformation: For each head, three different linear transformations are applied, and the corresponding weight matrices are to generate the query (Q), key (K), and value (V) matrices.

[0101]

[0102] Attention score: compute the dot product of Q and K, then apply a scaling factor To zoom:

[0103]

[0104] 2) Residual connection;

[0105] Combine the output of multi-head self-attention with the input z i Perform residual connection.

[0106] M i =A i +z i

[0107] 3) Feed-Forward Network (FFN);

[0108] For each M i , applying a feed-forward neural network consisting of two linear transformations and a ReLU activation function.

[0109] F i =max(0,A i W1)W2

[0110] Among them, W1 and W2 are the weight matrices of the two linear layers of the feedforward neural network.

[0111] Similarly, the output of FFN is compared with A i Perform residual connection to obtain the final output O of this layer i :

[0112] O i =F i +M i

[0113] In a specific embodiment, risk assessment: the medical record information of similar patients obtained by the graph retrieval model is combined with the information of the current patient, and input into a defined large model to perform kidney cancer risk assessment on the individual, and output the patient's risk results, which include high risk, moderate risk, and low risk.

[0114] In one embodiment, the clinical data of the subject is input into a graph retrieval model to calculate the weighted cosine similarity to obtain the first three similar case data, and the first three similar case data and the clinical data of the subject are input into an evaluation model for risk assessment to obtain any one of high, medium and low risks.

[0115] In one embodiment, the clinical data of the subject is input into a graph retrieval model to calculate the first three similar case data through nonlinear cosine similarity, and the first three similar case data and the clinical data of the subject are input into an evaluation model for risk assessment to obtain any one of high, medium and low risks.

[0116] The clinical data of the subject is input into the graph retrieval model to calculate the first three similar case data through weighted cosine similarity and nonlinear similarity. The first three similar case data and the clinical data of the subject are input into the evaluation model for risk assessment to obtain any one of high, medium and low risks.

[0117] The disclosed embodiments of the present invention further provide a computer program product or system, including a computer program, which implements the above-mentioned steps of the method for assessing kidney cancer risk based on graph retrieval when executed by a processor.

[0118] Figure 2 A schematic diagram of a kidney cancer risk assessment system based on graph retrieval provided by an embodiment of the present invention specifically includes:

[0119] Acquisition unit: acquiring clinical data of the subject to be tested;

[0120] Retrieval unit: inputting the clinical data into the graph retrieval model to perform similarity retrieval to obtain similar case data;

[0121] Evaluation unit: input the clinical data and the similar case data into the evaluation model to obtain the evaluation results of high risk, medium risk and low risk of kidney cancer.

[0122] Figure 3 A schematic diagram of a kidney cancer risk assessment device based on graph retrieval provided by an embodiment of the present invention specifically includes:

[0123] A memory and a processor; the memory is used to store program instructions; the processor is used to call the program instructions, and when the program instructions are executed, any one of the above-mentioned graph retrieval-based kidney cancer risk assessment methods is executed.

[0124] The disclosed embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it is any one of the above-mentioned graph retrieval-based kidney cancer risk assessment methods.

[0125] The verification results of this verification embodiment show that assigning inherent weights to indications can improve the performance of the method relative to the default setting. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. The unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units. A person of ordinary skill in the art may understand that all or part of the steps in the various methods of the above-mentioned embodiments may be completed by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0126] A person of ordinary skill in the art can understand that all or part of the steps in the above-mentioned embodiment method can be implemented by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. The above-mentioned medium storage can be a read-only memory, a disk or an optical disk, etc.

[0127] The above is a detailed introduction to a computer device provided by the present invention. For a person skilled in the art, according to the concept of the embodiments of the present invention, there may be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for assessing kidney cancer risk based on graph retrieval, characterized in that: include: Obtain clinical data of the subjects; Inputting the clinical data into a graph retrieval model to perform similarity retrieval to obtain similar case data; The clinical data and the similar case data are input into the evaluation model to obtain the evaluation results of high risk, medium risk and low risk of renal cancer.

2. The method for assessing kidney cancer risk based on graph retrieval according to claim 1, characterized in that: The similarity search calculates the weighted similarity between data vectors by weighted cosine similarity to obtain similar case data. The weighted cosine similarity is obtained by assigning different weights to data of different dimensions when calculating the similarity between data vectors of different dimensions. Optionally, the calculation of the weighted cosine similarity S_T is expressed as: Among them, w i is the weight of the i-th dimension, p is the embedding vector of the query graph, and p is the embedding vector of the medical record graph; optionally, the weight of the weighted cosine similarity is dynamically adjusted based on the characteristics of the data.

3. The method for assessing kidney cancer risk based on graph retrieval according to claim 1, characterized in that: The similarity search calculates the nonlinear similarity between data vectors by nonlinear similarity to obtain similar case data, wherein the nonlinear similarity is obtained by first performing similarity calculation on the vectors, and then performing data conversion on the similarity by a nonlinear function to obtain the nonlinear similarity; Optionally, the calculation of the nonlinear similarity S_L is expressed as: Wherein, k and θ are parameters for controlling the shape and position of nonlinear similarity, and S_C represents cosine similarity; optionally, the nonlinear similarity is obtained by mapping the nonlinear similarity through a similarity model, and the similarity model is trained by learning the relationship between the original similarity and the nonlinear similarity.

4. The method for assessing kidney cancer risk based on graph retrieval according to claim 1, characterized in that: The similarity search is to first obtain weighted similarity by weighted similarity calculation, then perform nonlinear similarity calculation on the weighted similarity to obtain comprehensive similarity, and obtain similar case data based on comprehensive similarity retrieval; Optionally, the calculation formula of the comprehensive similarity is: Among them, S_T represents the weighted similarity, k and θ are parameters that control the shape and position of the non-linear similarity.

5. The method for assessing kidney cancer risk based on graph retrieval according to claim 1, characterized in that: The graph retrieval model includes a query graph module, an index module, a calculation module, and a graph database module. The clinical data is input into the query graph module and converted into a query graph. The query graph is input into the index module and parsed into a retrieval feature or a retrieval attribute. According to the retrieval feature or retrieval attribute, the graph data is retrieved in the graph database module and similarity is calculated by the calculation module to obtain similar case data. Optionally, the similarity is calculated by a graph embedding method, which is to map graph structure data to vectors in a low-dimensional space, and obtain the similarity by calculating the distance between the vectors. Optionally, the graph database construction process of the graph retrieval model is: obtaining a case data set and converting the case data set into graph data to obtain a graph data set; the graph data set includes data entities, entity relationships, and data attributes, and a graph database is constructed based on the graph data set.

6. The method for assessing kidney cancer risk based on graph retrieval according to claim 1, characterized in that: The evaluation model is a large language model, including one or more of the following: GPT2.0, GPT-4, ChatGPT, Wenxin Yiyan, Tongyi Qianwen, and Tencent Hunyuan.

7. The method for assessing kidney cancer risk based on graph retrieval according to claim 1, characterized in that: The evaluation model includes an input layer, a coding module, and an output layer. The coding module is composed of N coding layers, where N is a natural number greater than 1. The coding layer includes a multi-head attention layer, a feedforward neural network layer, and a residual connection and a normalization layer between the multi-head attention layer and the feedforward neural network layer. The input data passes through the input layer and is input to the multi-head attention layer to obtain attention data. The attention data and the input data are fused through a residual connection to obtain attention fusion data. The attention fusion data is input to the feedforward neural network after passing through a normalization layer to obtain feedforward data. The feedforward data and the attention fusion data are fused through a residual connection to obtain feedforward fusion data. The feedforward fusion data passes through the normalization layer and is output through the output layer. Optionally, the feedforward neural network includes M linear transformation layers and M-1 activation layers, where M is a natural number greater than or equal to 2.

8. A computer program product comprising a computer program or instructions, characterized in that include: The computer program or instructions are executed by a processor to implement the graph retrieval-based kidney cancer risk assessment method according to any one of claims 1 to 7.

9. A computer device comprising a memory, a processor and a computer program or instruction stored in the memory, characterized in that: include: The computer program or instructions are executed by a processor to implement the graph retrieval-based kidney cancer risk assessment method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: include: The computer program or instructions are executed by a processor to implement the graph retrieval-based kidney cancer risk assessment method according to any one of claims 1 to 7.

Images