A method for predicting the risk of delayed medical treatment in patients with gastric cancer

Abstract

The present invention discloses a method for predicting the risk of medical delay for gastric cancer patients, belonging to the field of artificial intelligence technology, comprising the following steps: obtaining medical treatment data of different gastric cancer patients and constructing prediction target variables and prediction features respectively; selecting significant prediction features using a significant feature objective function; performing feature grouping and label assignment based on a decision tree to obtain group labels for significant prediction features; constructing a gastric cancer patient medical delay risk prediction model; constructing a prediction loss function; training the gastric cancer patient medical delay risk prediction model based on a significant prediction feature dataset according to the prediction loss function to obtain a trained medical delay risk prediction model; obtaining significant prediction features based on the medical treatment data of the gastric cancer patient to be predicted, and inputting the significant prediction features into the trained medical delay risk prediction model to obtain a medical delay risk result for the gastric cancer patient. The present invention solves the problem of the difficulty in accurately predicting the risk of medical delay for gastric cancer patients.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to a method for predicting the risk of delayed medical treatment for gastric cancer patients. Background Art

[0002] Gastric cancer has a high morbidity and mortality rate. The prognosis of gastric cancer is closely related to the timing of diagnosis and treatment. Early diagnosis and treatment are key factors in improving the prognosis of gastric cancer patients. Delayed medical treatment not only leads to a poor prognosis for gastric cancer patients, affecting their quality of life and survival, but also increases medical costs.

[0003] At present, research on delayed medical treatment for gastric cancer patients mainly focuses on the current situation and influencing factors, and there is a lack of prediction of delayed medical treatment for gastric cancer patients. At the same time, due to the time and cost of collecting clinical data, the data collected on delayed medical treatment for gastric cancer patients are generally small. The factors that lead to delayed medical treatment for gastric cancer are complex. Existing prediction models are mostly based on logistic regression models, which are prone to overfitting problems on small data sets, resulting in insufficient model generalization, manifested as high accuracy on training data and low accuracy on test data. Although this problem can be alleviated to a certain extent by reducing model complexity and weight decay, these methods lack direct constraints on the feature space, making it difficult to achieve a balance between modeling ability and generalization ability, and it is difficult to meet the requirements for accurate prediction of the risk of delayed medical treatment for gastric cancer. Summary of the Invention

[0004] In response to the above-mentioned deficiencies in the prior art, the present invention provides a method for predicting the risk of delayed medical treatment for gastric cancer patients, which solves the problem of difficulty in accurately predicting the risk of delayed medical treatment for gastric cancer patients.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] The present invention provides a method for predicting the risk of delayed medical treatment for gastric cancer patients, comprising the following steps:

[0007] S1. Obtain medical data of different gastric cancer patients and construct prediction target variables and prediction features respectively;

[0008] S2. According to the Lasso regression algorithm, significant prediction features are selected from the prediction features using the significant feature objective function, and a significant prediction feature dataset is constructed;

[0009] S3. performing feature grouping and label assignment for each significant prediction feature in the significant prediction feature dataset based on the decision tree to obtain a group label for the significant prediction feature;

[0010] S4. Construct a risk prediction model for delayed medical treatment in gastric cancer patients;

[0011] S5. Construct a prediction loss function based on the group labels of significant prediction features;

[0012] S6. Training a risk prediction model for delayed medical treatment for gastric cancer patients based on the significant prediction feature dataset according to the prediction loss function to obtain a trained risk prediction model for delayed medical treatment;

[0013] S7. Based on the medical treatment data of the gastric cancer patient to be predicted, obtain its significant prediction features, and input the significant prediction features into the trained medical treatment delay risk prediction model to obtain the medical treatment delay risk result of the gastric cancer patient.

[0014] The beneficial effects of the present invention are as follows: the present invention provides a method for predicting the risk of delayed medical treatment for gastric cancer patients, which extracts prediction features from a variety of data that mainly affect the delay in medical treatment for gastric cancer patients based on the medical treatment data of different gastric cancer patients obtained, and uses the results that lead to the risk of delayed medical treatment as the prediction target variable, thereby fully considering the objective factors that affect the timely medical treatment of gastric cancer patients; the present invention obtains significant prediction features through the Lasso algorithm and the significant feature objective function screening, which provides a basis for reducing the overfitting effect on small data sets, thereby further ensuring the accuracy of predicting the risk of delayed medical treatment for gastric cancer patients; the present invention assigns grouping labels to each significant feature based on the decision tree, which helps to evaluate the feature space similarity of significant prediction features; the prediction loss function constructed by the present invention not only realizes direct constraints on the feature space, but also realizes a balance between modeling ability and generalization ability, effectively improving the training effect of the gastric cancer patient medical delay risk prediction model, so that the trained medical delay risk prediction model can accurately predict the medical delay risk results of gastric cancer patients.

[0015] Furthermore, the medical treatment data of gastric cancer patients in S1 include the time interval data from the first discovery of cancer symptoms to the first medical treatment, gender data, ethnicity data, age data, education level data, place of residence data, medical insurance type data, marital status data, per capita annual income data, spicy food consumption data, long-term drinking data, relatives within three generations suffering from rectal cancer, previous cancer data, previous stomach disease data, regular physical examination data, family support score data, health literacy score data, anxiety score data, depression score data, medical coping style score data of facing, medical coping style score data of avoiding and medical coping style score data of giving in.

[0016] The beneficial effect of adopting the above-mentioned further scheme is: the present invention provides medical treatment data items for gastric cancer patients, fully considering the direct phenomenon characteristics and influencing factors that lead to the risk of delayed medical treatment for gastric cancer patients, and providing a data basis for further screening significant predictive features and accurately predicting the risk of delayed medical treatment for gastric cancer patients.

[0017] Furthermore, the S1 includes the following steps:

[0018] S11. Let the obtained medical data of gastric cancer patients be N sets, where each patient has a unique set of medical data, and N is a positive integer;

[0019] S12. Extract the time interval from the patient's first discovery of cancer symptoms to the first visit to the doctor from each medical record as the diagnosis interval;

[0020] S13, setting a prediction target variable based on the time interval between consultations;

[0021] The calculation expression of the predicted target variable is as follows:

[0022] ,

[0023] in, represents the i-th predicted target variable, represents the time interval between the first discovery of cancer-related symptoms and the first visit to a medical institution for cancer by the i-th patient, where i is a positive integer;

[0024] S14. Combining the other data in the medical treatment data of the same patient except the time interval data from the first discovery of cancer symptoms to the first visit to the doctor as the corresponding prediction features of the patient.

[0025] The beneficial effect of adopting the above-mentioned further scheme is as follows: the present invention provides a method for constructing predictive target variables and predictive features based on the medical data of different gastric cancer patients, and realizes that the factors that may cause delayed medical treatment for gastric cancer patients are associated with the direct phenomenon characteristics of delayed medical treatment based on the medical data, thereby providing a basis for accurately predicting the risk of delayed medical treatment for gastric cancer patients.

[0026] Furthermore, the S2 includes the following steps:

[0027] S21. Set feature weights based on the Lasso regression algorithm;

[0028] The calculation expression of the feature weight is as follows:

[0029] ,

[0030] in, represents the feature weight, Represents the 1-dimensional feature weight in the prediction feature, Represents the 2D feature weight in the prediction feature, Represents the D-dimensional feature weight in the prediction feature, T represents transposition, ∈ represents belonging, R D represents the prediction feature dataset, where D represents the number of feature dimensions, D=21;

[0031] S22. Construct a significant feature objective function based on feature weights;

[0032] The calculation expression of the significant feature objective function is as follows:

[0033] ,

[0034] , ,

[0035] in, represents the weight of significant features, represents the feature weight minimization operation, represents the salient feature objective function, represents the transpose of feature weights, represents the i-th prediction feature, Indicates adjustable parameters, represents the feature weight of the d-dimensional feature in the prediction feature, || represents the L1 norm, where d is a positive integer;

[0036] S23. According to the significant feature objective function, features corresponding to dimensions with non-zero significant feature weights are selected as significant prediction features, and a significant prediction feature dataset is constructed.

[0037] The beneficial effects of adopting the above-mentioned further scheme are as follows: the present invention constructs a significant feature objective function based on the Lasso regression algorithm, and adds an L1 norm regularization term on the basis of the least squares method. It can extract characteristic factors that are significantly related to the delay in medical treatment for gastric cancer patients through feature weights, and construct a corresponding significant prediction feature data set, providing a basis for training a network model that can accurately predict the risk of delayed medical treatment for gastric cancer patients.

[0038] Furthermore, the S3 includes the following steps:

[0039] S31. Construct a feature classification binary tree based on the decision tree algorithm;

[0040] S32. Perform feature grouping and label assignment for each significant prediction feature in the significant prediction feature dataset using a feature classification binary tree to obtain group labels for the significant prediction features.

[0041] The beneficial effect of adopting the above further scheme is: the present invention provides a method for feature grouping and label classification for significant prediction features based on a decision tree. By assigning the same grouping labels to similar significant prediction features, it helps to directly constrain the similarity of spatial features, thereby effectively improving the robustness of model prediction even when the amount of medical data of gastric cancer patients is small, and accurately predicting the risk of delayed medical treatment for gastric cancer patients.

[0042] Furthermore, the risk prediction model for delayed medical treatment for gastric cancer patients in S4 includes a first feature extraction module, a second feature extraction module, a third feature extraction module and a feature output module connected in sequence;

[0043] The first feature extraction module, the second feature extraction module and the third feature extraction module each include a linear mapping layer and a Relu activation layer; the feature output module includes a linear mapping layer and a Sigmoid activation layer.

[0044] The beneficial effect of adopting the above-mentioned further scheme is as follows: the present invention provides a risk prediction model for delayed medical treatment for gastric cancer patients. By constructing a neural network model with a simple structure, it can effectively reduce weight attenuation. On the basis of significant prediction features and prediction loss functions, it effectively ensures accurate prediction of the risk of delayed medical treatment for gastric cancer patients.

[0045] Furthermore, the calculation expression of the prediction loss function in S5 is as follows:

[0046] ,

[0047] ,

[0048] ,

[0049] ,

[0050] ,

[0051] Among them, L represents the prediction loss function, L CE represents the cross entropy loss function, Represents the feature space similarity loss weight coefficient, L FS represents the feature space similarity loss function, represents the risk prediction result of medical delay corresponding to the i-th significant prediction feature, Parameter indicating that the i-th significant predictive feature and the j-th significant predictive feature have the same grouping label, The exponential parameter representing the similarity between the i-th significant prediction feature and the j-th significant prediction feature, Indicates the variable adjustment amount, Parameter indicating that the i-th significant prediction feature and the k-th significant prediction feature have the same grouping label, The exponential parameter representing the similarity between the i-th significant prediction feature and the k-th significant prediction feature, if represents if, g i represents the group label of the i-th significant prediction feature, g jRepresents the grouping label of the jth significant prediction feature, otherwise represents other cases, represents the similarity between the i-th significant prediction feature and the j-th significant prediction feature, f i represents the intermediate feature corresponding to the i-th significant prediction feature, f j represents the intermediate feature corresponding to the j-th significant prediction feature, where j and k are both positive integers.

[0052] The beneficial effect of adopting the above-mentioned further scheme is as follows: the present invention provides a calculation method for predicting the instantaneous function, and the loss function constructed based on the degree of proximity between the risk prediction result and the predicted target value, and the similarity of the feature space. It can make the feature space more regular while ensuring the prediction accuracy, and make the model more robust to changes in the input features, thereby improving the generalization ability of the model.

[0053] Furthermore, the S6 includes the following steps:

[0054] S61. Inputting the significant prediction features in the significant prediction feature dataset into a risk prediction model for delayed medical treatment for gastric cancer patients;

[0055] S62, extracting the input significant prediction feature using the linear mapping layer in the first feature extraction module, and activating the input significant prediction feature using the Relu activation layer in the first feature extraction module to obtain a first feature extraction result corresponding to the significant prediction feature;

[0056] S63, using the linear mapping layer in the second feature extraction module to perform feature extraction on the input first feature extraction result, and using the Relu activation layer in the second feature extraction module to activate it, to obtain a second feature extraction result corresponding to the significant prediction feature;

[0057] S64, using the linear mapping layer in the third feature extraction module to perform feature extraction on the input second feature extraction result, and using the Relu activation layer in the third feature extraction module to activate it, to obtain a third feature extraction result corresponding to the significant prediction feature as the intermediate feature;

[0058] S65. Extracting features from the intermediate features using the linear mapping layer in the feature output module and activating them using the Sigmoid activation layer in the feature output module to obtain risk prediction results corresponding to the significant prediction features, which are used as the risk results for delayed medical treatment for gastric cancer patients.

[0059] S66. Based on minimizing the prediction loss function, repeatedly execute S61-S65 for the preset number of training times, and optimize the parameters of the medical delay risk prediction model for gastric cancer patients through an adaptive momentum estimation algorithm to obtain a trained medical delay risk prediction model.

[0060] The beneficial effect of adopting the above-mentioned further scheme is as follows: the present invention provides a method for training a risk prediction model for delayed medical treatment for gastric cancer patients. Based on the prediction loss function provided by the present invention, the feature space can be effectively constrained during the model training process, thereby ensuring the balance between modeling ability and generalization ability, and providing a basis for accurately predicting the risk of delayed medical treatment for gastric cancer patients.

[0061] Other advantages of the present invention will be analyzed in more detail in subsequent embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1 This is a flowchart of the steps of a method for predicting the risk of delayed medical treatment for gastric cancer patients in an embodiment of the present invention.

[0064] Figure 2 Schematic diagram of a risk prediction model for delayed medical treatment for gastric cancer patients in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0066] like Figure 1 As shown, in one embodiment of the present invention, the present invention provides a method for predicting the risk of delayed medical treatment for gastric cancer patients, comprising the following steps:

[0067] S1. Obtain medical data of different gastric cancer patients and construct prediction target variables and prediction features respectively;

[0068] The medical data of gastric cancer patients in S1 include the time interval data from the first discovery of cancer symptoms to the first visit to the doctor, gender data, ethnicity data, age data, education level data, place of residence data, medical insurance type data, marital status data, per capita annual income data, spicy food consumption data, long-term drinking data, relatives within three generations suffering from rectal cancer, previous cancer data, previous stomach disease data, regular physical examination data, family support score data, health literacy score data, anxiety score data, depression score data, medical coping style score data of facing, medical coping style score data of avoiding and medical coping style score data of giving in.

[0069] In this embodiment, the medical treatment data indicators and value selection rules of gastric cancer patients are shown in Table 1:

[0070] Table 1:

[0071]

[0072]

[0073] Based on the medical data indicators and related value-selection rules of gastric cancer patients, different influencing factors can be effectively distinguished and divided, thus providing a basis for improving the classification accuracy of significant predictive features.

[0074] The S1 comprises the following steps:

[0075] S11. Let the obtained medical data of gastric cancer patients be N sets, where each patient has a unique set of medical data, and N is a positive integer;

[0076] S12. Extract the time interval from the patient's first discovery of cancer symptoms to the first visit to the doctor from each medical record as the diagnosis interval;

[0077] S13, setting a prediction target variable based on the time interval between consultations;

[0078] The calculation expression of the predicted target variable is as follows:

[0079] ,

[0080] in, represents the i-th predicted target variable, represents the time interval between the first discovery of cancer-related symptoms and the first visit to a medical institution for cancer by the i-th patient, where i is a positive integer;

[0081] S14. Combining the other data in the medical treatment data of the same patient except the time interval data from the first discovery of cancer symptoms to the first visit to the doctor as the corresponding prediction features of the patient.

[0082] S2. According to the Lasso regression algorithm, significant prediction features are selected from the prediction features using the significant feature objective function, and a significant prediction feature dataset is constructed;

[0083] The S2 comprises the following steps:

[0084] S21. Set feature weights based on the Lasso regression algorithm;

[0085] The calculation expression of the feature weight is as follows:

[0086] ,

[0087] in, represents the feature weight, Represents the 1-dimensional feature weight in the prediction feature, Represents the 2D feature weight in the prediction feature, Represents the D-dimensional feature weight in the prediction feature, T represents transposition, ∈ represents belonging, R D represents the prediction feature dataset, where D represents the number of feature dimensions, D=21;

[0088] S22. Construct a significant feature objective function based on feature weights;

[0089] The calculation expression of the significant feature objective function is as follows:

[0090] ,

[0091] , ,

[0092] in, represents the weight of significant features, represents the feature weight minimization operation, represents the salient feature objective function, represents the transpose of feature weights, represents the i-th prediction feature, Indicates adjustable parameters, represents the feature weight of the d-dimensional feature in the prediction feature, || represents the L1 norm, where d is a positive integer;

[0093] S23. According to the significant feature objective function, features corresponding to dimensions with non-zero significant feature weights are selected as significant prediction features, and a significant prediction feature dataset is constructed.

[0094] In this embodiment, the calculation expression of the significant prediction feature in S23 is as follows:

[0095] ,

[0096] Among them, z i represents the i-th significant predictive feature, R M Represents a significant prediction feature dataset, and M represents the number of feature dimensions with non-zero significant feature weights.

[0097] S3. performing feature grouping and label assignment for each significant prediction feature in the significant prediction feature dataset based on the decision tree to obtain a group label for the significant prediction feature;

[0098] The S3 comprises the following steps:

[0099] S31. Construct a feature classification binary tree based on the decision tree algorithm;

[0100] In this embodiment, the feature classification binary tree includes one root node, several intermediate nodes and leaf nodes; when the significant prediction feature is input into the feature classification binary tree, the significant prediction feature starts from the root node, performs classification judgment when passing through the intermediate nodes, and finally reaches a leaf node. The significant prediction features that reach the same leaf node have high similarity. The solution of the present invention assigns the same grouping label to the significant prediction features with high similarity, thereby forming several data groups.

[0101] S32. Perform feature grouping and label assignment for each significant prediction feature in the significant prediction feature dataset using a feature classification binary tree to obtain group labels for the significant prediction features.

[0102] In this embodiment, the calculation expression of the grouping label of the significant prediction feature is as follows:

[0103] ,

[0104] Among them, g i represents the grouping label of the i-th significant prediction feature, l represents the l-th leaf node, where l=1, 2, ..., L ’ , L ’ is the total number of leaf nodes.

[0105] S4. Construct a risk prediction model for delayed medical treatment in gastric cancer patients;

[0106] like Figure 2 As shown, the risk prediction model for delayed medical treatment for gastric cancer patients in S4 includes a first feature extraction module, a second feature extraction module, a third feature extraction module and a feature output module connected in sequence;

[0107] The first feature extraction module, the second feature extraction module and the third feature extraction module each include a linear mapping layer and a Relu activation layer; the feature output module includes a linear mapping layer and a Sigmoid activation layer.

[0108] In this embodiment, the input dimension of the linear mapping layer of the first feature extraction module is M, and the output dimension is 24; the input dimension of the linear mapping layer of the second feature extraction module is 24, and the output dimension is 16; the input dimension of the linear mapping layer of the third feature extraction module is 16, and the output dimension is E, where E=8, which is the feature dimension extracted after passing through three layers of feature extraction modules; the input dimension of the linear mapping layer of the feature output module is E, and the output dimension is 1.

[0109] S5. Construct a prediction loss function based on the group labels of significant prediction features;

[0110] The calculation expression of the prediction loss function in S5 is as follows:

[0111] ,

[0112] ,

[0113] ,

[0114] ,

[0115] ,

[0116] Among them, L represents the prediction loss function, L CE represents the cross entropy loss function, Represents the feature space similarity loss weight coefficient, L FS represents the feature space similarity loss function, represents the risk prediction result of medical delay corresponding to the i-th significant prediction feature, Parameter indicating that the i-th significant predictive feature and the j-th significant predictive feature have the same grouping label, The exponential parameter representing the similarity between the i-th significant prediction feature and the j-th significant prediction feature, Indicates the variable adjustment amount, Parameter indicating that the i-th significant prediction feature and the k-th significant prediction feature have the same grouping label, The exponential parameter representing the similarity between the i-th significant prediction feature and the k-th significant prediction feature, if represents if, g i represents the group label of the i-th significant prediction feature, g j Represents the grouping label of the jth significant prediction feature, otherwise represents other cases, represents the similarity between the i-th significant prediction feature and the j-th significant prediction feature, f i represents the intermediate feature corresponding to the i-th significant prediction feature, fj represents the intermediate feature corresponding to the j-th significant prediction feature, where j and k are both positive integers.

[0117] In this embodiment, the cross entropy loss function is used to supervise the degree of closeness between the risk prediction result output by the model and the predicted target value, and the feature space similarity loss function is used to constrain the feature space of the neural network to improve the generalization ability of the model; in the feature space similarity loss function, the more similar the features with the same grouping label are, the smaller the numerator is, and the smaller the corresponding loss value is; conversely, the more dissimilar the features with different grouping labels are, the larger the numerator is, and the larger the corresponding loss value is. Therefore, by minimizing the value of the feature space similarity loss function, features with the same grouping label can be made more similar, thereby making the feature space more regular and making the model more robust to changes in input features, thereby improving the generalization ability of the model.

[0118] S6. Training a risk prediction model for delayed medical treatment for gastric cancer patients based on the significant prediction feature dataset according to the prediction loss function to obtain a trained risk prediction model for delayed medical treatment;

[0119] The S6 comprises the following steps:

[0120] S61. Inputting the significant prediction features in the significant prediction feature dataset into a risk prediction model for delayed medical treatment for gastric cancer patients;

[0121] S62, extracting the input significant prediction feature using the linear mapping layer in the first feature extraction module, and activating the input significant prediction feature using the Relu activation layer in the first feature extraction module to obtain a first feature extraction result corresponding to the significant prediction feature;

[0122] S63, using the linear mapping layer in the second feature extraction module to perform feature extraction on the input first feature extraction result, and using the Relu activation layer in the second feature extraction module to activate it, to obtain a second feature extraction result corresponding to the significant prediction feature;

[0123] S64, using the linear mapping layer in the third feature extraction module to perform feature extraction on the input second feature extraction result, and using the Relu activation layer in the third feature extraction module to activate it, to obtain a third feature extraction result corresponding to the significant prediction feature as the intermediate feature;

[0124] S65. Extracting features from the intermediate features using the linear mapping layer in the feature output module and activating them using the Sigmoid activation layer in the feature output module to obtain risk prediction results corresponding to the significant prediction features, which are used as the risk results for delayed medical treatment for gastric cancer patients.

[0125] S66. Based on minimizing the prediction loss function, repeatedly execute S61-S65 for the preset number of training times, and optimize the parameters of the medical delay risk prediction model for gastric cancer patients through an adaptive momentum estimation algorithm to obtain a trained medical delay risk prediction model.

[0126] S7. Based on the medical treatment data of the gastric cancer patient to be predicted, obtain its significant prediction features, and input the significant prediction features into the trained medical treatment delay risk prediction model to obtain the medical treatment delay risk result of the gastric cancer patient.

[0127] The risk of delayed medical treatment for gastric cancer patients is reflected in whether the time from the first discovery of cancer symptoms to the first visit to the doctor exceeds 3 months. If it exceeds 3 months, there is a risk of delayed medical treatment. Otherwise, it is a timely medical treatment situation, which can effectively avoid the risk of delayed medical treatment.

[0128] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for predicting the risk of delayed medical treatment in gastric cancer patients, characterized in that: The steps include: S1. Obtain medical data of different gastric cancer patients and construct prediction target variables and prediction features respectively; S2. According to the Lasso regression algorithm, significant prediction features are selected from the prediction features using the significant feature objective function, and a significant prediction feature dataset is constructed; S3. performing feature grouping and label assignment for each significant prediction feature in the significant prediction feature dataset based on the decision tree to obtain a group label for the significant prediction feature; The S3 comprises the following steps: S31. Construct a feature classification binary tree based on the decision tree algorithm; S32, using a feature classification binary tree to perform feature grouping and label assignment for each significant prediction feature in the significant prediction feature dataset, to obtain a group label for the significant prediction feature; S4. Construct a risk prediction model for delayed medical treatment in gastric cancer patients; The risk prediction model for delayed medical treatment for gastric cancer patients in S4 comprises a first feature extraction module, a second feature extraction module, a third feature extraction module and a feature output module connected in sequence; The first feature extraction module, the second feature extraction module and the third feature extraction module each include a linear mapping layer and a Relu activation layer; the feature output module includes a linear mapping layer and a Sigmoid activation layer; S5. Construct a prediction loss function based on the group labels of significant prediction features; The calculation expression of the prediction loss function in S5 is as follows: ， ， ， ， ， in, L represents the prediction loss function, L CE represents the cross entropy loss function, represents the feature space similarity loss weight coefficient, L FS represents the feature space similarity loss function, Indicates the i The risk prediction results of medical delay corresponding to the significant predictive features are Indicates the i The significant predictive features j The parameters of the significant predictive features have the same group label, Indicates the i The significant predictive features j An index parameter that significantly predicts the similarity between features, Indicates the variable adjustment amount, Indicates the i The significant predictive features k The parameters of the significant predictive features have the same group label, Indicates the i The significant predictive features k An index parameter that significantly predicts the similarity between features, if Indicates if, g i Indicates the i The group labels of the significant predictive features, g j Indicates the j The group labels of the significant predictive features, otherwise Indicates other situations, Indicates the i The significant predictive features j The similarity between the significant predictive features, f i Indicates the i The intermediate features corresponding to the significant prediction features, f j Indicates the j The intermediate features corresponding to the significant prediction features, where j , k are all positive integers; S6. Training a risk prediction model for delayed medical treatment for gastric cancer patients based on the significant prediction feature dataset according to the prediction loss function to obtain a trained risk prediction model for delayed medical treatment; S7. Based on the medical treatment data of the gastric cancer patient to be predicted, obtain its significant prediction features, and input the significant prediction features into the trained medical treatment delay risk prediction model to obtain the medical treatment delay risk result of the gastric cancer patient.

2. The method for predicting the risk of delayed medical treatment for gastric cancer patients according to claim 1, characterized in that: The medical data of gastric cancer patients in S1 include the time interval data from the first discovery of cancer symptoms to the first visit to the doctor, gender data, ethnicity data, age data, education level data, place of residence data, medical insurance type data, marital status data, per capita annual income data, spicy food consumption data, long-term drinking data, relatives within three generations suffering from rectal cancer, previous cancer data, previous stomach disease data, regular physical examination data, family support score data, health literacy score data, anxiety score data, depression score data, medical coping style score data of facing, medical coping style score data of avoiding and medical coping style score data of giving in.

3. The method for predicting the risk of delayed medical treatment for gastric cancer patients according to claim 2, characterized in that: The S1 comprises the following steps: S11. Let the medical data of gastric cancer patients be N Each patient has a unique medical data set. N is a positive integer; S12. Extract the time interval from the patient's first discovery of cancer symptoms to the first visit to the doctor from each medical record as the diagnosis interval; S13, setting a prediction target variable based on the time interval between consultations; The calculation expression of the predicted target variable is as follows: ， in, Indicates the i predictor target variables, Indicates the i The time interval between the first discovery of cancer-related symptoms and the first visit to a medical institution for cancer i is a positive integer; S14. Combining the other data in the medical treatment data of the same patient except the time interval data from the first discovery of cancer symptoms to the first visit to the doctor as the corresponding prediction features of the patient.

4. The method for predicting the risk of delayed medical treatment for gastric cancer patients according to claim 2, characterized in that: The S2 comprises the following steps: S21. Set feature weights based on the Lasso regression algorithm; The calculation expression of the feature weight is as follows: ， in, represents the feature weight, Represents the 1-dimensional feature weight in the prediction feature, Represents the 2D feature weight in the prediction feature, Represents the predicted features D dimensional feature weights, T represents transposition, ∈ represents belonging to, R D Represents the prediction feature dataset, where D Indicates the number of feature dimensions, D =21; S22. Construct a significant feature objective function based on feature weights; The calculation expression of the significant feature objective function is as follows: ， ， ， in, represents the weight of significant features, represents the feature weight minimization operation, represents the salient feature objective function, represents the transpose of feature weights, Indicates the i Predictive features, Indicates adjustable parameters, Indicates the prediction feature d The feature weight of the dimension feature, || represents the L1 norm, where, d is a positive integer; S23. According to the significant feature objective function, features corresponding to dimensions with non-zero significant feature weights are selected as significant prediction features, and a significant prediction feature dataset is constructed.

5. The method for predicting the risk of delayed medical treatment for gastric cancer patients according to claim 1, characterized in that: The S6 comprises the following steps: S61. Inputting the significant prediction features in the significant prediction feature dataset into a risk prediction model for delayed medical treatment for gastric cancer patients; S62, extracting the input significant prediction feature using the linear mapping layer in the first feature extraction module, and activating the input significant prediction feature using the Relu activation layer in the first feature extraction module to obtain a first feature extraction result corresponding to the significant prediction feature; S63, using the linear mapping layer in the second feature extraction module to perform feature extraction on the input first feature extraction result, and using the Relu activation layer in the second feature extraction module to activate it, to obtain a second feature extraction result corresponding to the significant prediction feature; S64, using the linear mapping layer in the third feature extraction module to perform feature extraction on the input second feature extraction result, and using the Relu activation layer in the third feature extraction module to activate it, to obtain a third feature extraction result corresponding to the significant prediction feature as the intermediate feature; S65. Extracting features from the intermediate features using the linear mapping layer in the feature output module and activating them using the Sigmoid activation layer in the feature output module to obtain risk prediction results corresponding to the significant prediction features, which are used as the risk results for delayed medical treatment for gastric cancer patients. S66. Based on minimizing the prediction loss function, repeatedly execute S61-S65 for the preset number of training times, and optimize the parameters of the medical delay risk prediction model for gastric cancer patients through an adaptive momentum estimation algorithm to obtain a trained medical delay risk prediction model.

Images