Medical information prompting device and method, equipment and storage medium

Abstract

The invention relates to the technical field of medical information processing, and discloses a medical information prompting device, method and equipment and a storage medium, and the device comprises a parameter obtaining module which is used for obtaining an inflammation biomarker of a target object, the inflammation biomarker comprises one or more of a ratio of neutrophil to lymphocyte, a ratio of platelet to lymphocyte, a ratio of lymphocyte to monocyte and an international standardized ratio; the prediction module is used for inputting the inflammation biomarker into a neural network model to obtain the output of the neural network model; the determination module is used for determining corresponding prompt information according to the output of the neural network model; and the display module is used for displaying the prompt information. According to the invention, health risk prompting can be carried out based on the immune cell ratio, so that people can carry out health management in an earlier stage.

Description

Technical Field

[0001] The present invention relates to the technical field of medical information processing, and in particular to a medical information prompting device, method, equipment and storage medium. Background Art

[0002] As people pay more and more attention to their health, more and more people are beginning to realize the importance of regular physical examinations. Regular examinations can not only detect potential diseases early, but also take timely measures to intervene and treat them, thereby effectively preventing the disease from worsening. Therefore, it is particularly necessary to provide risk warnings based on the examination results. Such risk warnings can not only help individuals better understand their health status, but also guide them to take appropriate preventive and treatment measures to reduce health risks.

[0003] However, in the related art, health risk warnings can only be given based on some clear abnormalities in the examination parameters. If there are no clear abnormalities in the examination parameters, there will be no warnings, which is not conducive to people's health management. Summary of the invention

[0004] In view of this, the present invention provides a medical information prompting device, method, equipment and storage medium to solve the problem of the current single health check result prompting method.

[0005] In a first aspect, the present invention provides a medical information prompting device, the device comprising:

[0006] a parameter acquisition module, configured to acquire inflammation biomarkers of a target object, wherein the inflammation biomarkers include one or more of a neutrophil to lymphocyte ratio, a platelet to lymphocyte ratio, a lymphocyte to monocyte ratio, and an international normalized ratio;

[0007] A prediction module, used for inputting the inflammatory biomarker into a neural network model to obtain an output of the neural network model;

[0008] A determination module, used to determine corresponding prompt information according to the output of the neural network model;

[0009] The display module is used to display the prompt information.

[0010] In an optional embodiment, the neural network model includes a layer with 5 neurons, using a rectified linear unit as an activation function;

[0011] The neural network model uses the Adam optimizer to update weights during training, uses full batch training, and the training iterations are controlled by the maximum number of iterations parameter.

[0012] In an optional implementation, the prediction module includes:

[0013] a target parameter acquisition unit, configured to acquire target parameters, wherein the target parameters include the ALBI score, AFP level, age, and tumor size of the target subject;

[0014] A prediction unit is used to input the target parameter and the inflammatory biomarker into a neural network model to obtain an output of the neural network model.

[0015] In an optional embodiment, the sample data acquisition module is used to acquire a plurality of sample data, wherein the sample data includes candidate variables and corresponding survival times, and the candidate variables include at least the target parameter and the inflammatory biomarker;

[0016] An extraction module, used for extracting a plurality of training subsets from the plurality of sample data by adopting a sampling method with replacement;

[0017] A first construction module is used to construct a survival decision tree for each of the training subsets, and select a preset number of candidate variables for splitting at each node of the survival decision tree;

[0018] A second building module is used to integrate the plurality of survival decision trees into a random survival forest model;

[0019] A performance evaluation module, for randomly disrupting the value of each target candidate variable in the candidate variables in the out-of-bag sample data, and re-evaluating the prediction performance of the random survival forest model;

[0020] A scoring calculation module, used to calculate the difference in prediction performance of the random survival forest model before and after the shuffle as the importance score of the target candidate variable;

[0021] A screening module is used to screen out the target parameter and the inflammatory biomarker from the candidate variables according to the importance score.

[0022] In an optional implementation, the first building block includes:

[0023] A selection unit, configured to evaluate each of the candidate variables and select a variable that can maximize the survival rate difference as a split point;

[0024] A recursive splitting unit, used for splitting the data in the training subset layer by layer, wherein each layer is divided according to different candidate variables and thresholds until a stopping condition is met;

[0025] The decision tree construction unit is used to divide the data into a multi-layer structure according to the splitting of different candidate variables to form a decision tree.

[0026] In an optional implementation, the determining module includes:

[0027] An output acquisition unit, used to obtain a probability value output by the neural network model;

[0028] A prompt information determination unit, used to determine the corresponding prompt information according to the magnitude relationship between the probability value and a preset probability threshold, wherein the prompt information includes high-risk prompt information and low-risk prompt information;

[0029] Among them, the preset probability threshold is the threshold corresponding to when the AUC of the neural network model is the highest.

[0030] In a second aspect, the present invention provides a medical information prompting method, the method comprising:

[0031] Obtaining inflammation biomarkers of the target subject, wherein the inflammation biomarkers include one or more of a neutrophil to lymphocyte ratio, a platelet to lymphocyte ratio, a lymphocyte to monocyte ratio, and an international normalized ratio;

[0032] Inputting the inflammatory biomarker into a neural network model to obtain an output of the neural network model;

[0033] Determining corresponding prompt information according to the output of the neural network model;

[0034] The prompt information is displayed.

[0035] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the medical information prompt method of the second aspect or any corresponding embodiment thereof by executing the computer instructions.

[0036] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the medical information prompting method of the second aspect or any corresponding embodiment thereof.

[0037] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the medical information prompting method of the second aspect or any corresponding embodiment thereof.

[0038] The medical information prompting device, method, equipment and storage medium provided in the embodiments of the present invention use a machine learning model to analyze inflammatory biomarkers, output prediction results, and then determine corresponding prompting information based on the prediction results, so that people can monitor and control health risks at an earlier stage. Compared with the prior art, which only prompts abnormal items of examination indicators when there are clear abnormalities in the examination indicators, the present application can provide health risk prompts instead of just prompts for abnormal items of examination indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related technologies, the drawings required for use in the specific embodiments or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 is a structural block diagram of a medical information prompting device according to an embodiment of the present invention;

[0041] Figure 2 is a schematic diagram of the architecture of a neural network model according to an embodiment of the present invention;

[0042] Figure 3 is a schematic diagram of a prediction factor-importance score based on random survival forest analysis according to an embodiment of the present invention;

[0043] Figure 4 is a schematic diagram of the importance ranking of out-of-bag variables based on the prediction factor and random survival forest analysis according to an embodiment of the present invention;

[0044] Figure 5 is a schematic diagram of prediction factors-error rates based on random survival forest analysis according to an embodiment of the present invention;

[0045] Figure 6 The decision tree model according to an embodiment of the present invention shows the 5-year overall survival probability of patients with hepatocellular carcinoma;

[0046] Figure 7 is a schematic diagram of a patient inclusion process according to an embodiment of the present invention;

[0047] Figure 8 is a schematic diagram of ROC curves for predicting 1-year, 3-year and 5-year mortality rates based on training group data using a neural network model according to an embodiment of the present invention;

[0048] Fig. 9 is a schematic diagram of ROC curves for predicting 1-year, 3-year and 5-year mortality rates based on validation group data using a neural network model according to an embodiment of the present invention;

[0049] Fig.10 is a schematic diagram of a calibration curve for predicting 5-year overall survival rate based on training group data using a neural network model according to an embodiment of the present invention;

[0050] Fig.11 is a schematic diagram of a calibration curve for predicting 5-year overall survival rate based on validation group data using a neural network model according to an embodiment of the present invention;

[0051] Fig.12 is a schematic diagram of a decision curve analysis of a neural network model according to an embodiment of the present invention predicting a 5-year mortality rate based on training group data;

[0052] Fig.13 is a schematic diagram of a decision curve analysis of a neural network model according to an embodiment of the present invention predicting a 5-year mortality rate based on validation group data;

[0053] Fig.14 is a schematic diagram of a Kaplan-Meier survival curve of a training group according to an embodiment of the present invention;

[0054] Fig.15 is a schematic diagram of a survival curve in a validation group according to an embodiment of the present invention;

[0055] Fig.16 is a schematic diagram of a survival curve of HBV-infected patients according to an embodiment of the present invention;

[0056] Fig.17 is a schematic diagram of a survival curve of a patient without HBV infection according to an embodiment of the present invention;

[0057] Fig.18 is a schematic diagram of a survival curve for a Child-Pugh A stage patient according to an embodiment of the present invention;

[0058] Fig.19 is a schematic diagram of a survival curve for a Child-Pugh B stage patient according to an embodiment of the present invention;

[0059] Fig. 20 is a schematic diagram of a survival curve of a patient with AFP≤400ng / ml according to an embodiment of the present invention;

[0060] Fig.21 is a schematic diagram of a survival curve for patients with AFP>400ng / ml according to an embodiment of the present invention;

[0061] Fig. 22 Schematic diagram of the difference in lymphocyte-monocyte ratio (LMR) levels between the survival group and the non-survival group of patients according to an embodiment of the present invention;

[0062] Fig.23is a schematic diagram of the difference in platelet-to-lymphocyte ratio (PLR) levels between the survival group and the non-survival group of patients according to an embodiment of the present invention;

[0063] Fig.24 Schematic diagram of the difference in neutrophil-to-lymphocyte ratio (NLR) levels between the survival group and the non-survival group according to an embodiment of the present invention;

[0064] Fig.25 is a schematic flow chart of a medical information prompting method according to an embodiment of the present invention;

[0065] Fig.26 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0067] According to an embodiment of the present invention, a medical information prompt method embodiment is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of executable computer instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0068] In this embodiment, a medical information prompting device is provided. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0069] This embodiment provides a medical information prompting device, such as Figure 1 As shown, including:

[0070] The parameter acquisition module 101 is used to acquire inflammatory biomarkers of the target object, wherein the inflammatory biomarkers include one or more of the neutrophil-to-lymphocyte ratio (NLR), the platelet-to-lymphocyte ratio (PLR), the lymphocyte-to-monocyte ratio (LMR), and the international normalized ratio (INR).

[0071] The prediction module 102 is used to input the inflammatory biomarker into a neural network model (ie, an artificial neural network (ANN) model) to obtain the output of the neural network model.

[0072] Specifically, the above neural network model uses a multi-layer perceptron (MLP) structure, including input neurons (I1-I7), hidden layers (H1-H3) and output neurons (O1, O2). The artificial neural network (ANN) model is built using the Python package 'PyTorch'. The model performance can be optimized by adding three hidden layers through multiple tests.

[0073] The determination module 103 is used to determine the corresponding prompt information according to the output of the neural network model. The target object may be a BCLC0-A liver cancer patient, where BCLC is the Barcelona Clinical Liver Cancer Staging System. The output of the neural network model may be a survival rate, such as a one-year survival rate, a three-year survival rate, and a five-year survival rate.

[0074] The display module 104 is used to display the prompt information.

[0075] The medical information prompting device provided in the embodiment of the present invention can be implemented as a medical information prompting website.

[0076] The medical information prompting device provided in this embodiment uses a machine learning model to analyze inflammatory biomarkers, outputs prediction results, and then determines corresponding prompting information based on the prediction results, so that people can monitor and control health risks at an earlier stage. Compared with the prior art, which only prompts abnormal items of examination indicators when there are clear abnormalities in the examination indicators, the present application can provide health risk prompts instead of just prompts for abnormal items of examination indicators.

[0077] In addition, the neural network model is a non-invasive, low-cost model based on machine learning that is capable of processing large and complex data sets and generating more accurate and personalized prediction models through its unique training algorithm. The neural network model manages nonlinear statistical relationships better than traditional analytical methods.

[0078] In some optional embodiments, the neural network model includes a layer with 5 neurons, using a rectified linear unit (ReLU) as an activation function;

[0079] The neural network model uses the Adam optimizer to update weights during training, uses full batch training, and the training iterations are controlled by the maximum number of iterations (stepmax) parameter.

[0080] In the embodiment of the present invention, the training process of the neural network model includes:

[0081] 1. Sample data preparation

[0082] Divide the sample data set: Divide the original sample data set into a training set and a test set. The test set can account for 10% to 30% of the total data set.

[0083] Cross-validation segmentation: further divide the data subsets used for cross-validation from the training set. In the embodiment of the present invention, 5-fold cross-validation is used, so the training set is divided into 5 equal parts, 4 of which are used as training data each time, and the remaining 1 part is used as validation data.

[0084] 2. Model setup

[0085] Select model architecture: The artificial neural network model selected in the embodiment of the present invention includes a layer with 5 neurons and uses a rectified linear unit (ReLU) as an activation function.

[0086] Initialize Weights: Randomly initialize the weights of the model.

[0087] Set the optimizer: Select Adam optimizer and set the initial learning rate to 0.01.

[0088] Define loss function: Choose an appropriate loss function based on the task type (classification).

[0089] Configure Regularization: Apply Dropout regularization to reduce the risk of overfitting the model.

[0090] 3. Hyperparameter Tuning

[0091] Set the hyperparameter range: including but not limited to learning rate, Dropout ratio, number of hidden layer nodes, etc.

[0092] Perform 5-fold cross validation: For each hyperparameter combination, use 4 / 5 of the data to train the model in each round, and the remaining 1 / 5 of the data for validation. Record the validation performance of each round.

[0093] Evaluate and select the best hyperparameters: Compare the average validation performance under different hyperparameter combinations and select the best performing set of hyperparameters.

[0094] 4. Final model training

[0095] Retrain the model on the full training set using the selected best hyperparameter combination.

[0096] During the training process, the performance indicators of the model, such as accuracy or loss value, are monitored until the preset maximum number of iterations stepmax is reached or other stopping conditions are met.

[0097] 5. Model Evaluation

[0098] Evaluate the performance of the final model on an independent test set to examine how well the model generalizes to unseen data.

[0099] Record various performance indicators on the test set, such as accuracy, recall, F1 score, etc.

[0100] In some optional specific implementations, the prediction module 102 includes:

[0101] a target parameter acquisition unit, configured to acquire target parameters, wherein the target parameters include an albumin-bilirubin (ALBI) score, an alpha-fetoprotein (AFP) level, an age (Age), and a tumor size of the target subject;

[0102] A prediction unit is used to input the target parameter and the inflammatory biomarker into a neural network model to obtain an output of the neural network model.

[0103] Specifically, the neural network model uses a multi-layer perceptron (MLP) structure, such as Figure 2 As shown, it includes input neurons (I1-I7), hidden layers (H1-H3) and output neurons (O1, O2). The seven input variables are: albumin-bilirubin score (ALBI score), alpha-fetoprotein level (AFP level), age (Age), tumor size (Tumor_size), neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR) and lymphocyte-to-monocyte ratio (LMR).

[0104] In some optional specific implementations, the medical information prompting device further includes:

[0105] A plurality of sample data are obtained, wherein the sample data include candidate variables and corresponding survival times, wherein the candidate variables include at least the target parameters and the inflammatory biomarkers; the candidate parameters may also be other clinical variables and biomarkers, such as gender, medical history, treatment methods, etc., and other biomarkers are mainly indicators related to disease status, inflammation level, and immune response, such as C-reactive protein (CRP) in blood test results, etc.

[0106] A plurality of training subsets are extracted from the plurality of sample data using a sampling method with replacement (such as bootstrap);

[0107] For each of the training subsets, construct a survival decision tree;

[0108] At each node of the survival decision tree, selecting a preset number of the candidate variables for splitting;

[0109] Integrate the plurality of survival decision trees into a random survival forest model;

[0110] For each target candidate variable among the candidate variables, randomly shuffle the value of the target candidate variable in the out-of-bag sample data, and re-evaluate the prediction performance of the random survival forest model;

[0111] Calculating the difference in prediction performance of the random survival forest model before and after the disruption as the importance score of the target candidate variable;

[0112] According to the importance score, the target parameter and the inflammatory biomarker are screened out from the candidate variables.

[0113] In the embodiment of the present invention, the random survival forest model resamples multiple sample data by self-sampling to generate multiple training subsets, and constructs a survival decision tree for each training subset. A total of 500 decision trees (ntree=500) are constructed, and the minimum node size is 1 (nodesize=1) to control the depth of the tree. At each node split, 5 candidate variables are randomly selected (mtry=5). The importance of the variable is calculated by the increase in the out-of-bag (OOB) prediction error, and the OOB data is used to evaluate the model performance and prevent overfitting.

[0114] Variable Importance (VIMP), that is, importance scores are used to prioritize variables (such as Figure 3 and Figure 4The final model included variables with importance scores greater than 0.02, including ALBI score, AFP level, age, tumor size, and three inflammatory markers (NLR, PLR, LMR). Parameter tuning showed that as the number of trees (ntree) value increased to 300, the model error rate stabilized (as shown in Figure 2). Figure 5 shown).

[0115] The embodiment of the present invention uses a random survival forest model to determine the risk factors that affect the long-term survival of the target object, thereby improving the prediction accuracy of the neural network model. The random survival forest model is created using the 'randomForest' package in the R software.

[0116] In some optional specific implementations, the first building block in the medical information prompting device includes:

[0117] A selection unit is used to evaluate each of the candidate variables and select the variable that can maximize the survival rate difference as the split point. In the random survival forest model, the core of node splitting is to select a variable that can best distinguish the survival situation as the split point. The commonly used method is to evaluate the splitting effect of the candidate variable based on the log-rank statistic or the Cox score. The specific steps are as follows:

[0118] Candidate variable selection: At each node, a portion of candidate variables (mtry variables) are randomly selected. This step is to introduce randomness and increase the generalization ability of the model.

[0119] Evaluate each candidate variable: For each candidate variable, try different split points (usually different values ​​of the variable). For each split point of each candidate variable, the data will be divided into two subsets (left node and right node).

[0120] Calculate the split score:

[0121] The log-rank statistic is used to calculate the difference in survival rate after the split. The log-rank statistic is used to compare the significant difference between the survival curves in two subsets.

[0122] Specifically, the log-rank statistic measures the difference in the distribution of survival times in the two subsets. The higher the score, the better the split point is at distinguishing survival times.

[0123] Select the best variable and split point: For all candidate variables and corresponding split points, select the variable and split point with the highest score as the splitting criterion for the current node.

[0124] Summary: The random survival forest model selects the candidate variable that maximizes the difference in survival rate as the split point. This process is essentially looking for the variable and split threshold that most significantly separates the survival curves between the two subsets.

[0125] The recursive splitting unit is used to split the data in the training subset layer by layer, and each layer is divided according to different candidate variables and thresholds until a stopping condition is met.

[0126] The decision tree construction unit is used to divide the data into a multi-layer structure according to the splitting of different candidate variables to form a decision tree.

[0127] The specific process of stratification of the decision tree model includes:

[0128] 1. Variable evaluation: Each variable is evaluated and the variable that maximizes the difference in survival rate is selected as the split point.

[0129] 2. Recursive splitting: Split the data layer by layer, and each layer is divided according to different variables and thresholds until the stopping condition is met.

[0130] Specifically, starting from the root node, the data in the training subset is recursively divided into two subsets. That is, at each node, the variable that can maximize the survival rate difference and the corresponding split point are selected to split the data. The split data will form two child nodes: the left node and the right node. Continue to perform the same splitting process on each child node until the stopping condition is met, such as the number of samples in the node is less than a certain threshold (for example, 5), or the maximum tree depth reaches a preset value, or the current node has no significant split point (that is, the split scores of all candidate variables are very low).

[0131] 3. Generate a hierarchical structure: The model divides the data into a multi-layer structure based on the split of different variables to form a decision tree.

[0132] 4. Variable importance: Determine the importance of each variable by calculating the contribution of the candidate variable to the prediction result during the splitting process. Specifically, for each tree in the random forest, first use the out-of-bag data (OOB data) to calculate the prediction accuracy of the model and record it as Accuracy_original. Randomly shuffle a feature column (such as feature i), and then use the shuffled OOB data to recalculate the prediction accuracy of the model and record it as Accuracy_permuted. The importance of the feature is defined as the difference in accuracy before and after shuffling, that is: Importance(X i)=Accuracy_original-Accuracy_permuted. For each feature, repeat the above process and take the average value of all decision trees to get the final variable importance score, the formula is as follows: Importance(X i )=Accuracy_original-Accuracy_permuted.

[0133] Through recursive splitting, the decision tree gradually divides the data set into multiple subsets. Each split point corresponds to a different variable, which constitutes the hierarchical structure of the decision tree.

[0134] In the embodiment of the present invention, the decision tree model is divided into multiple levels according to the contribution of the variables to the results. The decision tree is constructed using the R packages 'rpart' and 'rpart.plot'.

[0135] In some optional specific implementations, the determining module 103 includes:

[0136] An output acquisition unit, used to obtain a probability value output by the neural network model;

[0137] A prompt information determination unit, used to determine the corresponding prompt information according to the magnitude relationship between the probability value and a preset probability threshold, wherein the prompt information includes high-risk prompt information and low-risk prompt information;

[0138] Among them, the preset probability threshold is the threshold corresponding to when the AUC of the neural network model is the highest.

[0139] Specifically, the preset probability threshold is also the optimal cutoff value. In a binary classification problem, the model will output a probability value, usually between 0 and 1. The optimal cutoff value is the dividing line used to convert the predicted probability into a binary classification. For example, if the output probability of the model (the probability of belonging to the high-risk group) is greater than or equal to this threshold, the prediction result is positive (such as the high-risk group); if it is less than this threshold, it is negative (such as the low-risk group). The optimal cutoff value is a parameter that divides patients into low-risk or high-risk groups based on the probability predicted by the model.

[0140] The ROC curve shows the sensitivity (True Positive Rate, TPR) of the model under different thresholds. The AUC (area under the curve) is the area of ​​the ROC curve. The closer the AUC value is to 1, the better the model performance. In the embodiment of the present invention, the threshold corresponding to the highest AUC value is selected as the optimal critical value.

[0141] In order to verify the importance of immune markers (i.e., inflammatory biomarkers) in predicting 5-year overall survival (OS), an embodiment of the present invention constructed a decision tree model that included seven independent risk factors identified by the random forest method. These risk factors include NLR (<1.50 vs. ≥1.50), LMR (<4.71 vs. ≥4.71), PLR level (<118 vs. ≥118), age (<65 years vs. ≥65 years), ALBI grade, AFP (<400ng / ml vs. ≥400ng / ml) and tumor size (<2cm vs. ≥2cm). The decision tree model stratifies the variables into multiple levels according to their contribution to the results ( Figure 6 As shown in Figure 2 ). AFP level was placed at the first level of the decision tree, showing the greatest contribution. As important immune markers, the contributions of LMR and NLR ranked second and third, respectively, suggesting that special attention should be paid to the levels of these immune markers, as they may be directly associated with the poor long-term prognosis of early HCC patients.

[0142] The training process of the neural network model provided in this embodiment includes:

[0143] 1. Get training data.

[0144] First, patients diagnosed with BCLC stage 0-A hepatocellular carcinoma (HCC) from September 2008 to September 2018 were screened. Patient exclusion criteria included: (1) suffering from other malignancies; (2) BCLC BD stage; (3) incomplete data or lack of follow-up; (4) under 18 years old or over 80 years old; (5) with severe heart, kidney or other primary diseases; (6) diagnosis of autoimmune liver disease or acquired immunodeficiency syndrome. On this basis, a total of 984 patients were selected for further evaluation and randomly divided into a training group and a validation group in a ratio of 2:1. The patient inclusion process was as follows: Figure 7 The primary outcome was mortality within 5 years or at the end of the 5-year follow-up period. Follow-up was mainly performed using hospital records and telephone interviews.

[0145] Then, the above 984 patients were used as the object samples, and the corresponding target parameters and inflammatory biomarkers were obtained as training and validation samples, and the mortality rate of the object samples within 5 years or at the end of the 5-year follow-up period was used as the label.

[0146] 2. Use the training data obtained in the previous step to train and verify the neural network model.

[0147] like Figure 8As shown in Figure 2, in the training group, the AUC values ​​of the neural network model at 1, 3, and 5 years were 0.87 (95% CI: 0.82-0.92), 0.82 (95% CI: 0.78-0.86), and 0.84 (95% CI: 0.81-0.87), respectively. Fig. 9 As shown in Figure 2, in the validation group, the AUC values ​​of the neural network model were 0.75 (95% CI: 0.64-0.85), 0.80 (95% CI: 0.73-0.86), and 0.77 (95% CI: 0.72-0.83), respectively. Fig.10 and Fig.11 As shown in Figure 1, the calibration curve shows a high consistency between the predicted survival probability and the actual survival probability. In the training and validation groups, the comparison of AUC and C index showed that the neural network model was also superior to the classic liver cancer models such as the Okuda staging system (Okuda), Japanese comprehensive staging (JIS), Chinese University Prognostic Index (CUPI) and Italian Liver Cancer Project (CLIP) staging (as shown in Table 1, DeLong test p < 0.05). Finally, as Fig.12 and Fig.13 As shown, the decision curve analysis (DCA) curve shows that the neural network model has the highest clinical net benefit and stronger clinical practicality compared with other classic liver cancer models.

[0148] Table 1. Comparison of artificial neural network (ANN) models with traditional models in terms of performance and discrimination ability.

[0149]

[0150] Using the optimal cutoff value of the neural network model, the patients were divided into low-risk group and high-risk group. The 5-year mortality rate of the high-risk group was 63.7% in the training group ( Fig.14 In the validation group, it was 61.5% ( Fig.15 The neural network model accurately distinguished high-risk patients in both groups. Subgroup analysis showed that the neural network model was effective in predicting different etiologies (with or without HBV infection) ( Fig.16 and Fig.17 As shown), liver function status (Child-Pugh A or Child-Pugh B, C) ( Fig.18 and Fig.19 As shown) and AFP levels (AFP < 400 ng / ml or AFP ≥ 400 ng / ml) ( Fig. 20 and Fig.21 The 5-year prognosis of early HCC patients (shown in Figure 2) remained stable and provided satisfactory discriminatory ability (p<0.001). P value was obtained by log-rank test.

[0151] Kaplan-Meier curve analysis showed the effectiveness of the neural network model level as a predictor of 5-year mortality in patients with BCLC stage 0-A HCC, including training and validation sets, different etiologies, liver function, and AFP levels.

[0152] In summary, the AUC value of the ANN model for predicting the 1-5-year survival rate of HCC patients was significantly greater than that of the RSF model and traditional scoring systems such as Okuda, JIS, CUPI, and CLIP. The artificial neural network model based on the cutoff value showed excellent discrimination ability in patients with different liver functions and etiologies. According to the cutoff value, the hazard ratio (HR) of high-risk to low-risk patients in the training group was 6.67 (95% CI: 5.01-8.88), while the hazard ratio in the validation group was 4.75 (95% CI: 3.14-7.19). When excluding patients with a good 5-year prognosis, the low cutoff value of the ANN model obtained a negative predictive value (NPV) of 92.71% and 83.06% in the training group and validation group, respectively.

[0153] In addition, immune-inflammation related markers were compared between the survival group and the death group of early HCC patients. The results showed that the NLR and PLR in the death group were higher than those in the survival group, and the LMR was lower (p<0.05) ( Fig. 22 , Fig.23 and Fig.24 shown).

[0154] The medical information prompting device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0155] In this embodiment, a medical information prompting method is provided, which can be used in various computer devices, and the computer device can be a medical device. The method is implemented by the medical information prompting device of the above embodiment. Fig.25 is a flow chart of a medical information prompting method according to an embodiment of the present invention. Fig.25 As shown, the process includes the following steps:

[0156] Step S2501, obtaining inflammatory biomarkers of the target object, wherein the inflammatory biomarkers include one or more of a neutrophil-to-lymphocyte ratio (NLR), a platelet-to-lymphocyte ratio (PLR), a lymphocyte-to-monocyte ratio (LMR), and an international normalized ratio (INR).

[0157] Step S2502, inputting the inflammatory biomarker into a neural network model (ie, an artificial neural network (ANN) model) to obtain an output of the neural network model.

[0158] Specifically, the above neural network model uses a multi-layer perceptron (MLP) structure, including input neurons (I1-I7), hidden layers (H1-H3) and output neurons (O1, O2). The artificial neural network (ANN) model is built using the Python package 'PyTorch'. The model performance can be optimized by adding three hidden layers through multiple tests.

[0159] Step S2503, determining corresponding prompt information according to the output of the neural network model. The target object may be a BCLC0-A liver cancer patient, where BCLC is the Barcelona Clinical Liver Cancer Staging System. The output of the neural network model may be a survival rate, such as a one-year survival rate, a three-year survival rate, and a five-year survival rate.

[0160] Step S2504: display the prompt information.

[0161] The embodiment of the present invention also provides a computer device having the above Figure 1 The medical information prompting device shown.

[0162] See also Fig.26 , Fig.26 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Fig.26 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Fig.26 A processor 10 is taken as an example.

[0163] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0164] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0165] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0166] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0167] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Fig.26 The example of connecting through bus is taken in the following.

[0168] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0169] The computer device also includes a communication interface, which is used for the computer device to communicate with other devices or a communication network.

[0170] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0171] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0172] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A medical information prompting device, characterized in that: The device comprises: a parameter acquisition module, configured to acquire inflammation biomarkers of a target object, wherein the inflammation biomarkers include one or more of a neutrophil to lymphocyte ratio, a platelet to lymphocyte ratio, a lymphocyte to monocyte ratio, and an international normalized ratio; A prediction module, used for inputting the inflammatory biomarker into a neural network model to obtain an output of the neural network model; A determination module, used to determine corresponding prompt information according to the output of the neural network model; The display module is used to display the prompt information.

2. The device according to claim 1, characterized in that The neural network model includes a layer with 5 neurons and uses a rectified linear unit as an activation function; The neural network model uses the Adam optimizer to update weights during training, uses full batch training, and the training iterations are controlled by the maximum number of iterations parameter.

3. The device according to claim 1, characterized in that The prediction module comprises: a target parameter acquisition unit, configured to acquire target parameters, wherein the target parameters include the ALBI score, AFP level, age, and tumor size of the target subject; A prediction unit is used to input the target parameter and the inflammatory biomarker into a neural network model to obtain an output of the neural network model.

4. The device according to claim 3, characterized in that Also includes: A sample data acquisition module, used to acquire a plurality of sample data, wherein the sample data includes candidate variables and corresponding survival times, wherein the candidate variables include at least the target parameter and the inflammatory biomarker; An extraction module, used for extracting a plurality of training subsets from the plurality of sample data by adopting a sampling method with replacement; A first construction module is used to construct a survival decision tree for each of the training subsets, and select a preset number of candidate variables for splitting at each node of the survival decision tree; A second building module is used to integrate the plurality of survival decision trees into a random survival forest model; A performance evaluation module, for randomly disrupting the value of each target candidate variable in the candidate variables in the out-of-bag sample data, and re-evaluating the prediction performance of the random survival forest model; A scoring calculation module, used to calculate the difference in prediction performance of the random survival forest model before and after the shuffle as the importance score of the target candidate variable; A screening module is used to screen out the target parameter and the inflammatory biomarker from the candidate variables according to the importance score.

5. The device according to claim 4, characterized in that The first building block comprises: A selection unit, configured to evaluate each of the candidate variables and select a variable that can maximize the survival rate difference as a split point; A recursive splitting unit, used for splitting the data in the training subset layer by layer, wherein each layer is divided according to different candidate variables and thresholds until a stopping condition is met; The decision tree construction unit is used to divide the data into a multi-layer structure according to the splitting of different candidate variables to form a decision tree.

6. The device according to claim 1, characterized in that The determination module comprises: An output acquisition unit, used to obtain a probability value output by the neural network model; A prompt information determination unit, used to determine the corresponding prompt information according to the magnitude relationship between the probability value and a preset probability threshold, wherein the prompt information includes high-risk prompt information and low-risk prompt information; Among them, the preset probability threshold is the threshold corresponding to when the AUC of the neural network model is the highest.

7. A medical information prompting method, characterized in that: The method comprises: Obtaining inflammation biomarkers of the target subject, wherein the inflammation biomarkers include one or more of a neutrophil to lymphocyte ratio, a platelet to lymphocyte ratio, a lymphocyte to monocyte ratio, and an international normalized ratio; Inputting the inflammatory biomarker into a neural network model to obtain an output of the neural network model; Determining corresponding prompt information according to the output of the neural network model; The prompt information is displayed.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the medical information prompting method according to claim 7 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the medical information prompting method described in claim 7.

10. A computer program product, characterized in that It includes computer instructions, and the computer instructions are used to make a computer execute the medical information prompting method described in claim 7.

Images