Mediterranean anemia prediction method based on extreme random tree

Abstract

The invention relates to the field of medical information engineering, in particular to a thalassemia prediction method based on an extreme random tree, which comprises the steps of S1, constructing a thalassemia data set, S2, carrying out data preprocessing on the data set, the data preprocessing comprising data standardization, resampling and feature screening, S3, dividing data of the data set into a training set and a test set, and S4, carrying out feature extraction on the training set and the test set. The method comprises the following steps: S1, training an extreme random tree model, S4, collecting data of a to-be-tested person, processing the data, importing the data into the model, and carrying out classification prediction on the data of the to-be-tested person. The method has the beneficial effects that the problem of data set imbalance is solved through a synthetic minority class oversampling and self-adaptive synthetic sampling method; selecting index data with relatively high importance on predicting thalassemia from a blood routine examination report through feature screening to construct a feature set for training a classification model and improving the robustness of the model; and classification prediction is carried out by using an integrated classifier extreme random tree, so that the classification accuracy and the model generalization ability are improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information engineering, and in particular to a thalassemia prediction method based on an extremely random tree. Background Art

[0002] Thalassemia, also known as thalassemia or thalassemia syndrome, is a common hereditary blood disease. It is caused by impaired hemoglobin synthesis, and its main characteristics are abnormal erythropoiesis and anemia. Thalassemia is more common in certain parts of the world, especially in the Mediterranean region, Southeast Asia and Africa, and is highly prevalent in my country, Guangdong, Guangxi and Yunnan. In these high-incidence areas, the proportion of thalassemia carriers is relatively high, which means that even if they do not show clinical symptoms, a considerable number of people carry abnormal genes. Therefore, it is particularly important to screen for thalassemia in people in these areas, which can prevent the birth of severely ill children, achieve early diagnosis and treatment, optimize the allocation of medical resources, and reduce the economic burden on society and families.

[0003] Although genetic testing is the gold standard for diagnosing thalassemia, its long testing cycle and high cost limit its feasibility for large-scale population screening in resource-limited areas. In addition, the shortage of equipment and professional personnel often faced by primary hospitals also affects the widespread application of genetic testing. Therefore, it is particularly important to develop a rapid and economical screening method for thalassemia.

[0004] Routine blood tests have the advantages of wide availability, mature technology, simple operation procedures and low cost, making them an ideal choice for thalassemia screening. By using routine blood tests for screening, the efficiency of screening can be effectively improved and the cost can be significantly reduced, especially in resource-constrained environments. The implementation of this method will help to widely identify carriers of thalassemia, thereby promoting early intervention and the implementation of preventive measures.

[0005] In thalassemia screening, the amount of data from healthy people is often much larger than that from patients, resulting in an imbalanced training data set. Due to the problem of data imbalance, existing thalassemia screening models based on machine learning, such as the cuckoo algorithm combined with the random forest method, are prone to overfitting the majority class data during the training process, making the model's recognition ability for minority classes (i.e., patients) insufficient, resulting in low model prediction accuracy and poor model generalization ability, thereby reducing the effectiveness and reliability of the screening model in practical applications. Therefore, it is necessary to further improve and optimize existing methods to improve prediction accuracy and generalization ability, and provide more effective technical support for the early detection and treatment of thalassemia.

[0006] Therefore, it is necessary to invent a thalassemia prediction method based on extremely randomized trees. Summary of the invention

[0007] To this end, the present invention provides a thalassemia prediction method based on an extremely random tree. The method solves the problem of data set imbalance by adopting minority class oversampling and adaptive synthetic sampling methods. At the same time, it screens out indicator data that are more important for predicting thalassemia to construct feature vectors, and uses important features to build a prediction model. Finally, an extremely random tree is used for prediction to improve classification accuracy and model generalization. This method provides effective technical support for the early diagnosis and treatment of thalassemia.

[0008] In order to achieve the above object, the present invention provides the following technical solution: a method for predicting thalassemia based on an extremely random tree, specifically comprising the following steps:

[0009] S1, constructing the thalassemia dataset;

[0010] S2, preprocessing the data set, including data standardization, resampling and feature screening;

[0011] S3, divide the data set into training set and test set, and train the extreme random tree model;

[0012] S4, collects the data of the subjects, processes the data, and then imports it into the model to perform classification prediction on the data of the subjects.

[0013] Preferably, in the step S1, a thalassemia data set is constructed through the blood test records of the subject, and the blood test records of the subject include the subject's basic personal information, including gender, age, and blood routine index test data.

[0014] Preferably, the blood routine index detection data include white blood cell count (WBC), red blood cell count (RBC), hemoglobin level (HGB), hematocrit (HCT), mean corpuscular volume (MCV), mean hemoglobin amount (MCH), mean hemoglobin concentration (MCHC), mean red blood cell distribution width coefficient of variation (RDW-CV), red blood cell volume distribution width standard deviation (RDW-SD), and platelets (PLT).

[0015] Preferably, in the step S2, the collected data is cleaned, missing values ​​and outliers are processed, necessary data conversion and standardization are performed, and a balanced distribution of categories is achieved through resampling techniques such as synthetic minority class oversampling and adaptive synthetic sampling, and the features that have the greatest impact on thalassemia screening are screened out from multiple indicators of routine blood tests.

[0016] Preferably, said characteristics include red blood cell count, hemoglobin level, mean corpuscular volume.

[0017] Preferably, in step S3, 90% of the data are randomly selected for training, and the remaining 10% of the data are used for testing.

[0018] Preferably, in the step S3, the model is trained using the training set data, the model performance is optimized by adjusting the model parameters, the performance of the model is evaluated on the test set, the accuracy, recall and precision indicators are used to measure the prediction effect of the model, and the model structure and parameters are optimized and adjusted according to the evaluation results.

[0019] Preferably, the optimized model is applied in actual screening work to assist doctors in the diagnosis and risk assessment of thalassemia.

[0020] Preferably, the blood routine index detection process is to first collect 2 ml of fasting elbow venous blood from the patient in the morning, place the blood sample in an ethylenediaminetetraacetic acid anticoagulation tube for anticoagulation, and then use a blood cell analyzer to detect and record the white blood cell count (WBC), red blood cell count (RBC), hemoglobin level (HGB), hematocrit (HCT), mean corpuscular volume (MCV), mean hemoglobin amount (MCH), mean hemoglobin concentration (MCHC), mean red blood cell distribution width coefficient of variation (RDW-CV), red blood cell volume distribution width standard deviation (RDW-SD), and platelet (PLT) indicator data.

[0021] The beneficial effects of the present invention are as follows: the problem of data set imbalance is solved by synthesizing minority class oversampling and adaptive synthetic sampling methods; feature screening is used to screen indicator data with high importance for predicting thalassemia in routine blood tests to construct a feature set, so as to obtain the optimal feature set for classifier training; the indicator data with high importance for predicting thalassemia is selected from the routine blood test report through feature screening to construct a feature set for training a classification model, thereby improving the robustness of the model; and the classification prediction is performed using an integrated classifier extreme random tree, thereby improving the classification accuracy and the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of a thalassemia prediction method based on an extremely random tree provided by the present invention. DETAILED DESCRIPTION

[0023] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0024] Please refer to the attached Figure 1 The present invention provides a thalassemia prediction method based on an extremely random tree, which specifically comprises the following steps:

[0025] S1, construct a thalassemia dataset. The thalassemia dataset is constructed through the blood test records of the subjects. The blood test records of the subjects include the basic personal information of the subjects, including gender, age, and blood routine index test data. The blood routine index test data include white blood cell count (WBC), red blood cell count (RBC), hemoglobin level (HGB), hematocrit (HCT), mean corpuscular volume (MCV), mean hemoglobin (MCH), mean hemoglobin concentration (MCHC), mean red blood cell distribution width coefficient of variation (RDW-CV), red blood cell volume distribution width standard deviation (RDW-SD ), platelets (PLT), the blood routine index detection process is to first collect 2ml of fasting elbow venous blood from the patient in the morning, place the blood sample in an ethylenediaminetetraacetic acid anticoagulation tube for anticoagulation, and then use a blood cell analyzer to detect and record the white blood cell count (WBC), red blood cell count (RBC), hemoglobin level (HGB), hematocrit (HCT), mean corpuscular volume (MCV), mean hemoglobin (MCH), mean hemoglobin concentration (MCHC), mean red blood cell distribution width coefficient of variation (RDW-CV), red blood cell volume distribution width standard deviation (RDW-SD), platelet (PLT) index data;

[0026] S2, preprocessing the data set, including data standardization, resampling and feature screening, cleaning the collected data, processing missing values ​​and outliers, performing necessary data conversion and standardization, achieving balanced distribution of categories through resampling techniques of synthetic minority class oversampling and adaptive synthetic sampling, and screening out the most influential features for thalassemia screening from multiple indicators of routine blood examination, including red blood cell count, hemoglobin level, and mean corpuscular volume;

[0027] Since in clinical scenarios, the number of patients with thalassemia among the subjects is far less than the number of non-thalassemia patients, there is a sample imbalance problem, which leads to overfitting of the model and reduces the prediction accuracy. In order to solve this problem, this patent proposes data resampling, which achieves balanced distribution of categories through resampling technology of synthetic minority class oversampling and adaptive synthetic sampling;

[0028] The synthetic minority class oversampling in this patent is a statistical technique that can be used to balance the categories of a data set. It solves the overfitting problem by adding new instances to the minority class. It randomly selects a data sample from the minority class and finds the nearest neighbor sample of the data sample. In this patent, the frequency of selecting a random data point and creating a new sample data for the row at the selected point is set to k=5;

[0029] At the same time, this patent adopts an adaptive synthetic sampling method to generate samples with more learning value. It selects random points for generating minority class samples by finding linear correlation values, thereby generating synthetic minority class samples. For each sample x in the minority class i , the synthetic sample can be calculated as follows:

[0030] S i =x i +(x u -x i )×λ

[0031] Where λ is a random number, λ∈[0,1], x u -x i is the vector difference in n-dimensional space;

[0032] Through the above two data resampling methods, the distribution of data can be improved, so that the number of thalassemia patients and non-thalassemia patients in the data set is basically the same, reducing the deviation caused by class imbalance, reducing model overfitting, and further improving the prediction accuracy of the model;

[0033] It should be noted that routine blood test is a commonly used detection method in medical diagnosis, which can provide important information about the blood status. For the screening of thalassemia, the following are some blood routine indicators related to the disease:

[0034] Red blood cell count (RBC): In patients with thalassemia minor (such as thalassemia carriers), the RBC value may be within the normal range or have only slight changes. In patients with thalassemia intermedia or major, the RBC value may be significantly reduced, leading to anemia, due to bone marrow hematopoietic dysfunction and abnormal destruction of red blood cells.

[0035] Hemoglobin levels (HGB): The hemoglobin levels of patients with thalassemia are usually lower than the normal range because the condition affects the synthesis of hemoglobin;

[0036] Hematocrit (HCT) is the percentage of red blood cell volume in the blood to the total blood volume. It is usually used to assess the blood status of an individual. Due to impaired hemoglobin synthesis, the production and life span of red blood cells in patients with thalassemia may be affected, which may lead to a lower HCT value.

[0037] Mean Corpuscular Volume (MCV): The MCV value of patients with thalassemia is usually low, which means that the average volume of red blood cells is smaller than normal;

[0038] Mean corpuscular hemoglobin (MCH): This index reflects the average amount of hemoglobin contained in each red blood cell. The MCH value of patients with thalassemia is often low.

[0039] Mean Corpuscular Hemoglobin Concentration (MCHC): This refers to the average concentration of hemoglobin in red blood cells. The MCHC of patients with thalassemia may be reduced.

[0040] The mean red cell distribution width coefficient of variation (RDW-CV) is a statistical parameter that measures the variation in red blood cell size. It is the ratio of the standard deviation of the red blood cell distribution width (RDW) to the mean corpuscular volume (MCV), usually expressed as a percentage. The red blood cells of patients with thalassemia usually show characteristics of varying sizes. This is because abnormal hemoglobin synthesis affects the maturation and morphology of red blood cells. Therefore, the patient's RDW-CV may be elevated, reflecting the increased heterogeneity of red blood cell size. This elevated RDW-CV is associated with the pathological state of thalassemia, especially in patients with α-thalassemia and β-thalassemia.

[0041] The red blood cell volume distribution width standard deviation (RDW-SD) is an indicator of the variability of red blood cell volume. It reflects the heterogeneity of red blood cell size in the blood. In the context of thalassemia, RDW-SD may show specific changes. The red blood cells of patients with thalassemia may show abnormal size and shape. This is because abnormal hemoglobin synthesis affects the production and maturation of red blood cells. Therefore, the patient's RDW-SD may increase, showing a wide variation in red blood cell size, which is consistent with the pathological characteristics of thalassemia.

[0042] S3, divide the data set into a training set and a test set, randomly select 90% of the data for training, and the remaining 10% of the data for testing, train the extreme random tree model, use the training set data to train the model, and optimize the model performance by adjusting the model parameters. Use {T(K,X,D)} to represent the extreme random tree, where T represents the final classifier model, D is the data sample set, K is the number of base classifiers, and each base classifier generates a prediction result based on the input sample X={x1,x2,…,xN}. Finally, the final classification category is determined by voting. The specific training steps are as follows:

[0043] Step 1: Given the original data set D, the number of samples N, and the number of features M, each base classifier is trained using all the samples;

[0044] Step 2: Generate a base classifier based on the CART decision tree algorithm. To enhance randomness, randomly select m features from M features when splitting each node. Select the optimal attribute for each node to split the node. No pruning is done during the splitting process. For each decision tree, select the features that can reduce entropy to the greatest extent for node splitting, that is, maximize the gain to select the best split point.

[0045] The entropy value is calculated as follows:

[0046]

[0047] Where S is the data set, c is the number of class labels, and is the probability of the i-th category in the data set;

[0048] The gain can be expressed as:

[0049]

[0050] Where wl is the weight of the left child node, W is the sum of sample weights, Il is the sample set contained in the left child node, Δyi is the difference between the target value of sample i and the predicted value of the current model, and λ is the regularization parameter;

[0051] Iterate step 2 again on the data subsets generated by the split until a decision tree is generated;

[0052] Step 3: Repeat steps 1 and 2 for K iterations to generate K decision trees and extreme random trees;

[0053] Step 4: Use the test sample to generate prediction results for the generated extreme random tree, perform statistics on the prediction results of all base classifiers, and use the voting decision method to generate the final classification result;

[0054]

[0055] F(x) is the model's prediction value for sample x, Tj(x) is the prediction value of the jth decision tree for sample x, and a j is the weight of the jth tree;

[0056] Evaluate the performance of the model on the test set, use accuracy, recall, and precision to measure the prediction effect of the model, optimize the model structure and parameters based on the evaluation results, and apply the optimized model to actual screening work to assist doctors in diagnosing and assessing the risk of thalassemia;

[0057] It should be noted that extreme random trees are an efficient ensemble learning algorithm. The algorithm is based on the gradient boosting decision tree framework and introduces a variety of optimization strategies to improve the performance and computing speed of the model. Unlike typical decision trees and random forests, extreme random trees can use the entire data set instead of a subset of the data set when constructing each decision tree. Instead of selecting the best split point from all features, a certain number of features are randomly selected as candidate sets. At each decision node, the algorithm evaluates and selects the best split feature from these candidate features to construct each decision tree. By integrating multiple relatively independent decision trees, extreme random trees can integrate the prediction results of each decision tree to obtain a more accurate and robust final output. The randomization process of extreme random trees reduces the correlation between trees, which not only helps to reduce the degree of model fitting to training data, but also enhances the generalization ability of the model on unknown data, which helps to improve the stability and prediction performance of the model. Extreme random trees perform well in processing large-scale data sets and can effectively process numerical and categorical features. Especially when processing complex and high-dimensional data sets, it can effectively improve the accuracy of classification.

[0058] S4, collects the data of the subjects, processes the data, and then imports it into the model to perform classification prediction on the data of the subjects.

[0059] The above is only a preferred embodiment of the present invention. Any person skilled in the art can modify the present invention by using the above technical solution or modify it into an equivalent technical solution. Therefore, any simple modification or equivalent replacement made according to the technical solution of the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A thalassemia prediction method based on extremely randomized trees, characterized by: The specific steps include: S1, constructing the thalassemia dataset; S2, preprocessing the data set, including data standardization, resampling and feature screening; S3, divide the data set into training set and test set, and train the extreme random tree model; S4, collects the data of the subjects, processes the data, and then imports it into the model to perform classification prediction on the data of the subjects.

2. The method for predicting thalassemia based on an extremely random tree according to claim 1, characterized in that: In the step S1, a thalassemia data set is constructed through the blood test records of the subject, and the blood test records of the subject include the basic personal information of the subject, including gender, age, and blood routine index test data.

3. The method for predicting thalassemia based on extremely randomized trees according to claim 2, characterized in that: The blood routine index detection data include white blood cell count (WBC), red blood cell count (RBC), hemoglobin level (HGB), hematocrit (HCT), mean corpuscular volume (MCV), mean hemoglobin content (MCH), mean hemoglobin concentration (MCHC), mean red blood cell distribution width coefficient of variation (RDW-CV), red blood cell volume distribution width standard deviation (RDW-SD), and platelets (PLT).

4. The method for predicting thalassemia based on extremely randomized trees according to claim 3, characterized in that: In the step S2, the collected data is cleaned, missing values ​​and outliers are processed, necessary data conversion and standardization are performed, and a balanced distribution of categories is achieved through resampling techniques such as synthetic minority class oversampling and adaptive synthetic sampling, and the features that have the greatest impact on thalassemia screening are screened out from multiple indicators of routine blood tests.

5. The method for predicting thalassemia based on extremely randomized trees according to claim 4, characterized in that: The characteristics include red blood cell count, hemoglobin level, mean corpuscular volume.

6. The method for predicting thalassemia based on extremely randomized trees according to claim 1, characterized in that: In the S3 step, 90% of the data are randomly selected for training, and the remaining 10% of the data are used for testing.

7. The method for predicting thalassemia based on extremely randomized trees according to claim 1, characterized in that: In the step S3, the model is trained using the training set data, the model performance is optimized by adjusting the model parameters, the performance of the model is evaluated on the test set, the accuracy, recall and precision indicators are used to measure the prediction effect of the model, and the model structure and parameters are optimized and adjusted according to the evaluation results.

8. The method for predicting thalassemia based on extremely randomized trees according to claim 7, characterized in that: The optimized model is applied in actual screening work to assist doctors in the diagnosis and risk assessment of thalassemia.

9. The method for predicting thalassemia based on extremely randomized trees according to claim 3, characterized in that: The blood routine index detection process is to first collect 2 ml of fasting elbow venous blood from the patient in the morning, place the blood sample in an ethylenediaminetetraacetic acid anticoagulation tube for anticoagulation, and then use a blood cell analyzer to detect and record the white blood cell count (WBC), red blood cell count (RBC), hemoglobin level (HGB), hematocrit (HCT), mean corpuscular volume (MCV), mean hemoglobin (MCH), mean hemoglobin concentration (MCHC), mean red blood cell distribution width coefficient of variation (RDW-CV), red blood cell volume distribution width standard deviation (RDW-SD), and platelet (PLT) index data.

Images