Assessment method, device and equipment for severed finger replantation postoperative necrosis risk

Abstract

The invention discloses a severed finger replantation postoperative necrosis risk assessment method, device and equipment and a readable storage medium, and relates to the technical field of artificial intelligence. Comprising the following steps: firstly, acquiring severed finger replantation related data of a sample patient; screening out target data of which the influence degree on the severed finger replantation postoperative necrosis risk assessment result is greater than a preset value from the severed finger replantation related data by using Lasso regression and a Boruta algorithm; then training candidate models based on the target data, and screening out a target model from the candidate models according to the prediction performance of each candidate model; and finally, obtaining target data of a target patient, and inputting the target data of the target patient into the target model to obtain an evaluation result of the severed finger replantation postoperative necrosis risk of the target patient. And the evaluation accuracy of the severed finger replantation postoperative necrosis risk is obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a method and device for evaluating the risk of necrosis after finger replantation, and a readable storage medium. BACKGROUND

[0002] The development of manufacturing industry cannot be separated from numerous manual laborers. With the rapid development of mechanization and industrialization, finger amputation as a common hand injury has seen a growing number of patients in recent years. Finger replantation is the main way to treat completely or incompletely amputated fingers at the present stage. With the continuous improvement of microsurgery technology and the updating and improvement of equipment, the survival rate of finger replantation has reached more than 90%, and the indications for surgery have gradually expanded. Various complex and different types of amputated fingers have survived. However, necrosis still occurs in some patients after replantation, which will have a great impact on the psychological health, appearance and function of the patient's hand. Therefore, it is also important to evaluate the risk of necrosis after finger replantation.

[0003] However, there are many factors affecting necrosis after finger replantation. At present, more research focuses on the repair of amputated fingers during surgery, and there is a lack of evaluation of the injury of amputated fingers of patients before replantation, strict control of replantation indications, and detection and inspection data of amputated fingers during the whole process of finger replantation. Comprehensive analysis results in inaccurate evaluation of the risk of necrosis after finger replantation. Therefore, there is an urgent need for a method that can accurately predict the risk of necrosis after finger replantation. SUMMARY

[0004] The present application aims to provide a method and device for evaluating the risk of necrosis after finger replantation, which comprehensively considers the basic characteristic data of patients, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data and composite inflammation indicators, and uses multiple candidate models for evaluation, significantly improving the accuracy of the evaluation of the risk of necrosis after finger replantation from both data and model aspects.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions: In a first aspect, the present application provides a method for evaluating the risk of necrosis after finger replantation, which comprises: Obtaining finger replantation related data of a sample patient; the finger replantation related data includes basic characteristic data, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data and composite inflammation indicators of the sample patient; Using Lasso regression and Boruta algorithm to screen target data from the finger replantation related data, the influence degree of the target data on the evaluation result of the risk of necrosis after finger replantation being greater than a preset value; training the candidate model based on the target data, and selecting a target model from the candidate model according to a prediction performance of each candidate model; obtaining target data of a target patient, and inputting the target data of the target patient into the target model to obtain an evaluation result of a necrosis risk of the target patient after finger replantation.

[0006] In some embodiments, the basic feature data of the sample patient includes age, gender, smoking history, hypertension, and diabetes; the preoperative risk factor data includes injured finger, injury plane, injury mechanism, number of amputations, method of preserving amputated fingers, whether combined with other injuries, degree of amputation, bone and joint injury, condition of digital artery injury, nerve injury, tendon injury plane, and dorsal skin injury of the nail bed; the intraoperative risk factor data includes ischemia time, operation time, doctor's seniority, vein graft, number of arterial anastomosis, number of venous anastomosis, ratio of anastomotic artery to vein, and whether a skin flap covers the wound; the postoperative risk factor data includes whether a vascular crisis occurs, hemoglobin, platelet count, neutrophil count, lymphocyte count, mononuclear cell count, neutrophil percentage, albumin, and D-dimer; the composite inflammation indicators include neutrophil-lymphocyte ratio (NLR), mononuclear cell-lymphocyte ratio (MLR), platelet-lymphocyte ratio (PLR), neutrophil-platelet ratio (NPR), systemic inflammatory response index (SIRI), systemic immune inflammation index (SII), systemic inflammatory syndrome index (AISI), prognostic nutrition index (PNI), and neutrophil percentage to albumin ratio (NPAR).

[0007] In some embodiments, the target data having an influence degree greater than a preset value on the evaluation result of the necrosis risk after finger replantation is screened from the finger replantation related data by using Lasso regression and Boruta algorithm, including: the first data having an influence degree greater than a preset value on the evaluation result of the necrosis risk after finger replantation is screened from the finger replantation related data by using Lasso regression; the second data having an influence degree greater than a preset value on the evaluation result of the necrosis risk after finger replantation is screened from the finger replantation related data by using Boruta algorithm; the intersection of the first data and the second data is obtained to obtain the target data.

[0008] In some embodiments, the candidate models include: Logistic regression (LR), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), Neural Network, Random Forest (RF), eXtreme Gradient Boosting (Xgboost), K Nearest Neighbor Machine (KNN), AdaBoost, Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost).

[0009] In some embodiments, the candidate models are trained based on the target data, and a target model is selected from the candidate models according to the prediction performance of each candidate model, including: The candidate models are trained based on the target data, and evaluation indexes of each candidate model are obtained; the evaluation indexes include accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, area under curve, calibration curve, clinical decision curve, and confusion matrix; The prediction performance of each candidate model is evaluated based on the evaluation indexes; The target model is selected from the candidate models according to the prediction performance of each candidate model.

[0010] In some embodiments, the target model is selected from the candidate models according to the prediction performance of each candidate model, including: The best model is selected from the candidate models according to the prediction performance of each candidate model; The SHAP method is used to perform local interpretation and global interpretation on the best model, and the target model is obtained.

[0011] In a second aspect, the present application further provides an evaluation device for post-amputation necrosis risk, which comprises: A data acquisition module is configured to acquire amputation-related data of a sample patient; the amputation-related data includes basic characteristic data, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data, and composite inflammatory indicators of the sample patient. A data screening module is configured to screen target data with an influence degree greater than a preset value on the evaluation result of post-amputation necrosis risk from the amputation-related data by using Lasso regression and Boruta algorithm. The model screening module is configured to train candidate models based on the target data, and screen a target model from the candidate models according to the prediction performance of each candidate model. The risk assessment module is configured to obtain target data of the target patient, and input the target data of the target patient into the target model to obtain an evaluation result of the necrosis risk of the target patient after the finger replantation.

[0012] In a third aspect, the present application further provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method for evaluating the necrosis risk after the finger replantation according to the first aspect when executing the computer program.

[0013] In a fourth aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the method for evaluating the necrosis risk after the finger replantation according to the first aspect.

[0014] In a fifth aspect, the present application further provides a computer program product, which comprises a computer program, and the computer program is executable on a processor to implement the method for evaluating the necrosis risk after the finger replantation according to the first aspect.

[0015] The method for evaluating the necrosis risk after the finger replantation provided by the present application comprises the following steps: first, obtaining finger replantation related data of sample patients; the finger replantation related data comprises basic characteristic data, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data, and a composite inflammation index of the sample patients; then, using Lasso regression and Boruta algorithm to screen target data with an influence degree greater than a preset value on the evaluation result of the necrosis risk after the finger replantation from the finger replantation related data; then, training candidate models based on the target data, and screening a target model from the candidate models according to the prediction performance of each candidate model; finally, obtaining target data of a target patient, and inputting the target data of the target patient into the target model to obtain an evaluation result of the necrosis risk of the target patient after the finger replantation. The basic characteristic data, the preoperative risk factor data, the intraoperative risk factor data, the postoperative risk factor data, and the composite inflammation index of the patient are comprehensively considered, and a plurality of candidate models are used for evaluation, so that the evaluation accuracy of the necrosis risk after the finger replantation is significantly improved from the aspects of data and model.

[0016] The above description is only a summary of the technical scheme of the present application, in order to more clearly understand the technical means of the present application, and the content of the specification can be implemented. The following describes the preferred embodiments of the present application in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1A flowchart of a method for evaluating the risk of post-replantation necrosis of a severed finger according to an embodiment of the present application; Figure 2 A flowchart of another method for evaluating the risk of post-replantation necrosis of a severed finger according to an embodiment of the present application; Figure 3 A structural diagram of an apparatus for evaluating the risk of post-replantation necrosis of a severed finger according to an embodiment of the present application; Figure 4 A structural diagram of another apparatus for evaluating the risk of post-replantation necrosis of a severed finger according to an embodiment of the present application; Figure 5 An electronic device structure diagram is provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] The technical solutions of the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application. It should be noted that the description of "one embodiment", "embodiment", "example embodiment" and the like in the specification means that the described embodiment can include specific features, structures or characteristics, but not every embodiment must include these specific features, structures or characteristics. In addition, such expressions do not mean the same embodiment. Further, when a specific feature, structure or characteristic is described in combination with an embodiment, it is indicated that such a feature, structure or characteristic is combined with other embodiments within the knowledge of those skilled in the art, whether or not it is explicitly described.

[0019] In addition, the technical features involved in different embodiments of the present application described below can be combined with each other as long as there is no conflict.

[0020] In some embodiments, as shown in Figure 1 A method for evaluating the risk of post-replantation necrosis of a severed finger is provided, and the specific method includes: S101, acquiring the data related to the replantation of the severed finger of the sample patient.

[0021] The data related to finger replantation include basic characteristic data of sample patients, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data and composite inflammation indicators. The basic characteristic data of the sample patients includes age, gender, smoking history, hypertension, diabetes; the preoperative risk factor data includes injured finger, injury plane, injury mechanism, number of dislocations, finger preservation method, whether combined with other injuries, degree of dislocation, bone and joint injury, finger artery injury, nerve injury, tendon injury plane, nail bed dorsal skin injury; the intraoperative risk factor data includes ischemia time, operation time, doctor's seniority, vein graft, number of arterial anastomosis, number of venous anastomosis, anastomotic artery-vein ratio, whether the flap covers the wound; the postoperative risk factor data includes whether the vascular crisis occurs, hemoglobin, platelet count, neutrophil count, lymphocyte count, monocyte count, neutrophil percentage, albumin, D-dimer; the composite inflammation indicators include neutrophil-lymphocyte ratio (NLR), monocyte-lymphocyte ratio (MLR), platelet-lymphocyte ratio (PLR), neutrophil-platelet ratio (NPR), systemic inflammatory response index (SIRI), systemic immune inflammation index (SII), systemic inflammatory syndrome index (AISI), prognostic nutrition index (PNI), neutrophil percentage-albumin ratio (NPAR). The sample patients are patients who have undergone finger replantation at a historical time.

[0022] Specifically, the age, gender, smoking history, hypertension, diabetes, injured finger, injury plane, injury mechanism, number of dislocations, finger preservation method, whether combined with other injuries, degree of dislocation, bone and joint injury, finger artery injury, nerve injury, tendon injury plane, nail bed dorsal skin injury, ischemia time, operation time, doctor's seniority, vein graft, number of arterial anastomosis, number of venous anastomosis, anastomotic artery-vein ratio, whether the flap covers the wound, whether the vascular crisis occurs, hemoglobin, platelet count, neutrophil count, lymphocyte count, monocyte count, neutrophil percentage, albumin, D-dimer, neutrophil-lymphocyte ratio (NLR), monocyte-lymphocyte ratio (MLR), platelet-lymphocyte ratio (PLR), neutrophil-platelet ratio (NPR), systemic inflammatory response index (SIRI), systemic immune inflammation index (SII), systemic inflammatory syndrome index (AISI), prognostic nutrition index (PNI), neutrophil percentage-albumin ratio (NPAR) and other data of the sample patients can be obtained from the electronic medical record system of the hospital.

[0023] In S102, Lasso regression and Boruta algorithm are used to screen target data with an influence degree greater than a preset value on the necrosis risk evaluation result after finger replantation from the data related to finger replantation.

[0024] Among them, Lasso regression (Least Absolute Shrinkage and Selection Operator Regression) is a linear regression model that introduces L1 regularization (i.e., Lasso penalty term) to compress the coefficients in the model, making some coefficients shrink to zero, thereby achieving feature selection and model sparsity. Lasso regression is mainly used to deal with the case of too many variables and small sample size, which can effectively prevent overfitting and solve the problem of multicollinearity. Boruta algorithm is a feature selection method based on random forest, aiming to select all features that are significantly related to the dependent variable, rather than only optimizing the loss function of a specific model. Its core idea is to compare the importance of original features with randomly generated "shadow features", combined with statistical tests and iterative strategies, to ensure the stability and comprehensiveness of the screening results.

[0025] Optionally, Lasso regression can be used to screen first data from the data related to finger replantation, which has an impact on the evaluation result of the risk of post-finger replantation necrosis greater than a preset value; Boruta algorithm can be used to screen second data from the data related to finger replantation, which has an impact on the evaluation result of the risk of post-finger replantation necrosis greater than a preset value; and the intersection of the first data and the second data is obtained to obtain target data.

[0026] For example, Lasso regression, as a compression estimation method, completes variable selection and complexity adjustment by formulating an optimization objective function containing a penalty term. The present application uses LASSO regression to identify characteristic factors, identifies characteristic factors such as age (seniority), neutrophil count (NEU), vascular crisis (vascular_crisis), systemic immune inflammation index (SII), platelets (PLT), arteriovenous ratio (arteriovenous_ratio), number of injuries (injury_number), percentage of neutrophils (NEU%), D-dimer (D-Dimer), and performs cross-validation. Boruta is a feature selection method based on random forest, which compares the importance of original features with randomly generated "shadow features", aiming to screen features significantly related to the dependent variable, rather than only optimizing the loss function of a specific model. Boruta screens variables related to finger replantation necrosis. By comparing and analyzing the screening results of LASSO regression and Boruta algorithm, we decide to use a common subset of characteristic variables selected by the two methods, that is, age (seniority), neutrophil count (NEU), vascular crisis (vascular_crisis), systemic immune inflammation index (SII), platelets (PLT), arteriovenous ratio (arteriovenous_ratio), number of injuries (injury_number), percentage of neutrophils (NEU%), D-dimer (D-Dimer), and finally use these nine data as target data.

[0027] S103, training the candidate model based on the target data, and screening the target model from the candidate model according to the prediction performance of each candidate model.

[0028] The candidate models include: Logistic regression (LR), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), Neural Network, Random Forest (RF), eXtreme Gradient Boosting (Xgboost), K Nearest Neighbor Machine (KNN), AdaBoost, Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost). Logistic regression (LR) uses a linear function wT x + b to map to 0-1 through Sigmoid, and maximizes the log-likelihood; it is simple to implement, fast to train, and the result is interpretable; it naturally outputs probability; it is excellent for linearly separable problems but can only fit linear boundaries; it requires manual feature engineering; it is sensitive to outliers and multicollinearity. Support Vector Machine (SVM) can maximize the "margin" between positive and negative classes; when non-linear, it implicitly maps samples to high dimensions through kernel functions; it is still robust in small sample high dimensions; kernel trick is flexible; it is not sensitive to outliers (soft margin) but slow to train O(n 2 ~n 3The advantages of Gradient Boosting Machine (GBM) are: 1) Difficult kernel and hyperparameter tuning; probability output requires additional platscaling. 2) It adds weak learners (usually CART) sequentially, fitting the negative gradient (residual) of the current loss function at each step; it can handle arbitrary differentiable losses; it has high accuracy; it can characterize complex nonlinear relationships; however, training is slow; it has many hyperparameters; it is prone to overfitting; and it is sensitive to outliers and noise. Neural Network (NN) uses multiple learnable nonlinear transformations (activation function + backpropagation) to approximate the target function end-to-end. Its advantages are: universal approximator; state-of-the-art performance with massive data and high computing power; it supports transfer / multimodal learning but requires a large amount of data, computing power, and parameter tuning; it is a black box and difficult to interpret; it is vulnerable to noisy labels and adversarial examples. Random Forest (RF) works by using Bagging (sampling with replacement + random sub-features) to generate a large number of decision trees in parallel, with voting / averaging output. Its advantages include almost no hyperparameter tuning required; resistance to overfitting; ability to provide feature importance; and support for missing values ​​and categorical variables. However, its accuracy is lower than Boosting when the boundary conditions are complex; the model size is large, and prediction latency is high. Extreme Gradient Boosting (XGBoost) adds second-order Taylor expansion, regularization terms, column sampling, and parallel approximate quantiles to GBM. Its advantages include being a powerful tool for Kaggle competitions; automatic handling of missing values; regularization to prevent overfitting; and CPU / GPU parallelism. However, it has many hyperparameters; categorical variables require one-hot encoding or coding; and its memory usage is higher than LightGBM. The K-Nearest Neighbor Machine (KNN) operates on the principle of lazy learning, calculating the distance to the k nearest neighbors in the training set during prediction and then voting / averaging the results. Its advantages include: no training required; easy implementation; incremental learning; and friendliness to nonlinear boundaries. However, it suffers from slow prediction speed (O(n), the curse of high dimensionality), requires feature normalization, and is prone to bias in imbalanced data. Adaptive Boosting (AdaBoost) sequentially adjusts sample weights, increasing the weight of misclassified samples in each round. The weak learner obtains coefficients based on weighted errors, and finally, weighted voting is performed. Its algorithm is simple; it significantly improves weak classifiers; it is less prone to overfitting (weak learners are shallow), but it is extremely sensitive to noise and outliers. For multi-class classification, extensions (such as AdaBoost.MH) are needed. The Light Gradient Boosting Machine (LightGBM) is an improvement on GBDT by Microsoft, using histogram binning, leaf-wise growth, GOSS gradient sampling, and EFB feature packing to achieve efficient training. Its advantages include fast training / prediction, low memory usage, support for categorical variables, and accuracy comparable to XGBoost. However, leaf-wise methods are prone to overfitting (requiring adjustment of max_depth / num_leaves), and its advantages are not obvious on small datasets.The principle of the category gradient boosting (CatBoost) is: a GBDT variant of Yandex, which processes category features with Ordered Target Statistics and adds Ordered Boosting to reduce overfitting; the advantages are that category variables are directly fed; the default parameters are good to use; it is robust to high base numbers and missing values; GPU acceleration; but the training speed is generally slower than LightGBM; the memory occupation is high; the hyperparameter grid is large. Therefore, under the premise that different candidate models have different characteristics, the target model most suitable for the present application needs to be selected from the candidate models.

[0029] Optionally, the method for determining the target model can be based on training the candidate models on the target data and obtaining evaluation indicators of each candidate training model; evaluating the prediction performance of each candidate model based on the evaluation indicators; and selecting the target model from the candidate models according to the prediction performance of each candidate model.

[0030] Among them, the evaluation indicators include accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, area under the curve, calibration curve, calibration curve, clinical decision curve and confusion matrix; according to the prediction performance of each candidate model, the target model is selected from the candidate models, including: according to the prediction performance of each candidate model, the best model is selected from the candidate models; the SHAP method is used for local interpretation and global interpretation of the best model to obtain the target model.

[0031] For example, in order to obtain the best prediction performance, the present application constructs 10 models, including: Logistic regression (LR), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), Neural Network, Random Forest (RF), eXtreme Gradient Boosting (Xgboost), K Nearest Neighbor Machine (KNN), AdaBoost, Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost). The performance of these models is evaluated using various indicators, including accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, area under the curve, calibration curve, calibration curve, clinical decision curve, and confusion matrix, to select the best model. The final results show that the RF model performs excellent prediction performance in the training set, with an area under the curve of 1.00 (95% CI: 1.000-1.000), indicating high accuracy of the prediction. In contrast, the area under the curve values of the remaining nine models are as follows: 0.961, (95% CI: 0.946-0.976) for LR, 0.959, (95% CI: 0.946-0.973) for SVM, 0.966, (95% CI: 0.952-0.980) for Neural Network, 0.960, (95% CI: 0.946-0.975) for Xgboost, 0.994, (95% CI: 0.991-0.997) for KNN, 0.920, (95% CI: 0.896-0.945) for AdaBoost, 0.999, (95% CI: 0.998-1.000) for LightGBM, 0.982, (95% CI: 0.974-0.991) for CatBoost. The RF model threshold is 0.5, with an accuracy of 1, sensitivity of 1, specificity of 1, precision of 1, and F1 score of 1. The LightGBM model threshold is 0.481, with an accuracy of 0.999, sensitivity of 1, specificity of 0.999, precision of 0.994, and F1 score of 0.997. The RF and LightGBM models perform too well, which may be overfitting. Therefore, the GBM model is the best model in the training set data.

[0032] In the validation set, the area under the curve (AUC) for both the GBM and CatBoost models was 0.978 (95% CI: 0.961–0.995). Both GBM and CatBoost demonstrated excellent predictive performance. However, in terms of accuracy (0.939 vs 0.922), sensitivity (0.944 vs 0.931), specificity (0.938 vs 0.920), precision (0.731 vs 0.667), and F1 score (0.824 vs 0.784), the GBM model outperformed the CatBoost model. In both the training and validation sets, the calibration curves showed a satisfactory consistency between the predicted and observed risks of the GBM model. Decision curve analysis indicated that the GBM model has high clinical value. These results strongly indicate that the GBM model significantly outperforms other models across various performance parameters, thus making it the optimal model. To further confirm GBM as the best model, data on finger replantation from other patients in other hospitals were obtained for validation. The results showed that the area under the curve (AUC) for the GBM model was 0.992 (95% CI: 0.983-1.000), demonstrating good predictive performance. Furthermore, the GBM model also outperformed other models in accuracy (0.98), sensitivity (0.958), specificity (0.981), precision (0.821), and F1 score (0.885). Decision curve analysis showed that the GBM model exhibited high clinical value across the entire threshold range (0-0.9), therefore, the GBM model was selected as the optimal model.

[0033] The SHAP algorithm is then used to interpret the model output by calculating the contribution of each variable to the predicted value, providing both a global interpretation of the model at the feature level and a local interpretation at the individual level.

[0034] Shapley Additive Explanations (SHAP) is a model interpretation method based on Shapley values ​​in game theory, aiming to quantify the contribution of each feature to the prediction results of a machine learning model. Its core principle is to fairly distribute the "merit" or "responsibility" of the model output to each input feature through mathematical allocation principles. This applies whether it's a global interpretation analyzing the importance of each feature in the overall model (e.g., the average impact of a feature on prediction) or a local interpretation of the prediction results of a single sample (e.g., the specific contribution value of each feature in a particular prediction).

[0035] Exemplarily, the contribution of features to the model can be evaluated by the average absolute SHAP value of the influence of different features on the model output, and finally the key features in the model can be identified in descending order of the average value: D-dimer (D-Dimer), neutrophil count (NEU), platelet (PLT), age (seniority), neutrophil percentage (NEU%), systemic immune inflammation index (SII), number of injuries (injury_number), arteriovenous ratio (arteriovenous_ratio), vascular crisis (vascular_crisis).

[0036] Local interpretation can be achieved by calculating and displaying the contribution of each feature to the prediction result of a single sample, and the contribution of each feature to the final prediction result can be intuitively understood. When the seniority of the doctor is high, the positive contribution value to the result is +0.0212, and when NEU=4.07, NEU%=62.7%, PLT=253, SII=507, D-Dimer=0.34, the negative contribution values to the result are -0.061, -0.0277, -0.0208, -0.0166, and -0.0154, respectively. Finally, the target model is obtained.

[0037] In S104, the target data of the target patient is obtained, and the target data of the target patient is input into the target model to obtain an evaluation result of the risk of post-replantation necrosis of the target patient.

[0038] Specifically, when it is necessary to evaluate the risk of post-replantation necrosis of the target patient, the age (seniority), neutrophil count (NEU), vascular crisis (vascular_crisis), systemic immune inflammation index (SII), platelet (PLT), arteriovenous ratio (arteriovenous_ratio), number of injuries (injury_number), neutrophil percentage (NEU%), and D-dimer (D-Dimer) of the target patient can be directly obtained, and the data is input into the target model. The target model outputs the evaluation result of the risk of post-replantation necrosis of the target patient.

[0039] Optionally, in order to facilitate use, the target model can also be applied to a WEB program, and by inputting the target data of the replanted patient, the application program can automatically calculate the risk of necrosis of the replanted patient.

[0040] The method for evaluating the risk of post-replantation necrosis in the above embodiment comprises the following steps: acquiring the data related to the finger replantation of a sample patient; the data related to the finger replantation of the sample patient comprises basic characteristic data of the sample patient, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data and a composite inflammation index; using Lasso regression and Boruta algorithm to screen target data with an influence degree greater than a preset value on the evaluation result of the risk of post-replantation necrosis from the data related to the finger replantation; training a candidate model based on the target data, and screening a target model from the candidate model according to the prediction performance of each candidate model; and inputting the target data of a target patient into the target model to obtain the evaluation result of the risk of post-replantation necrosis of the target patient. The basic characteristic data of the patient, the preoperative risk factor data, the intraoperative risk factor data, the postoperative risk factor data and the composite inflammation index are comprehensively considered, and a plurality of candidate models are used for evaluation, thereby significantly improving the evaluation accuracy of the risk of post-replantation necrosis from the aspects of data and model.

[0041] In order to more comprehensively display the present scheme, the present embodiment gives an optional way of a method for evaluating the risk of post-replantation necrosis, as shown in Figure 2 S201, acquiring the data related to the finger replantation of a sample patient.

[0042] The data related to the finger replantation of the sample patient comprises basic characteristic data of the sample patient, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data and a composite inflammation index; the basic characteristic data of the sample patient comprises age, gender, smoking history, hypertension and diabetes; the preoperative risk factor data comprises injured finger, injury plane, injury mechanism, number of amputations, amputated finger preservation method, whether combined with other injuries, degree of amputation, bone and joint injury, finger artery injury, nerve injury, tendon injury plane and dorsal skin injury of the nail bed; the intraoperative risk factor data comprises ischemia time, operation time, doctor's seniority, vein transplantation, number of arterial anastomosis, number of venous anastomosis, anastomotic arteriovenous ratio and whether the flap covers the wound; the postoperative risk factor data comprises whether a vascular crisis occurs, hemoglobin, platelet count, neutrophil count, lymphocyte count, mononuclear cell count, neutrophil percentage, albumin and D-dimer; and the composite inflammation index comprises neutrophil-lymphocyte ratio (NLR), mononuclear cell-lymphocyte ratio (MLR), platelet-lymphocyte ratio (PLR), neutrophil-platelet ratio (NPR), systemic inflammatory response index (SIRI), systemic immune inflammation index (SII), systemic inflammation comprehensive index (AISI), prognosis nutrition index (PNI) and neutrophil percentage to albumin ratio (NPAR).

[0043] ​S202, screening first data from the data related to the finger replantation surgery by using Lasso regression, the first data having an influence degree on the evaluation result of the risk of post-finger replantation necrosis greater than a preset value.

[0044] S203, screening second data from the data related to the finger replantation surgery by using Boruta algorithm, the second data having an influence degree on the evaluation result of the risk of post-finger replantation necrosis greater than a preset value.

[0045] S204, obtaining target data by taking an intersection of the first data and the second data.

[0046] S205, training candidate models based on the target data, and obtaining evaluation indexes of each candidate training model.

[0047] The candidate models include: Logistic regression (LR), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), Neural Network, Random Forest (RF), eXtreme Gradient Boosting (Xgboost), K Nearest Neighbor Machine (KNN), AdaBoost, Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost). The evaluation indexes include accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, area under curve, calibration curve, calibration curve, clinical decision curve, and confusion matrix.

[0048] S206, evaluating the prediction performance of each candidate model based on the evaluation indexes.

[0049] S207, selecting the best model from the candidate models according to the prediction performance of each candidate model.

[0050] S208, performing local interpretation and global interpretation on the best model by using SHAP method, and obtaining a target model.

[0051] S209, obtaining target data of a target patient, and inputting the target data of the target patient into the target model to obtain an evaluation result of the risk of post-finger replantation necrosis of the target patient.

[0052] The specific process of S201-S209 can be referred to the description of the above method embodiments, and the implementation principle and technical effects are similar, which will not be repeated here.

[0053] Based on the same inventive concept, the embodiments of the present application also provide a device for evaluating the risk of necrosis after finger replantation. The implementation scheme of the device for solving the problem is similar to the implementation scheme described in the above method, so the specific limitations in one or more device embodiments for evaluating the risk of necrosis after finger replantation provided below can be referred to the limitations of the method for evaluating the risk of necrosis after finger replantation described above, which will not be repeated here.

[0054] In one embodiment, as shown in Figure 3 a device for evaluating the risk of necrosis after finger replantation is provided, the device comprises: a data acquisition module 30, configured to acquire finger replantation related data of a sample patient; the finger replantation related data comprises basic feature data, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data and composite inflammation indicators of the sample patient; a data screening module 31, configured to screen target data with an influence degree on the evaluation result of the risk of necrosis after finger replantation greater than a preset value from the finger replantation related data by using Lasso regression and Boruta algorithm; a model screening module 32, configured to train candidate models based on the target data, and screen a target model from the candidate models according to the prediction performance of each candidate model; a risk evaluation module 33, configured to acquire target data of a target patient, and input the target data of the target patient into the target model to obtain the evaluation result of the risk of necrosis after finger replantation of the target patient.

[0055] In another embodiment, the basic feature data of the sample patient includes: age, gender, smoking history, hypertension, diabetes; preoperative risk factor data includes injured finger, injury plane, injury mechanism, number of dislocations, method of preserving amputated finger, whether combined with other injuries, degree of dislocation, bone and joint injury, finger artery injury, nerve injury, tendon injury plane, nail bed dorsal skin injury; intraoperative risk factor data includes ischemia time, operation time, doctor's seniority, vein graft, number of arterial anastomosis, number of venous anastomosis, anastomotic arteriovenous ratio, whether the flap covers the wound; postoperative risk factor data includes: whether vascular crisis occurs, hemoglobin, platelet, neutrophil count, lymphocyte count, monocyte count, neutrophil percentage, albumin, D-dimer; composite inflammatory indicators include neutrophil-lymphocyte ratio (NLR), monocyte-lymphocyte ratio (MLR), platelet-lymphocyte ratio (PLR), neutrophil-platelet ratio (NPR), systemic inflammatory response index (SIRI), systemic immune inflammation index (SII), systemic inflammatory syndrome index (AISI), prognostic nutrition index (PNI), neutrophil percentage to albumin ratio (NPAR).

[0056] In another embodiment, the data screening module 31 in the above Figure 3 is specifically configured to: use Lasso regression to screen first data from the data related to the replantation of amputated fingers, the first data having an influence degree on the risk assessment result of post-replantation necrosis greater than a preset value; use Boruta algorithm to screen second data from the data related to the replantation of amputated fingers, the second data having an influence degree on the risk assessment result of post-replantation necrosis greater than a preset value; and obtain target data by taking an intersection of the first data and the second data.

[0057] In another embodiment, the candidate model includes: logistic regression (Logistic regression, LR), support vector machine (Support Vector Machine, SVM), gradient boosting tree (Gradient Boosting Machine, GBM), neural network (Neural Network), random forest (RandomFofest, RF), extreme gradient boosting (eXtreme Gradient Boosting, Xgboost), K nearest neighbor (KNearest Neighbor Machine, KNN), adaptive boosting (AdaBoost), light gradient boosting machine (Light Gradient Boosting Machine, LightGBM), and categorical boosting (Categorical Boosting, CatBoost).

[0058] In another embodiment, such as Figure 4 As shown above, Figure 3 The model filtering module 32 in the middle includes: The indicator acquisition unit 320 is used to train candidate models based on target data and acquire evaluation indicators for each candidate training model. The evaluation indicators include accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, area under the curve, calibration curve, clinical decision curve, and confusion matrix. The performance evaluation unit 321 is used to evaluate the predictive performance of each candidate model based on evaluation metrics. The model selection unit 322 is used to select the target model from the candidate models based on the prediction performance of each candidate model.

[0059] In another embodiment, the above Figure 4 The model selection unit 322 is specifically used to: select the best model from the candidate models based on the prediction performance of each candidate model; and use the SHAP method to perform local and global interpretations on the best model to obtain the target model.

[0060] This application also provides an electronic device, in some embodiments, referring to... Figure 5 As shown, the electronic device 700 includes an input unit 710, a memory 720, a processor 730, and an output unit 740. The memory 720 stores program instructions that can be executed on the processor 730. The processor 730 can execute the program instructions to perform the assessment method and / or technical solution for the risk of necrosis after finger replantation in the foregoing embodiments. The electronic device 700 can be a mobile terminal device such as a mobile phone or computer.

[0061] Furthermore, embodiments of this application also provide a computer-readable storage medium for storing a computer program that performs a method for assessing the risk of necrosis after finger replantation. For example, computer program instructions, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. The program instructions that invoke the methods of this application may be stored in a fixed or removable storage medium, and / or transmitted via data streams in broadcast or other signal carrying media, and / or stored in a storage medium that operates according to the program instructions.

[0062] Obviously, those skilled in the art should understand that the modules or steps of the present application described above can be realized by universal computing devices, and can be centralized on a single computing device or distributed on a network composed of multiple computing devices. Alternatively, they can be realized by program codes executable by the computing devices, so that they can be stored in storage devices and executed by the computing devices, or they can be respectively manufactured into individual integrated circuit modules, or multiple modules or steps among them can be manufactured into a single integrated circuit module. Therefore, the present application is not limited to any specific combination of hardware and software.

[0063] The technical features of the above embodiments can be integrated in any manner. In order to make the description simple, all possible integrations of the technical features in the above embodiments are not described, however, as long as the integration of the technical features does not exist contradictions, it should be considered as the scope of the present application.

[0064] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for those skilled in the art, some modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for assessing the risk of necrosis after finger replantation, characterized in that, The method includes: Obtain data related to finger replantation surgery from sample patients; the data related to finger replantation surgery includes basic characteristic data, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data, and composite inflammatory markers of the sample patients; Lasso regression and Boruta algorithm were used to screen out target data from the finger replantation surgery-related data that had a greater impact on the risk assessment of necrosis after finger replantation than a preset value. Candidate models are trained based on the target data, and a target model is selected from the candidate models based on the prediction performance of each candidate model. Obtain target data from the target patient and input the target data into the target model to obtain the assessment result of the risk of necrosis after finger replantation in the target patient.

2. The method for assessing the risk of necrosis after finger replantation as described in claim 1, characterized in that, The basic characteristics of the sample patients included: age, gender, smoking history, hypertension, and diabetes; the preoperative risk factor data included the injured finger, injury level, injury mechanism, number of severed fingers, method of preservation of severed fingers, whether there were other injuries, degree of amputation, bone and joint injuries, digital artery injury, nerve injury, tendon injury level, and nail bed and dorsal finger skin injuries; the intraoperative risk factor data included ischemic time, operation time, surgeon's seniority, vein transplantation, number of arterial anastomoses, number of venous anastomoses, arteriovenous ratio of anastomosed arteries, and whether a skin flap was used to cover the wound; the postoperative risk factor data included whether vascular crisis occurred, blood... Hemoglobin, platelets, neutrophil count, lymphocyte count, monocyte count, neutrophil percentage, albumin, D-dimer; the composite inflammatory indicators include neutrophil-lymphocyte ratio (NLR), monocyte-lymphocyte ratio (MLR), platelet-lymphocyte ratio (PLR), neutrophil-platelet ratio (NPR), systemic inflammatory response index (SIRI), systemic immune inflammatory index (SII), systemic inflammatory composite index (AISI), prognostic nutritional index (PNI), and neutrophil percentage to albumin ratio (NPAR).

3. The method for assessing the risk of necrosis after finger replantation as described in claim 2, characterized in that, Lasso regression and Boruta algorithm were used to screen target data from the finger replantation surgery-related data that had a greater than preset value impact on the risk assessment of necrosis after finger replantation, including: Lasso regression was used to screen the first data point from the data related to finger replantation surgery that had a greater impact on the risk assessment of necrosis after finger replantation than a preset value. The Boruta algorithm was used to filter out second data from the data related to finger replantation surgery that had a greater impact on the risk assessment of necrosis after finger replantation than a preset value. The target data is obtained by taking the intersection of the first data and the second data.

4. The method for assessing the risk of necrosis after finger replantation as described in claim 1, characterized in that, The candidate models include: Logistic regression (LR), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), Neural Network, Random Forest (RF), eXtreme Gradient Boosting (Xgboost), K-Nearest Neighbor Machine (KNN), Adaptive Boosting (AdaBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost).

5. The method for assessing the risk of necrosis after finger replantation as described in claim 4, characterized in that, Train candidate models based on the target data, and select a target model from the candidate models according to the prediction performance of each candidate model, including: Candidate models are trained based on the target data, and evaluation metrics for each candidate training model are obtained; the evaluation metrics include accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, area under the curve, calibration curve, clinical decision curve, and confusion matrix. The predictive performance of each candidate model is evaluated based on the evaluation metrics. The target model is selected from the candidate models based on their predictive performance.

6. The method for assessing the risk of necrosis after finger replantation as described in claim 5, characterized in that, Based on the prediction performance of each candidate model, a target model is selected from the candidate models, including: Based on the prediction performance of each candidate model, the best model is selected from the candidate models; The SHAP method is used to perform local and global interpretations on the optimal model to obtain the target model.

7. A device for assessing the risk of necrosis after finger replantation, characterized in that, The device includes: The data acquisition module is used to acquire data related to finger replantation surgery of sample patients; the data related to finger replantation surgery includes the basic characteristic data, preoperative risk factor data, intraoperative risk factor data, postoperative risk factor data and composite inflammatory indicators of the sample patients; The data filtering module is used to filter out target data from the data related to finger replantation surgery that have a greater impact on the risk assessment of necrosis after finger replantation than a preset value using Lasso regression and Boruta algorithm. The model selection module is used to train candidate models based on the target data, and select the target model from the candidate models according to the prediction performance of each candidate model. The risk assessment module is used to acquire target data of the target patient and input the target data of the target patient into the target model to obtain the assessment result of the risk of necrosis after finger replantation in the target patient.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for assessing the risk of necrosis after finger replantation as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for assessing the risk of necrosis after finger replantation as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for assessing the risk of necrosis after replantation of a severed finger as described in any one of claims 1 to 6.

Images