A multi-model fusion classification method based on blood diseases

Abstract

The present invention discloses a multi-model fusion classification method based on blood diseases, comprising the following steps: obtaining a data set; dividing the data set into k parts based on a k-fold cross-validation method to obtain a test set and a training set, where k is any constant greater than 1; selecting a plurality of machine learning binary classifier models for data set training; fusing a plurality of machine learning binary classifier models by linear addition to obtain a fused multi-classifier model; using the machine learning multi-classifier model to train a multi-label data set to obtain a model accuracy rate; using the fused multi-classifier model to train a multi-label data set to obtain an accuracy rate of the fused multi-classifier model, and comparing it with the above accuracy rate. The present invention adopts the above-mentioned multi-model fusion classification method based on blood diseases, which can quickly classify blood diseases to help doctors quickly diagnose, reduce doctors' workload, and improve their work efficiency.

Description

Technical Field

[0001] The present invention relates to the field of laboratory medicine and disease identification technology, and in particular to a multi-model fusion classification method based on blood diseases. Background Art

[0002] Blood diseases are diseases that originate in the hematopoietic system or affect the hematopoietic system with abnormal blood changes, characterized by anemia, bleeding, fever, and hepatosplenomegaly and lymphadenopathy. The disease has the characteristics of low morbidity, high mortality, unpredictability, involving multiple systems, and diagnosis relying on laboratory tests. It can be primary or secondary. Among them, the causes of malignant blood diseases such as leukemia, lymphoma, and aplastic anemia are very complex, and there are certain difficulties in treatment. According to statistics, there are about 500 million people in the world suffering from blood diseases to varying degrees. In addition, many people suffer from blood diseases without knowing it, and by the time they are discovered, they are often in a very serious stage. Blood diseases are mostly refractory diseases, with insidious onset and hidden symptoms. They are often discovered when seeking medical treatment for other diseases or health examinations. Therefore, it is particularly important to raise awareness of blood diseases so that they can be discovered and treated early to avoid unnecessary losses to health.

[0003] Blood diseases include many types of diseases, with a variety of symptoms and signs, which often lack specificity. It takes multiple tests to confirm the diagnosis. The clinical manifestations of blood diseases are similar, and clinicians need to have a lot of professional knowledge, rich clinical experience and keen observation when distinguishing. With the development of artificial intelligence, it has also been widely used in the medical field. At present, the diagnostic model for blood diseases is mainly based on bone puncture smear analysis of cell morphology, immunology and genetics to determine the type of disease, but bone puncture smear is image data, so the image needs to be collected and manually annotated. High-quality bone marrow puncture smear image data sets are not easy to obtain, and it requires a lot of time, energy and medical knowledge. Bone puncture smears mainly analyze cell morphology. The overlap, adhesion, blurred boundaries and unclear types between cells make the image more challenging, so there are certain problems and difficulties in using models to diagnose blood diseases. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-model fusion classification method based on blood diseases, which can quickly classify blood diseases to help doctors quickly diagnose them, reduce doctors' workload and improve their work efficiency.

[0005] To achieve the above object, the present invention provides a multi-model fusion classification method based on blood diseases, comprising the following steps:

[0006] S1 extracts blood routine test data related to blood diseases and obtains a data set;

[0007] S2 is based on the k-fold cross-validation method, which divides the data set into k parts to obtain the test set and the training set, where k is any constant greater than 1;

[0008] S3 selects multiple machine learning binary classifier models for data set training and assigns weight values ​​to each machine learning binary classifier model based on the number of data sets for different diseases;

[0009] S4 combines the weight values ​​of multiple machine learning binary classifier models with the corresponding model prediction values ​​to construct a formula, and fuses multiple machine learning binary classifier models through linear addition to obtain a fused multi-classifier model;

[0010] S5 uses a machine learning multi-classifier model to train a multi-label dataset and obtains the accuracy of the machine learning multi-classifier model;

[0011] S6 uses the fused multi-classifier model to train the multi-label data set to obtain the accuracy of the fused multi-classifier model and compares it with the accuracy of step S5;

[0012] S7 determines whether the effect of the fused multi-classifier model is better than that of the machine learning multi-classifier model. If so, the fusion is successful. If not, continue to modify the fusion formula of multiple machine learning binary classification models.

[0013] Preferably, the data set extracted in step S1 includes a leukemia data set, a myelodysplastic syndrome data set, a aplastic anemia data set and an infectious mononucleosis data set; and the leukemia data set is combined with the myelodysplastic syndrome data set, the aplastic anemia data set and the infectious mononucleosis data set to form three data sets.

[0014] Preferably, in step S2, k is generally set to 5 or 10.

[0015] Preferably, all data sets in step S3 are classified using a random forest binary classifier, and the parameters used in the model constructed for each data set are different. The classification effect is tested using the test set obtained in step S2 to obtain the sensitivity TPR, specificity TNR and accuracy ACC;

[0016] Among them, the calculation formulas of TPR, TNR and ACC are:

[0017] TPR=TP / (TP+FN) (1)

[0018] TNR=TN / (FP + TN) (2)

[0019] ACC=(TP+TN) / (TP+FP+FN+TN) (3)

[0020] In the formula, TP represents the number of patients identified by the classifier, FP represents the number of normal people mistakenly identified as patients by the classifier, FN represents the number of patients identified as normal people by the classifier, and TN represents the number of normal people identified by the classifier;

[0021] The weight value w of the model i (i=1, 2, 3) is set according to the ratio between the number of disease data sets.

[0022] Preferably, in step S4, different data sets are tested on different machine learning binary classifier models to obtain a predicted value α; the formula in S4 is:

[0023]

[0024] In the formula, m is the number of base classification algorithms, h i is the i-th base classification algorithm, w i is the weight parameter of the i-th base classification algorithm, α is the model prediction value, and x represents the data set.

[0025] Preferably, in step S5, the multi-label data set is to integrate the extracted blood disease data, mark sample labels for different blood diseases, and then use a random forest binary classifier to train the multi-label data set to obtain the model accuracy.

[0026] Therefore, the multi-model fusion classification method based on blood diseases of the present invention uses conventional test data to build a model to classify blood diseases. Compared with image classification, the classification of conventional test data is relatively simple, and conventional test data is relatively easy in data acquisition and collection. The overall implementation is convenient and fast, which is conducive to solving the difficult problem of blood disease classification. In terms of machine learning, it can solve the fitting problem between model fusions and improve the generalization performance of machine learning. It can quickly classify blood diseases to help doctors quickly diagnose, reduce doctors' workload and improve their work efficiency.

[0027] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flow chart of a multi-model fusion classification method based on blood diseases of the present invention. DETAILED DESCRIPTION

[0029] The present invention is further described below in conjunction with embodiments. Unless otherwise defined, the technical terms or scientific terms used in the present invention should be the common meanings understood by people with ordinary skills in the field to which the present invention belongs. The above-mentioned features or features mentioned in the specific examples mentioned in the present invention can be combined arbitrarily, and these specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0030] Figure 1 The figure is a flow chart of a multi-model fusion classification method based on blood diseases of the present invention. As shown in the figure, the present invention provides a multi-model fusion classification method based on blood diseases, comprising the following steps:

[0031] S1 extracts blood routine test data related to blood diseases and obtains a data set;

[0032] The data set extracted in step S1 includes a leukemia data set, a myelodysplastic syndrome data set, aplastic anemia data set and an infectious mononucleosis data set; and the leukemia data set is combined with the myelodysplastic syndrome data set, the aplastic anemia data set and the infectious mononucleosis data set to form three data sets.

[0033] S2 is based on the k-fold cross-validation method, which divides the data set into k parts to obtain a test set and a training set. The value of k is generally 5 or 10.

[0034] S3 selects multiple machine learning binary classifier models for data set training and assigns weight values ​​to each model based on the number of data sets for different diseases.

[0035] All data sets are classified using the random forest binary classifier. The parameters used in the model constructed for each data set are different. The classification effect is tested using the test set obtained in step S2 based on the k-fold cross-validation method to obtain TPR, TNR and ACC. The results are shown in Table 1. TPR is the proportion of all positive samples identified to all samples, i.e., sensitivity; TNR is the proportion of all negative samples identified to all negative samples, i.e., specificity; and ACC is the classification accuracy of the classifier.

[0036] Among them, the calculation formulas of TPR, TNR and ACC are:

[0037] TPR=TP / (TP+FN) (1)

[0038] TNR=TN / (FP+TN) (2)

[0039] ACC=(TP+TN) / (TP+FP+FN+TN) (3)

[0040] In the formula, TP represents the number of patients identified by the classifier, FP represents the number of normal people mistakenly identified as patients by the classifier, FN represents the number of patients identified as normal people by the classifier, and TN represents the number of normal people identified by the classifier;

[0041] The leukemia dataset was combined with the myelodysplastic syndrome dataset to form the first dataset;

[0042] The leukemia dataset was combined with the aplastic anemia dataset to form the second dataset;

[0043] The leukemia dataset was combined with the infectious mononucleosis dataset to form a third dataset.

[0044] The parameters used for each data set of the random forest binary classifier are:

[0045] The first random forest algorithm uses the RandomForestClassifier function, where class_weight is set to {0:5,1:67}, n_estimators is set to 11, min_samples_split is set to 2, min_samples_leaf is set to 2, and max_depth is set to 10;

[0046] The second random forest algorithm uses the RandomForestClassifier function, where class_weight is set to {0:50,1:270}, n_estimators is set to 15, min_samples_split is set to 2, min_samples_leaf is set to 2, and max_depth is set to 3;

[0047] The third random forest algorithm uses the RandomForestClassifier function with class_weight set to {0:0.1,1:60}, n_estimators set to 15, min_samples_split set to 2, min_samples_leaf set to 2, and max_depth set to 6.

[0048] The weight value w of the model i (i=1, 2, 3) is set according to the ratio between the number of disease data sets.

[0049] Table 1 Results of training the machine learning binary classifier model on three datasets

[0050]

[0051]

[0052] S4 constructs a formula by combining the weight values ​​of multiple machine learning binary classifier models with the corresponding model prediction values, and fuses multiple machine learning binary classifier models through linear addition, where the prediction value α refers to the value obtained by testing different data sets in different machine learning binary classifier models.

[0053] The formula is:

[0054]

[0055] In the formula, m is the number of base classification algorithms, h i is the i-th base classification algorithm, w i is the weight parameter of the i-th base classification algorithm, α is the model prediction value, and x represents the data set.

[0056] S5 uses a machine learning multi-classifier model to train a multi-label dataset and obtains the accuracy of the machine learning multi-classifier model.

[0057] The multi-label dataset integrates the extracted blood disease data, marks sample labels for different blood diseases, and then uses a random forest multi-classifier to train the multi-label dataset to obtain the model accuracy.

[0058] The parameters used for the random forest classifier are:

[0059] The random forest algorithm uses the RandomForestClassifier function with n_estimators set to 15, min_samples_split set to 2, min_samples_leaf set to 2, and max_depth set to 20.

[0060] S6 uses the fused multi-classifier model to train the multi-label dataset to obtain the accuracy of the fused multi-classifier model and compares it with the accuracy of step S5, see Table 2.

[0061] Table 2 Comparison of evaluation indicators of the two models

[0062]

[0063]

[0064] S7 determines whether the effect of the fused multi-classifier model is better than that of the machine learning multi-classifier model. If so, the fusion is successful. If not, continue to modify the fusion formula of multiple machine learning binary classification models.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A multi-model fusion classification method based on blood diseases, characterized in that: The following steps are involved: S1 extracts blood routine test data related to blood diseases and obtains a data set; The extracted datasets include the leukemia dataset, the myelodysplastic syndrome dataset, the aplastic anemia dataset, and the infectious mononucleosis dataset; The leukemia dataset was combined with the myelodysplastic syndrome dataset, aplastic anemia dataset, and infectious mononucleosis dataset to form three datasets; S2 is based on the k-fold cross-validation method, which divides the data set into k parts to obtain the test set and the training set, where k is any constant greater than 1; S3 selects multiple machine learning binary classifier models for data set training and assigns weight values ​​to each machine learning binary classifier model based on the number of data sets for different diseases; All data sets in step S3 are classified using random forest binary classifiers. The parameters used in the model constructed for each data set are different, and the weight values ​​of the model are , =1,2,3, is set according to the ratio between the number of data sets for each disease; S4 combines the weight values ​​of multiple machine learning binary classifier models with the corresponding model prediction values ​​to construct a formula, and fuses multiple machine learning binary classifier models to obtain a fused multi-classifier model; In step S4, different data sets are tested on different machine learning binary classifier models to obtain predicted values ; The formula in S4 is: ； In the formula, m is the number of base classification algorithms, is the i-th base classification algorithm, For the The weight parameter of the seed-based classification algorithm, is the model prediction value, Represents a dataset; S5 uses a machine learning multi-classifier model to train a multi-label dataset and obtains the accuracy of the machine learning multi-classifier model; S6 uses the fused multi-classifier model to train the multi-label data set to obtain the accuracy of the fused multi-classifier model and compares it with the accuracy of step S5; S7 determines whether the effect of the fused multi-classifier model is better than that of the machine learning multi-classifier model. If so, the fusion is successful. If not, continue to modify the fusion formula of multiple machine learning binary classification models.

2. A multi-model fusion classification method based on blood diseases according to claim 1, characterized in that: In step S2, k is generally set to 5 or 10.

3. The multi-model fusion classification method based on blood diseases according to claim 1 is characterized in that: In step S5, the multi-label data set integrates the extracted blood disease data, labels the samples of different blood diseases, and then uses a random forest binary classifier to train the multi-label data set to obtain the model accuracy.

Images