Risk prediction method for malnutrition of gastric cancer surgery patient

Abstract

The invention discloses a risk prediction method for dystrophy of a patient in a gastric cancer operation, and belongs to the technical field of nutrition support and management of cancer patients, and the method comprises the following steps: obtaining an initial dystrophy risk monitoring data set; performing data cleaning and prediction label and data mask setting on the initial malnutrition risk monitoring data set to obtain a malnutrition risk dynamic time sequence training set; constructing a malnutrition risk prediction model; according to the dystrophy risk prediction loss function, training the dystrophy risk prediction model by using preoperative data and postoperative periodic report data of each gastric cancer patient in the dystrophy risk dynamic time sequence training set to obtain a trained dystrophy risk prediction model; and acquiring new malnutrition risk monitoring data of the gastric cancer patient, and performing risk prediction and early warning on the gastric cancer patient by using the trained malnutrition risk prediction model. The method solves the problem that the accuracy rate of dystrophy risk prediction of the patient in the gastric cancer operation is insufficient.

Description

Technical Field

[0001] The present invention belongs to the technical field of nutrition support and management for cancer patients, and in particular relates to a method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery. Background Art

[0002] The morbidity and mortality rates of gastric cancer are among the highest among all malignant tumors. Subtotal gastrectomy or total gastrectomy is the main treatment for gastric cancer. Since gastric cancer directly affects the function of the digestive system, and surgery can damage the physiological function of the intestines and affect nutrient absorption, patients undergoing gastric cancer surgery are prone to various gastrointestinal symptoms, such as nausea, bloating, abdominal pain, etc., which lead to loss of appetite or a feeling of fullness after eating a small amount of food, further affecting the patient's nutritional intake. The above factors lead to a high incidence of malnutrition and poor prognosis in patients undergoing gastric cancer surgery, and seriously affect the quality of life of patients. Therefore, effectively predicting the risk of malnutrition in patients with gastric cancer and taking timely intervention measures are of great significance to improving the treatment effect of patients undergoing gastric cancer surgery and improving the quality of life of patients.

[0003] The malnutrition risk prediction method is a method for predicting the possibility of malnutrition in patients based on artificial intelligence and machine learning technology. The prior art proposes a method for predicting malnutrition in patients with nasopharyngeal carcinoma, a method for predicting the risk of malnutrition in patients with gastric cancer chemotherapy and elderly patients with early gastric cancer endoscopic treatment, and a method for predicting malnutrition in patients with gastric cancer 3 months after surgery. In general, the existing methods for predicting the risk of malnutrition in gastric cancer patients generally predict the risk of subsequent malnutrition based on the patient's data at a certain time point. At the same time, the existing prediction models are mostly based on the Logistic regression model, which has limited modeling capabilities and low prediction accuracy. 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 malnutrition in patients undergoing gastric cancer surgery. Based on the preoperative data and periodic report data of gastric cancer patients, a dynamic malnutrition risk prediction model is constructed using a recurrent neural network, thereby solving the problem of insufficient accuracy in predicting the risk of malnutrition in patients undergoing gastric cancer surgery.

[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 malnutrition in patients undergoing gastric cancer surgery, comprising the following steps:

[0007] S1. Obtain malnutrition risk monitoring data of several gastric cancer patients to form an initial malnutrition risk monitoring data set;

[0008] S2. Clean the initial malnutrition risk monitoring data set and set prediction labels and data masks to obtain a malnutrition risk dynamic time series training set;

[0009] S3. Construct a malnutrition risk prediction model;

[0010] S4. According to the malnutrition risk prediction loss function, the malnutrition risk prediction model is trained using the preoperative data and postoperative periodic report data of each gastric cancer patient in the malnutrition risk dynamic time series training set to obtain a trained malnutrition risk prediction model;

[0011] S5. Obtain the malnutrition risk monitoring data of new gastric cancer patients, and use the trained malnutrition risk prediction model to perform risk prediction and early warning for the gastric cancer patients.

[0012] Furthermore, the S1 comprises the following steps:

[0013] S11. Obtaining preoperative data of gastric cancer patients, wherein the preoperative data of gastric cancer patients includes demographic data and medical related data;

[0014] S12. Obtain postoperative periodic reporting data of gastric cancer patients;

[0015] S13, obtaining periodic nutritional status monitoring data of gastric cancer patients, wherein the periodic nutritional status monitoring data includes preoperative nutritional status monitoring results and several postoperative nutritional status monitoring results of gastric cancer patients;

[0016] S14. Combine the preoperative data, postoperative periodic reporting data and periodic nutritional status monitoring data of gastric cancer patients as the malnutrition risk monitoring data of the corresponding gastric cancer patients.

[0017] Furthermore, S2 comprises the following steps:

[0018] S21. If the preoperative nutritional status monitoring result of a gastric cancer patient is malnutrition in the initial malnutrition risk monitoring data set, the malnutrition risk monitoring data of the gastric cancer patient is deleted to obtain a first screening malnutrition risk monitoring data set;

[0019] S22. If the interval between adjacent postoperative periodic reporting dates of the same gastric cancer patient in the initial malnutrition risk monitoring data set exceeds a preset reporting interval threshold, or the adjacent postoperative nutritional status monitoring dates of the same gastric cancer patient exceed a preset monitoring interval threshold, the malnutrition risk monitoring data of the gastric cancer patient is deleted to obtain a second screening malnutrition risk monitoring data set;

[0020] S23, respectively setting prediction labels and data masks for the postoperative periodic report data of each gastric cancer patient in the second screening malnutrition risk monitoring data set to obtain a malnutrition risk dynamic time series training set;

[0021] The calculation expressions of the preset labels and data masks are as follows:

[0022] ,

[0023] ,

[0024] in, represents the predicted label of the i-th postoperative periodic report data of the k-th gastric cancer patient, e represents the exponential base constant, represents the i-th postoperative periodic report date of gastric cancer patients, represents the Jth postoperative nutritional status monitoring date corresponding to the kth gastric cancer patient suffering from malnutrition, The data mask representing the i-th postoperative periodic report data of the k-th gastric cancer patient.

[0025] Furthermore, the malnutrition risk prediction model in S3 includes:

[0026] The first multi-layer perceptron is used to perform feature mapping on the preoperative data of each gastric cancer patient in the malnutrition risk dynamic time series training set to obtain a first mapping feature;

[0027] The second multi-layer perceptron is used to perform feature mapping on the postoperative periodic report data of each gastric cancer patient in the malnutrition risk dynamic time series training set to obtain a second mapping feature;

[0028] A Concat module, used to perform Concat concatenation on the first mapping feature and the second mapping feature to obtain a combined feature;

[0029] The gated recurrent neural module is used to calculate the current hidden state using the combined features and the previous hidden state;

[0030] The third multi-layer perceptron is used to perform feature mapping on the current hidden state to obtain a hidden state mapping result;

[0031] The activation module is used to activate the hidden state mapping results to obtain the i-th postoperative malnutrition risk prediction result.

[0032] Furthermore, the malnutrition risk prediction model in S3 is used to iteratively predict and update the preoperative data and postoperative periodic report data of each gastric cancer patient in the malnutrition risk dynamic time series training set, and obtain each postoperative malnutrition risk prediction result of the gastric cancer patient and the hidden state of the postoperative malnutrition risk prediction;

[0033] The calculation expression of the malnutrition risk prediction model is as follows:

[0034] ,

[0035] , ,

[0036] in, represents the risk prediction result of malnutrition after surgery for the kth gastric cancer patient. represents the hidden state of the risk prediction of malnutrition after surgery for the kth gastric cancer patient, represents the malnutrition risk prediction model, represents the preoperative data of the kth gastric cancer patient, represents the periodic report data of the i-th postoperative period for the k-th gastric cancer patient, represents the hidden state of the risk prediction of malnutrition after surgery for the kth gastric cancer patient i-1 times, represents the sigmoid function, represents the multi-layer perceptron feature map, represents the initial hidden state.

[0037] Furthermore, the calculation expression of the malnutrition risk prediction loss function in S4 is as follows:

[0038] ,

[0039] ,

[0040] ,

[0041] Among them, L represents the loss function of malnutrition risk prediction results, represents the loss function of malnutrition risk prediction results, represents the weight adjustment parameter, represents the constraint mapping loss function, represents the identity matrix, represents the risk prediction result of malnutrition after the i-th surgery for gastric cancer patients, represents the first mapping feature matrix, represents the transpose of the first mapping feature matrix, represents the second mapping feature matrix, represents the transpose of the second mapping feature matrix, Represents the Frobenius norm of the matrix.

[0042] Furthermore, the S5 comprises the following steps:

[0043] S51. Obtain preoperative data of new gastric cancer patients after the surgical method is determined, as well as several postoperative periodic reporting data of the gastric cancer patients;

[0044] S52, inputting the preoperative data and each postoperative periodic report data of the new gastric cancer patient into the trained malnutrition risk prediction model, and predicting the malnutrition risk prediction result of the new gastric cancer patient after each postoperative periodic report;

[0045] S53. Compare the malnutrition risk prediction result of new gastric cancer patients after each postoperative periodic report with the malnutrition risk threshold. If the malnutrition risk prediction result of existing gastric cancer patients after each postoperative periodic report is greater than the malnutrition risk threshold, a malnutrition warning is generated for the new gastric cancer patient's postoperative periodic report.

[0046] The beneficial effects of the present invention are as follows: a method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery provided by the present invention obtains malnutrition risk monitoring data of several gastric cancer patients, constructs a dynamic time series training set of malnutrition risk in gastric cancer patients, and provides a data basis for training a neural network model that can accurately predict the malnutrition risk in gastric cancer patients; the present invention constructs a malnutrition risk prediction model based on a recurrent neural network, realizes feature embedding and extraction of low-order linear independent features for the input preoperative data and postoperative periodic report data of each gastric cancer patient, effectively reduces the complexity of the gated recurrent neural module, and improves the generalization ability of the malnutrition risk prediction model; the present invention can accurately predict and warn the risks of gastric cancer patients, and provide a basis for medical staff to carry out nutritional support intervention in a timely manner.

[0047] Other advantages of the present invention will be analyzed in more detail in subsequent embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 The present invention is a flowchart of the steps of a method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery according to an embodiment of the present invention.

[0050] Figure 2 Schematic diagram of a malnutrition risk prediction model in an embodiment of the present invention.

[0051] Figure 3This is an overall schematic diagram of risk prediction performed by the malnutrition risk prediction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] 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, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here 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 the 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 belong to the scope of protection of the present invention.

[0053] like Figure 1 As shown, in one embodiment of the present invention, the present invention provides a method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery, comprising the following steps:

[0054] S1. Obtain malnutrition risk monitoring data of several gastric cancer patients to form an initial malnutrition risk monitoring data set;

[0055] The S1 comprises the following steps:

[0056] S11. Obtaining preoperative data of gastric cancer patients, wherein the preoperative data of gastric cancer patients includes demographic data and medical related data;

[0057] In this plan, demographic data include age, gender, ethnicity, education, place of residence, marital status, and income; medical related data include the patient's preoperative BMI data, hypoproteinemia data, diabetes data, tumor type, tumor stage, and surgical method. The preoperative data of gastric cancer patients are collected after the patient's surgical plan is determined.

[0058] S12. Obtain postoperative periodic reporting data of gastric cancer patients;

[0059] In this program, the periodic reporting data of gastric cancer patients after surgery include the patients' postoperative BMI data, quality of life score FACT-Ga, generalized anxiety disorder score GAD-7, depression score (PHQ-9), dietary intake change data and protein powder consumption data.

[0060] BMI is the square of the quotient of weight / height, and a BMI less than 18.5 indicates malnutrition.

[0061] Quality of life score FACT-Ga consists of 30 items, including 5 functional scales, 3 symptom scales, 6 single measurement items and 1 overall quality of life scale. Items 29 and 30 are positively scored, with a rating scale of 1-7, 1 representing very poor and 7 representing very good; the other items are reversely scored, with a rating scale of 1-4, 1-4 points, where 1 point means no, 2 points means a little, 3 points means quite, and 4 points means very. The higher the scores in the functional domain and overall quality of life, the higher the quality of life, and the higher the scores in the symptom domain and single items, the worse the quality of life.

[0062] The Generalized Anxiety Disorder Score GAD-7 is a psychological scale with good reliability, validity, sensitivity and specificity, which is suitable for screening, identification and assessment of anxiety states. This questionnaire contains 7 questions, each of which has four options: not at all, a few days, more than half of the time, and almost every day, corresponding to 0 points, 1 point, 2 points and 3 points respectively. Add the scores of the 7 questions to get the final score of the anxiety screening scale. The total score of this scale is 21 points. The higher the score, the greater the possibility of anxiety and the higher the degree of anxiety. Specifically, 0 to 4 points indicate no anxiety, 5 to 9 points indicate mild anxiety, 10 to 13 points indicate moderate anxiety, 14 to 18 points indicate moderate to severe anxiety, and 19 to 21 points indicate severe anxiety.

[0063] Depression score PHQ-9 has a total of 9 questions. The options for each question are from left to right: almost never, a few days, more than half the time, and almost every day. The corresponding scores are 0, 1, 2, and 3. The final score of PHQ-9 is the sum of the scores of the 9 questions, with a total score of 27. The higher the score, the more severe the depression. 0 to 4 points represent no depressive symptoms, 5 to 9 points represent mild depression, 10 to 14 points represent moderate depression, 15 to 19 points represent moderate to severe depression, and 20 to 27 points represent severe depression.

[0064] The postoperative periodic reporting data of gastric cancer patients is recorded as , that is, the i-th postoperative periodic report data of the k-th gastric cancer patient. The postoperative periodic report date of the gastric cancer patient is recorded as , which is the i-th postoperative periodic report date of the k-th gastric cancer patient, where, , , K represents the total number of gastric cancer patients, and B represents the total number of reports; in this embodiment, the postoperative periodic report data of gastric cancer patients are continuously obtained once a week for at least 6 consecutive months.

[0065] S13, obtaining periodic nutritional status monitoring data of gastric cancer patients, wherein the periodic nutritional status monitoring data includes preoperative nutritional status monitoring results and several postoperative nutritional status monitoring results of gastric cancer patients;

[0066] In this plan, the periodic nutritional status monitoring data of gastric cancer patients are collected once before surgery and at least once a month after surgery. The data collection time is set when the gastric cancer patients are re-examined or re-admitted to the hospital for chemotherapy. According to the GLIM malnutrition diagnostic criteria, if the weight of gastric cancer patients decreases by more than 5% within 6 months or the BMI is less than 18.5 kg / m 2 , then the gastric cancer patient is judged to be malnourished, and the j-th nutritional status monitoring result of the k-th gastric cancer patient is recorded as , the date of monitoring the nutritional status of the kth gastric cancer patient is recorded as , i.e. the jth nutritional status monitoring date of the kth gastric cancer patient, , among which, when hour, is the preoperative nutritional status monitoring result of the kth gastric cancer patient, represents the preoperative nutritional status monitoring date of the kth gastric cancer patient. , then the preoperative nutritional status monitoring result of the kth gastric cancer patient is malnutrition, if , then the preoperative nutritional status monitoring result of the kth gastric cancer patient is no malnutrition. hour, is the monitoring result of the nutritional status of the kth gastric cancer patient after the jth surgery. , then the result of the nutritional status monitoring of the kth gastric cancer patient after the jth surgery is malnutrition. , then the result of the nutritional status monitoring of the kth gastric cancer patient after the jth surgery is that there is no malnutrition. represents the postoperative nutritional status monitoring date of gastric cancer patients, and C is the total number of nutritional status monitoring times of gastric cancer patients; in this embodiment, the periodic nutritional status monitoring data of gastric cancer patients is collected at least once a month for 6 consecutive months after surgery.

[0067] S14. Combine the preoperative data, postoperative periodic reporting data and periodic nutritional status monitoring data of gastric cancer patients as the malnutrition risk monitoring data of the corresponding gastric cancer patients.

[0068] S2. Clean the initial malnutrition risk monitoring data set and set prediction labels and data masks to obtain a malnutrition risk dynamic time series training set;

[0069] The S2 comprises the following steps:

[0070] S21. If the preoperative nutritional status monitoring result of a gastric cancer patient is malnutrition in the initial malnutrition risk monitoring data set, the malnutrition risk monitoring data of the gastric cancer patient is deleted to obtain a first screening malnutrition risk monitoring data set;

[0071] S22. If the interval between adjacent postoperative periodic reporting dates of the same gastric cancer patient in the initial malnutrition risk monitoring data set exceeds a preset reporting interval threshold, or the adjacent postoperative nutritional status monitoring dates of the same gastric cancer patient exceed a preset monitoring interval threshold, the malnutrition risk monitoring data of the gastric cancer patient is deleted to obtain a second screening malnutrition risk monitoring data set;

[0072] In this embodiment, the reporting interval threshold is 14 days, and the monitoring interval threshold is 60 days.

[0073] S23, respectively setting prediction labels and data masks for the postoperative periodic report data of each gastric cancer patient in the second screening malnutrition risk monitoring data set to obtain a malnutrition risk dynamic time series training set;

[0074] The calculation expressions of the preset labels and data masks are as follows:

[0075] ,

[0076] ,

[0077] in, represents the predicted label of the i-th postoperative periodic report data of the k-th gastric cancer patient, e represents the exponential base constant, represents the i-th postoperative periodic report date of gastric cancer patients, represents the Jth postoperative nutritional status monitoring date corresponding to the kth gastric cancer patient suffering from malnutrition, The data mask representing the i-th postoperative periodic report data of the k-th gastric cancer patient, that is, the gastric cancer patient was found to be malnourished on the J-th postoperative nutritional status monitoring date. The data mask is used to mark the postoperative periodic report data of each gastric cancer patient used for training, so as to ensure that only the data up to the time point of malnutrition is used when training the model, and the data after malnutrition is no longer needed for prediction and does not need to be input for training.

[0078] S3. Construct a malnutrition risk prediction model;

[0079] like Figure 2 As shown, the malnutrition risk prediction model in S3 includes:

[0080] The first multi-layer perceptron is used to train the preoperative data of each gastric cancer patient in the malnutrition risk dynamic time series training set. Perform feature mapping to obtain a first mapping feature;

[0081] The second multi-layer perceptron is used to perform feature mapping on the postoperative periodic report data of each gastric cancer patient in the malnutrition risk dynamic time series training set to obtain the second mapping feature; the i-th postoperative periodic report data of the gastric cancer patient , the second mapping feature corresponding to the postoperative periodic report data can be obtained;

[0082] In this embodiment, the first multi-layer perceptron and the second multi-layer perceptron are used to extract features from the preoperative data and postoperative periodic report data of gastric cancer patients respectively, which can reduce the dimension of the features and eliminate the correlation between the feature dimensions, thereby reducing the complexity and overfitting risk of the subsequent model and enhancing the generalization ability of the model.

[0083] Concat splicing module, used to concat the first mapping feature and the second mapping feature to obtain a combined feature ;

[0084] The gated recurrent neural module is used to calculate the current hidden state using the combined features and the previous hidden state; the previous hidden state of the hidden state of the i-th postoperative malnutrition risk prediction is the hidden state of the i-1-th postoperative malnutrition risk prediction ;

[0085] The calculation expression of the hidden state is as follows:

[0086] ,

[0087] ,

[0088] ,

[0089] ,

[0090] in, represents the hidden state of the risk prediction of postoperative malnutrition at the i-th time, represents the output of the update gate, represents Hadamard multiplication, represents the hidden state of the i-1th postoperative malnutrition risk prediction, represents the candidate hidden state for the i-th postoperative malnutrition risk prediction, represents the sigmoid function, represents the weight matrix of the update gate, represents the combined features, represents vector concatenation, represents the output of the reset gate, represents the weight matrix of the reset gate, represents the tanh function;

[0091] The third multi-layer perceptron is used to perform feature mapping on the current hidden state to obtain a hidden state mapping result;

[0092] The activation module is used to activate the hidden state mapping results and obtain the risk prediction results of postoperative malnutrition for the i-th time. .

[0093] In this embodiment, the input end of the first multilayer perceptron serves as the first input end of the malnutrition risk prediction model; the input end of the second multilayer perceptron serves as the second input end of the malnutrition risk prediction model; the output end of the first multilayer perceptron is connected to the first input end of the Concat splicing module; the output end of the second multilayer perceptron is connected to the second input end of the Concat splicing module; the output end of the Concat splicing module is connected to the first input end of the gated recurrent neural module; the second input end of the gated recurrent neural module obtains a hidden state; the output end of the gated recurrent neural module is connected to the input end of the third multilayer perceptron; the output end of the third multilayer perceptron is connected to the input end of the activation module; the output end of the activation module serves as the output end of the malnutrition risk prediction model.

[0094] The malnutrition risk prediction model in S3 is used to iteratively predict and update the preoperative data and postoperative periodic report data of each gastric cancer patient in the malnutrition risk dynamic time series training set, and obtain each postoperative malnutrition risk prediction result of the gastric cancer patient and the hidden state of the postoperative malnutrition risk prediction;

[0095] The calculation expression of the malnutrition risk prediction model is as follows:

[0096] ,

[0097] , ,

[0098] in, represents the risk prediction result of malnutrition after surgery for the kth gastric cancer patient. represents the hidden state of the risk prediction of malnutrition after surgery for the kth gastric cancer patient, represents the malnutrition risk prediction model, represents the preoperative data of the kth gastric cancer patient, represents the periodic report data of the i-th postoperative period for the k-th gastric cancer patient, represents the hidden state of the risk prediction of malnutrition after surgery for the kth gastric cancer patient i-1 times, represents the sigmoid function, represents the multi-layer perceptron feature map, represents the initial hidden state.

[0099] like Figure 3 As shown in this protocol, the first postoperative periodic reporting date for gastric cancer patients is When the preoperative data of gastric cancer patients were input into the malnutrition risk prediction model , first postoperative periodic report data and the initial hidden state , output the first postoperative malnutrition risk prediction results for gastric cancer patients and the hidden state of the first postoperative malnutrition risk prediction ; Similar processing is performed on each subsequent postoperative periodic reporting date of gastric cancer patients, that is, the preoperative data of gastric cancer patients are input into the malnutrition risk prediction model , the i-th postoperative periodic report data and the hidden state of the i-1th postoperative malnutrition risk prediction , we can get the risk prediction results of malnutrition after the i-th surgery for gastric cancer patients and the hidden state of the risk prediction of postoperative malnutrition at the i-th time , such as: to malnutrition risk prediction model Enter the preoperative data of gastric cancer patients Hidden state of risk prediction for postoperative malnutrition at the first time and the second postoperative periodic report data After that, the risk prediction results of malnutrition after the second surgery for gastric cancer patients can be obtained. and the hidden state of risk prediction for postoperative malnutrition after the second surgery , and then to the malnutrition risk prediction model Enter the preoperative data of gastric cancer patients Hidden state of risk prediction for postoperative malnutrition after the second surgery and the third postoperative periodic report data After that, the risk prediction results of malnutrition after the third surgery for gastric cancer patients can be obtained. and the hidden state of risk prediction of postoperative malnutrition at the 3rd time . The malnutrition risk prediction model based on the scheme of the present invention can predict a postoperative malnutrition risk prediction result based on each postoperative periodic report data of gastric cancer patients. The postoperative malnutrition prediction result not only depends on the physiological, psychological and dietary status of the gastric cancer patient at the time of the postoperative periodic report, but also depends on the demographic factors, disease, surgical conditions and previous historical information of the gastric cancer patient, among which the historical information refers to the hidden state of the last postoperative malnutrition risk prediction, so that the risk of malnutrition in patients can be more accurately predicted based on the gastric cancer patient's own information and physiological, psychological and dietary dynamic changes.

[0100] S4. According to the malnutrition risk prediction loss function, the malnutrition risk prediction model is trained using the preoperative data and postoperative periodic report data of each gastric cancer patient in the malnutrition risk dynamic time series training set to obtain a trained malnutrition risk prediction model;

[0101] The calculation expression of the malnutrition risk prediction loss function in S4 is as follows:

[0102] ,

[0103] ,

[0104] ,

[0105] Among them, L represents the loss function of malnutrition risk prediction results, represents the loss function of malnutrition risk prediction results, represents the weight adjustment parameter, represents the constraint mapping loss function, represents the identity matrix, represents the risk prediction result of malnutrition after the i-th surgery for gastric cancer patients, represents the first mapping feature matrix, represents the transpose of the first mapping feature matrix, represents the second mapping feature matrix, represents the transpose of the second mapping feature matrix, represents the Frobenius norm of the matrix, where each row in the first mapping feature matrix corresponds to the first mapping feature of the kth patient. , each row in the second mapping feature matrix corresponds to the second mapping feature of the kth patient .

[0106] In this scheme, the malnutrition risk prediction result loss function uses the absolute error function to calculate the malnutrition risk prediction result of the i-th postoperative gastric cancer patient Prediction labels of the i-th postoperative periodic report data of gastric cancer patients The constraint mapping loss function ensures the first mapping feature after constraint mapping. and the second mapping feature The dimensional features in are linearly independent. By minimizing the malnutrition risk prediction loss function and using the gradient descent algorithm to optimize the parameters of the malnutrition risk prediction model, a trained malnutrition risk prediction model is obtained.

[0107] S5. Obtain the malnutrition risk monitoring data of new gastric cancer patients, and use the trained malnutrition risk prediction model to perform risk prediction and early warning for the gastric cancer patients.

[0108] The S5 comprises the following steps:

[0109] S51. Obtain preoperative data of new gastric cancer patients after the surgical method is determined, as well as several postoperative periodic reporting data of the gastric cancer patients;

[0110] In this embodiment, new gastric cancer patients do not refer to patients who have just been diagnosed with gastric cancer, but rather gastric cancer patients who meet both the first screening malnutrition risk monitoring data set requirements in S21 and the second screening malnutrition risk monitoring data set requirements in S22.

[0111] S52, inputting the preoperative data and each postoperative periodic report data of the new gastric cancer patient into the trained malnutrition risk prediction model, and predicting the malnutrition risk prediction result of the new gastric cancer patient after each postoperative periodic report;

[0112] The calculation expression of the malnutrition risk prediction result of the new gastric cancer patient after each postoperative periodic report is as follows:

[0113] ,

[0114] in, represents the malnutrition risk prediction result of new gastric cancer patients after the i-th postoperative periodic report, represents the preoperative data of new gastric cancer patients, represents the periodic report data of the i-th postoperative period for new gastric cancer patients;

[0115] S53, respectively comparing the malnutrition risk prediction result of the new gastric cancer patient after each postoperative periodic report with the malnutrition risk threshold, if the malnutrition risk prediction result of the existing gastric cancer patient after each postoperative periodic report is greater than the malnutrition risk threshold, then generating a malnutrition warning for the postoperative periodic report of the new gastric cancer patient. In this embodiment, the value range of the malnutrition risk threshold is 0.5-0.75.

[0116] This program can accurately predict and analyze the malnutrition risk of gastric cancer patients for each postoperative periodic report, can promptly remind medical staff to provide nutritional support and intervention for gastric cancer patients, and can effectively maintain the nutritional health of gastric cancer patients.

[0117] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery, characterized in that: The steps include: S1. Obtain malnutrition risk monitoring data of several gastric cancer patients to form an initial malnutrition risk monitoring data set; S2. Clean the initial malnutrition risk monitoring data set and set prediction labels and data masks to obtain a malnutrition risk dynamic time series training set; S3. Construct a malnutrition risk prediction model; S4. According to the malnutrition risk prediction loss function, the malnutrition risk prediction model is trained using the preoperative data and postoperative periodic report data of each gastric cancer patient in the malnutrition risk dynamic time series training set to obtain a trained malnutrition risk prediction model; S5. Obtain the malnutrition risk monitoring data of new gastric cancer patients, and use the trained malnutrition risk prediction model to perform risk prediction and early warning for the gastric cancer patients.

2. The method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery according to claim 1, characterized in that: The S1 comprises the following steps: S11. Obtaining preoperative data of gastric cancer patients, wherein the preoperative data of gastric cancer patients includes demographic data and medical related data; S12. Obtain postoperative periodic reporting data of gastric cancer patients; S13, obtaining periodic nutritional status monitoring data of gastric cancer patients, wherein the periodic nutritional status monitoring data includes preoperative nutritional status monitoring results and several postoperative nutritional status monitoring results of gastric cancer patients; S14. Combine the preoperative data, postoperative periodic reporting data and periodic nutritional status monitoring data of gastric cancer patients as the malnutrition risk monitoring data of the corresponding gastric cancer patients.

3. The method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery according to claim 2, characterized in that: The S2 comprises the following steps: S21. If the preoperative nutritional status monitoring result of a gastric cancer patient is malnutrition in the initial malnutrition risk monitoring data set, the malnutrition risk monitoring data of the gastric cancer patient is deleted to obtain a first screening malnutrition risk monitoring data set; S22. If the interval between adjacent postoperative periodic reporting dates of the same gastric cancer patient in the initial malnutrition risk monitoring data set exceeds a preset reporting interval threshold, or the adjacent postoperative nutritional status monitoring dates of the same gastric cancer patient exceed a preset monitoring interval threshold, the malnutrition risk monitoring data of the gastric cancer patient is deleted to obtain a second screening malnutrition risk monitoring data set; S23, respectively setting prediction labels and data masks for the postoperative periodic report data of each gastric cancer patient in the second screening malnutrition risk monitoring data set to obtain a malnutrition risk dynamic time series training set; The calculation expressions of the preset labels and data masks are as follows: ， ， in, represents the predicted label of the i-th postoperative periodic report data of the k-th gastric cancer patient, e represents the exponential base constant, represents the i-th postoperative periodic report date of gastric cancer patients, represents the Jth postoperative nutritional status monitoring date corresponding to the kth gastric cancer patient suffering from malnutrition, The data mask representing the i-th postoperative periodic report data of the k-th gastric cancer patient.

4. The method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery according to claim 3, characterized in that: The malnutrition risk prediction model in S3 includes: The first multi-layer perceptron is used to perform feature mapping on the preoperative data of each gastric cancer patient in the malnutrition risk dynamic time series training set to obtain a first mapping feature; The second multi-layer perceptron is used to perform feature mapping on the postoperative periodic report data of each gastric cancer patient in the malnutrition risk dynamic time series training set to obtain a second mapping feature; A Concat module, used to perform Concat concatenation on the first mapping feature and the second mapping feature to obtain a combined feature; The gated recurrent neural module is used to calculate the current hidden state using the combined features and the previous hidden state; The third multi-layer perceptron is used to perform feature mapping on the current hidden state to obtain a hidden state mapping result; The activation module is used to activate the hidden state mapping results to obtain the i-th postoperative malnutrition risk prediction result.

5. The method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery according to claim 4, characterized in that: The malnutrition risk prediction model in S3 is used to iteratively predict and update the preoperative data and postoperative periodic report data of each gastric cancer patient in the malnutrition risk dynamic time series training set, and obtain each postoperative malnutrition risk prediction result of the gastric cancer patient and the hidden state of the postoperative malnutrition risk prediction; The calculation expression of the malnutrition risk prediction model is as follows: ， ， ， in, represents the risk prediction result of malnutrition after surgery for the kth gastric cancer patient. represents the hidden state of the risk prediction of malnutrition after surgery for the kth gastric cancer patient, represents the malnutrition risk prediction model, represents the preoperative data of the kth gastric cancer patient, represents the periodic report data of the i-th postoperative period for the k-th gastric cancer patient, represents the hidden state of the risk prediction of malnutrition after surgery for the kth gastric cancer patient i-1 times, represents the sigmoid function, represents the multi-layer perceptron feature map, represents the initial hidden state.

6. The method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery according to claim 5, characterized in that: The calculation expression of the malnutrition risk prediction loss function in S4 is as follows: ， ， ， Among them, L represents the loss function of malnutrition risk prediction results, represents the loss function of malnutrition risk prediction results, represents the weight adjustment parameter, represents the constraint mapping loss function, represents the identity matrix, represents the risk prediction result of malnutrition after the i-th surgery for gastric cancer patients, represents the first mapping feature matrix, represents the transpose of the first mapping feature matrix, represents the second mapping feature matrix, represents the transpose of the second mapping feature matrix, Represents the Frobenius norm of the matrix.

7. The method for predicting the risk of malnutrition in patients undergoing gastric cancer surgery according to claim 6, characterized in that: The S5 package contains the following steps: S51. Obtain preoperative data of new gastric cancer patients after the surgical method is determined, as well as several postoperative periodic reporting data of the gastric cancer patients; S52, inputting the preoperative data and each postoperative periodic report data of the new gastric cancer patient into the trained malnutrition risk prediction model, and predicting the malnutrition risk prediction result of the new gastric cancer patient after each postoperative periodic report; S53. Compare the malnutrition risk prediction result of new gastric cancer patients after each postoperative periodic report with the malnutrition risk threshold. If the malnutrition risk prediction result of existing gastric cancer patients after each postoperative periodic report is greater than the malnutrition risk threshold, a malnutrition warning is generated for the new gastric cancer patient's postoperative periodic report.

Images