Dynamic analysis method for personalized nutrition assessment of malignant tumor patient based on blood data

Abstract

The invention discloses a malignant tumor patient personalized nutrition assessment dynamic analysis method based on blood data, and relates to the technical field of personalized nutrition assessment, and the method comprises the steps: taking a multilayer perceptron neural network as a basic model to train a data set S2 of blood routine and biochemical data, and obtaining model1; adding the historical blood routine and biochemical data of the person to be tested into the data set S2 to form a data set S3, and re-assigning the weight to the data set S3; s3, the parameters of the second layer and the third layer of the model1 are activated, training is continued, the training epoch is 200, and model2 is obtained. The model structure and the parameters of the model1 are reloaded, and the parameters of the first layer of the model1 are fixed, and the parameters of the second layer and the third layer of the model1 are activated to continue training. According to the method, customized nutrition state evaluation is provided according to unique physiological changes of each person and data in the treatment process, compared with a general model, the method can better fit individual differences, and meanwhile, the model has the ability of re-training by continuously incorporating new data, and the prediction accuracy and adaptability of the model are continuously improved along with accumulation of historical data of patients.

Description

Technical Field

[0001] The present invention relates to the technical field of medical testing and disease identification, and in particular to a dynamic analysis method for personalized nutrition assessment of malignant tumor patients based on blood data. Background Art

[0002] As the incidence of malignant tumors gradually increases, the incidence of complications such as malnutrition is also on the rise. However, the rate of clinical nutritional treatment for patients with malignant tumors is low. Malnutrition caused by malignant tumors urgently needs to be paid attention to by patients and medical staff.

[0003] At present, although there are nutritional assessment scales such as NRS2002, this method has the following disadvantages when applied to the medical industry:

[0004] (1) The scale evaluation method is highly subjective and takes a long time to evaluate, and cannot timely detect the nutritional status of patients with malignant tumors;

[0005] (2) The evaluation scores of the scale are fixed and cannot achieve individual precision medicine for a particular person. The evaluation method for a particular person can only be evaluated using the same subscale, which is highly subjective.

[0006] Although there are currently artificial intelligence methods for nutritional assessment, some testers experience poor model testing results due to different data distribution, testing instruments, or insufficient model generalization performance. Summary of the invention

[0007] To solve the above problems, the present invention provides a dynamic analysis method for personalized nutritional assessment of malignant tumor patients based on blood data, which uses artificial intelligence methods to perform accurate case-by-case assessment of the nutritional status of tumor patients.

[0008] To achieve the above object, the present invention provides a dynamic analysis method for personalized nutritional assessment of malignant tumor patients based on blood data, comprising the following steps:

[0009] (1) After data sorting and screening, a data set S2 based on routine blood tests and biochemical data was obtained, in which the ratio of training set to test set was 4:1;

[0010] (2) Taking the multi-layer perceptron MLP as the basic model, the dataset S2 is trained to obtain model_1;

[0011] (3) The historical blood routine and biochemical data set D1 of the person to be tested is extracted, and the standardized data set D1 is added to the data set S2 to form the data set S3, and the data set S3 is re-weighted;

[0012] (4) Load the model structure and parameters of model_1, fix the first layer parameters of model_1, activate the second and third layer parameters of model_1, continue to train the data set S3, and obtain model_2;

[0013] (5) Model_2 is used to test the blood routine and biochemical data of the test subjects to obtain personalized classification results.

[0014] Preferably, in step (2), the input dimensions when training the data set S2 include:

[0015] 'sex','age','RBC','MCV','PDW','WBC','NEUT%','LYMPH%','EO%','BASO%','NEUT#','LYMPH#','BASO#','HGB','HCT','MCH','MCHC','R-CV','PLT','MPV','PCT','MONO#','MONO%','EO#','GLU','CREA','UREA','ALB','ALT','AST' and the presence or absence of malnutrition label.

[0016] Preferably, in step (2), the training epoch is 200.

[0017] Preferably, step (3) includes: adding the historical blood routine and biochemical data D1 of the person to be tested to the data set S2 after standardization to form the data set S3, D1 is other data sources that do not belong to S2, and the test set T1 of the S2 data set does not change; re-weighting the data set S3, D1:S2=1.05:1.

[0018] Preferably, in step (4), the epoch number when activating the second and third layer parameters of model_1 and continuing training is 200.

[0019] The present invention provides a dynamic analysis method for personalized nutrition assessment of malignant tumor patients based on blood data, which has the following beneficial effects:

[0020] (1) Provide customized nutritional status assessment based on each person's unique physiological changes and treatment data. Compared with general models, it is more in line with individual differences. At the same time, the model is able to be retrained by continuously incorporating new data. As the patient's historical data accumulates, the model's predictive accuracy and adaptability will continue to improve;

[0021] (2) During the incremental learning process of model_1, the first-layer parameters of model_1 were fixed, and the second-layer and third-layer parameters of model_1 were activated to continue training. The test accuracy of the obtained model_2 for the test data of the test patients was significantly improved.

[0022] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a schematic diagram of the overall process of a dynamic analysis method for personalized nutrition assessment of malignant tumor patients based on blood data of the present invention;

[0024] Figure 2 Select a flowchart for your data. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical scheme and advantages disclosed in the embodiments of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0026] Embodiment 1

[0027] A dynamic analysis method for personalized nutritional assessment of malignant tumor patients based on blood data, the process is as follows Figure 1 As shown, the specific steps include:

[0028] S1. Data selection and processing, such as Figure 2 shown.

[0029] The blood routine and biochemical data set S1 based on blood was selected, and the dimensions were sex, age, red blood cell count (RBC), mean corpuscular volume (MCV), platelet distribution width (PDW), white blood cell count (WBC), neutrophil percentage (NEUT%), lymphocyte percentage (LYMPH%), eosinophil percentage (EO%), basophil percentage (BASO%), absolute neutrophil value (NEUT#), absolute lymphocyte value (LYMPH#), absolute basophil value (BASO#), hemoglobin (HGB), and hematocrit (HCT). , mean corpuscular hemoglobin content (MCH), mean corpuscular hemoglobin concentration (MCHC), red blood cell volume distribution width (R-CV), platelet count (PLT), mean platelet volume (MPV), platelet packed volume (PCT), monocyte absolute value (MONO#), monocyte percentage (MONO%), eosinophil absolute value (EO#), glucose (GLU), creatinine (CREA), urea (UREA), albumin (ALB), alanine aminotransferase (ALT), aspartate aminotransferase (AST), a total of 3404 cases, the data set comes from Wuhai People's Hospital.

[0030] After the data set S1 was sorted, a total of 2969 malignant tumor cases were obtained, which is the data set S2. Among them, there are 669 malnutrition cases (positive cases) and 2300 normal nutrition cases (negative cases).

[0031] Inclusion criteria: ① patients diagnosed with tumors; ② aged 18 to 90 years old; ③ conscious, and the patients and their families are willing to cooperate with this survey.

[0032] Exclusion criteria: ① Patients with pacemaker or arterial stent implantation; ② Patients with severe complications such as massive bleeding and severe infection; ③ Patients unable to cooperate for other reasons.

[0033] After preprocessing and feature selection, the dataset S2 was divided into training set and test set. The training set T0 consisted of 2375 cases, and the test set T1 consisted of 594 cases.

[0034] S2, using the multi-layer perceptron MLP as the basic model, train the dataset S2. Select the input dimensions as 'sex', 'age', 'RBC', 'MCV', 'PDW', 'WBC', 'NEUT%', 'LYMP H%', 'EO%', 'BASO%', 'NEUT#', 'LYMPH#', 'BASO#', 'HGB', 'HCT', 'MCH', 'MCHC', 'R-CV', 'PLT', 'MPV', 'PCT', 'MONO#', 'MONO%', 'EO#', 'GLU', 'CR EA', 'UREA', 'ALB', 'ALT', 'AST' and the label of whether there is malnutrition. After 200 epochs, model_1 is obtained.

[0035] S3, extract the historical data of the blood routine and biochemical data of the person to be tested, use model_1 for incremental learning, add the person's data to the data set S2, and re-assign different weights to the data.

[0036] Specifically include:

[0037] S31. The historical data set D1 of the blood routine and biochemical data of the person to be tested is standardized and added to the data set S2 to form the data set S3. D1 is other data sources that do not belong to S2. The test set T1 of the S2 data set does not change.

[0038] S32. Re-weight the data set S3, D1:S2=1.05:1.

[0039] The amount of historical data of the test subjects is not large, so through different weighting ratios, a higher sampling ratio can be given to the test subjects' data in the model training phase, thereby increasing the emphasis on the test subjects' data.

[0040] S4, reload the model structure and parameters saved in model_1, and retrain the data set S3. In order to retain the generalization ability of the model, fix the first layer network (input layer) with the largest number of parameters, release the second layer (hidden layer) and the third layer (output layer) network for training, and obtain model_2.

[0041] The first layer model has the largest number of parameters, so retaining the first layer network can ensure that the model's fitting performance for data set S2 is not weakened. At the same time, the second and third layers of the network have small parameters and can fit the data of the test subjects well. This method can ensure that the generalization performance of the overall model will not be weakened in the process of fitting new data.

[0042] S5. Use the trained model_2 to test the data of the test subjects that are not in S2 to obtain personalized and accurate results. The evaluation indicators used for the test results are: TPR, TNR, and ACC.

[0043] The calculation formulas are:

[0044] TPR = TP / (TP+FP);

[0045] TNR = TN / (TN + FP);

[0046] ACC=(TP+TN) / (TP+FP+FN+TN);

[0047] Among them, TP represents the number of correctly identified positives; TN represents the number of correctly identified negatives; FP represents the number of falsely identified positives; FN represents the number of falsely identified negatives; TPR represents the sensitivity of the proportion of correctly identified positives to all positives; TNR represents the specificity of correctly identified negatives to all negatives; ACC represents the accuracy of the model predicting the classification results.

[0048] Test Example 1

[0049] The test set T1 is tested using model_1 and model_2. The comparison of the test results of the two models is shown in Table 1.

[0050] Table 1 Comparison of algorithm results

[0051]

[0052] Test Example 2

[0053] From the newly added data that model_1 and model_2 have not been trained on and that meet the inclusion and exclusion criteria, a test subject with a large number of experiments was selected, and his blood routine and biochemical data were used to test model_1 and model_2. The test results are shown in Table 2 below:

[0054] Table 2

[0055]

[0056]

[0057] From the analysis of Table 2, we can see that the test results of model_1 for the test person's data are very poor due to data distribution or instrument differences, with an accuracy of only 1 / 13, while the accuracy of the test results of the trained model_2 for the test person is improved to 8 / 13. According to Table 1, for the test set T1 constructed for the S2 data set, the accuracy of model_1 and model_2 is comparable, indicating that this method does not overfit or produce model deviation due to the addition of D1 data, but has better test results for data with poor test results due to data distribution or testing instruments while ensuring the generalization performance of the model.

[0058] Table 3

[0059]

[0060]

[0061] In Table 3, several testers with a large number of experiments were re-extracted, and their blood routine and biochemical data were used to test model_2. The test results are shown in Table 3 above. It can be seen that the scores are missing due to patient non-cooperation, sample problems or other factors. Traditionally, when faced with a vacancy in the real score, doctors often find it difficult to make accurate judgments, which not only affects the efficiency of diagnosis, but may also lead to the ineffectiveness of the patient's treatment plan or aggravate the condition. However, with the help of advanced machine learning algorithms, even in the absence of some real scores, reliable dynamic graded evaluation probability values ​​can be obtained to help doctors understand the possible distribution of the patient's status and obtain more comprehensive information to support their decision-making process. This shows that this method can perform personalized dynamic nutritional assessments for patients with malignant tumors and solve the problem of differences between patients.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A dynamic analysis method for personalized nutritional assessment of malignant tumor patients based on blood data, characterized in that: The steps include: (1) After data sorting and screening, a data set S2 based on routine blood tests and biochemical data was obtained, in which the ratio of training set to test set was 4:1; (2) Taking the multi-layer perceptron MLP as the basic model, the dataset S2 is trained to obtain model_1; (3) The historical blood routine and biochemical data set D1 of the person to be tested is extracted, and the standardized data set D1 is added to the data set S2 to form the data set S3, and the data set S3 is re-weighted; (4) Load the model structure and parameters of model_1, fix the first layer parameters of model_1, activate the second and third layer parameters of model_1, continue to train the data set S3, and obtain model_2; (5) Model_2 is used to test the blood routine and biochemical data of the test subjects to obtain personalized classification results.

2. According to claim 1, a dynamic analysis method for personalized nutritional assessment of malignant tumor patients based on blood data is characterized in that: In step (2), the input dimensions for training the dataset S2 include: 'sex','age','RBC','MCV','PDW','WBC','NEUT%','LYMPH%','EO%','BASO%','NEUT#','LYMPH#','BASO#','HGB','HCT','MCH','MCHC','R-CV','PLT','MPV','PCT','MONO#','MONO%','EO#','GLU','CREA','UREA','ALB','ALT','AST' and the presence or absence of malnutrition label.

3. According to claim 1, a dynamic analysis method for personalized nutritional assessment of malignant tumor patients based on blood data is characterized in that: In step (2), the training epoch is 200.

4. The method for dynamic analysis of personalized nutritional assessment of malignant tumor patients based on blood data according to claim 1, characterized in that: In step (3), data set D1 is other data sources that do not belong to S2, and the test set T1 of the S2 data set does not change; the data set S3 is re-weighted, D1:S2=1.05:

1.

5. According to claim 1, a dynamic analysis method for personalized nutritional assessment of malignant tumor patients based on blood data is characterized in that: In step (4), the second and third layer parameters of model_1 are activated and the training continues for epoch 200.

Images