CN109948732B - A method and system for classifying distant metastasis of abnormal cells based on non-equilibrium learning - Google Patents

Abstract

The disclosure provides a method and a system for classifying abnormal cell distant metastasis based on unbalanced learning, wherein a plurality of data sequences with certain cell distant metastasis and a plurality of data sequences without certain cell distant metastasis are obtained, the data set is divided into a training set and a testing set, the training set is used for training a model, and the testing set is used for testing the model; firstly, inputting a training set into a feature selection algorithm to be compared with the results of the classification of an original condition data set, and selecting p features with the best results; obtaining a training set with a positive-negative sample ratio of 1:1 by using an oversampling algorithm, respectively inputting the training set into a classification algorithm, testing by using a data sequence of a test set, and selecting an oversampling algorithm i of a training set Pi with an optimal evaluation result; and (3) inputting the training set into an oversampling algorithm for obtaining a training set Pi by adjusting the proportion of the positive and negative samples, gradually increasing the proportion of the positive and negative samples to a set proportion, and carrying out classification evaluation on the optimal proportion of the positive and negative samples. According to the technical scheme, the proportion of the positive samples is increased by adopting an oversampling algorithm, and better model evaluation indexes and the recall rate of a few positive samples are obtained.

Description

A method and system for classifying distant metastasis of abnormal cells based on non-equilibrium learning

technical field

The present disclosure relates to the technical field of machine learning and data mining, and in particular, to a method and system for classifying distant metastasis of abnormal cells based on unbalanced learning.

Background technique

Esophageal squamous cell carcinoma is one of the most common malignant tumors in the world, but its early symptoms are not obvious, and changes in the body are easy to be ignored. In clinical practice, doctors diagnose whether there is distant metastasis of cancer cells in patients with esophageal squamous cell carcinoma through imaging, even puncture and surgery. These three methods not only increase the cost of treatment for patients, but also take a long time. With the advent of the era of big data, in order to solve this problem, it is proposed to use blood cell analysis to predict whether a patient's cancer cells have metastasized. According to the relevant literature, it is known that the classification and prediction of lymph node metastasis has been published in the medical field, and the specificity and sensitivity are less than 50%, but no relevant research has been done on distant metastasis. SPSS, SAS) for P-test statistical analysis, and the present disclosure uses machine learning for analysis and prediction.

Since the collected data on patients with esophageal squamous cell carcinoma is not much, and the patients with distant metastasis of cancer cells account for a small proportion, there is a problem of category imbalance.

The inventor found in the research that in this imbalanced dataset, the standard classifier tends to obtain the maximum accuracy, while ignoring the minority class samples, which are the focus of attention, even if the results are very high. The accuracy of this analysis is also meaningless, and it is difficult to effectively predict whether a patient's cancer cells have metastasized. In real life, especially in the medical field, the problem of class imbalance is very common, which is mainly caused by morbidity. In this case, if the imbalanced data is not processed, the performance of the standard classifier will be severely affected.

SUMMARY OF THE INVENTION

The purpose of the embodiments of this specification is to provide a method for classifying distant metastases of abnormal cells based on unbalanced learning, using an oversampling algorithm to try to increase the proportion of positive samples, and to obtain better model evaluation indicators and recall rate of a few positive samples.

Embodiments of the present specification provide a method for classifying distant metastasis of abnormal cells based on non-equilibrium learning, including:

Obtain several data sequences with distant metastasis of a certain cell and several data sequences without distant metastasis of a certain cell, and form a training set;

Input the training set into k feature selection algorithms, respectively select the top p attributes as the features of the training set, input the classifier for training, compare the classification results, and select the p features with the best results;

Based on the oversampling algorithm, the training set achieves data balance at the data level, and the training set processed by the feature selection algorithm is input into n oversampling algorithms to obtain a training set with a positive and negative sample ratio of 1:1;

Input the training set with the ratio of positive and negative samples to 1:1 into the classification algorithm respectively, and then use the data sequence of the test set to test, and select the oversampling algorithm i that obtains the training set Pi with the best evaluation result;

By adjusting the ratio of positive and negative samples, input the training set to the oversampling algorithm to obtain the training set Pi, gradually increase the ratio of positive and negative samples to the set ratio, and classify and evaluate the optimal ratio of positive and negative samples.

Embodiments of the present specification provide a non-equilibrium learning-based cell distant metastasis classification system, including:

The training set acquisition unit is configured to: obtain several data sequences with distant metastasis of a certain cell and several data sequences without distant metastasis of a certain cell, and form a training set;

The feature selection unit is configured to: input the training set into k feature selection algorithms, respectively select the top p attributes as the features of the training set, input the classifier for training, compare the classification results, and select the result. the best p features;

The oversampling unit is configured to: make the training set achieve data balance at the data level based on the oversampling algorithm, input the training set processed by the feature selection algorithm into n oversampling algorithms, and obtain a positive and negative sample ratio of 1: 1 training set;

The optimal oversampling algorithm obtaining unit is configured to: input the training set with the positive and negative class sample ratio of 1:1 into the classification algorithm, and then use the data sequence of the test set for testing, and select the training set with the best evaluation result. Pi's oversampling algorithm i;

The optimal positive and negative sample ratio acquisition unit is configured to: by adjusting the ratio of positive and negative samples, input the training set to the oversampling algorithm to obtain the training set Pi, and gradually increase the ratio of positive and negative samples to a set ratio, The optimal proportion of positive and negative class samples for classification evaluation.

Compared with the prior art, the beneficial effects of the present disclosure are:

1. The data sequence in the technical solution of the present disclosure can be the blood cell analysis data of the routine examination in the hospital, and the acquisition of the data is relatively easy in terms of technical realization, which is convenient for subsequent data feature selection and oversampling processing.

2. The technical solution of the present disclosure uses an oversampling algorithm to try to increase the proportion of positive samples, so as to obtain better model evaluation indicators and the recall rate of a small number of positive samples.

Description of drawings

The accompanying drawings that constitute a part of the present disclosure are used to provide further understanding of the present disclosure, and the exemplary embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation of the present disclosure.

1 is a flowchart of a method for classifying distant metastasis of abnormal cells based on non-equilibrium learning according to an embodiment of the disclosure;

2 is a schematic diagram of a feature selection strategy for a method for classifying distant metastasis of abnormal cells based on non-equilibrium learning according to an embodiment of the present disclosure;

3 is a schematic diagram of an oversampling algorithm selection strategy for an abnormal cell distant metastasis classification method based on unbalanced learning according to an embodiment of the present disclosure;

FIG. 4 is a schematic diagram of a strategy for adjusting the proportion of positive and negative samples in a method for classifying distant metastasis of abnormal cells based on unbalanced learning according to an embodiment of the present disclosure.

Detailed ways

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the present disclosure. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural as well, furthermore, it is to be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates that There are features, steps, operations, devices, components and/or combinations thereof.

At present, there are two main solutions for dealing with the problem of unbalanced data classification: first, balanced data. At the data level, using appropriate methods to reconstruct training samples, over-sampling or under-sampling algorithms can be used to achieve data balance; second, improve or propose new algorithms. At the algorithm level, the existing classification algorithms are used to improve or new classification algorithms are proposed to make the samples of the minority class get more attention and improve the accuracy of the samples of the minority class. The technical solution of the embodiments of the present application is the first one, which balances data samples at the data level.

Example 1

This example discloses a method for classifying distant metastasis of abnormal cells based on unbalanced learning. The present disclosure first screens the available data sets. Diagnostic information, screen out the patients with clinical M stage for the first time from the diagnosis table. A clinical M stage of 0 indicates that the patient does not have cancer cells metastasized to other organs, and the label is set to 0, and a clinical M stage of non-0 indicates that the patient If cancer cells metastasize to other organs, set the label to 1, and then select the blood cell analysis and test data before the surgical treatment according to the operation time recorded in the operation table of these patients. If no surgical treatment is performed, select the diagnosis stage Blood cell analysis data on the day or before the time of diagnosis. The data on which this application is based are all patient data, and therefore, have nothing to do with diagnosis and treatment, but only predict the transfer of relevant cells based on relevant data.

In one embodiment, the filtered available samples are divided into 75% training set and 25% test set, the training set is input into a variety of feature selection methods, the top 8 attributes are selected as the features of the data set, and then the classification is input. The output model evaluation index AUC and recall are compared with the output results of the original situation, and the features with the best results are selected.

In this embodiment, by training two types of data through a classifier, the classification characteristics of the two types of data can be learned, and new data can be input, and the classifier can automatically identify which category belongs to.

The output model evaluation index AUC and recall are explained as follows:

AUC (Area Under Curve) is defined as the area enclosed by the coordinate axis under the ROC curve. Obviously, the value of this area will not be greater than 1. Also, since the ROC curve is generally above the straight line y=x, the value range of AUC is between 0.5 and 1. The AUC value is used as the evaluation criterion because in many cases the ROC curve does not clearly indicate which classifier is better, and as a value, the classifier with a larger AUC is better.

Recall, also known as recall rate (TPR), is a measure of the integrity of the classification results of imbalanced data, indicating the ratio of the actual number of minority class samples to the actual number of minority class samples.

Then, using the selected features, input different oversampling algorithms, so that the sample ratio of positive and negative classes is 1:1, input the classifier, and compare the output model evaluation index with the recall rate, and select the best result. oversampling algorithm.

Then, using the selected oversampling algorithm, try to increase the proportion of positive samples, from 1.1:1, 1.2:1 to 2:1, and then give two ratios with a large difference between 5:1 and 10:1 , compare the results, and select the appropriate proportion of positive and negative samples.

In specific implementation, referring to Figure 1, the method for classifying distant metastasis of abnormal cells based on non-equilibrium learning includes:

Step (1): Screen the data set, clean the dirty data after screening, delete the samples with missing data directly, and the red blood cell distribution width (CV) attribute is really serious, delete this column of attribute data before feature selection, Leave the complete dataset.

Specifically, the data set includes a training set and a test set, and the data in the training set includes several blood cell analysis data sequences with distant metastasis of cancer cells, and several blood cell analysis data sequences with no distant metastasis of cancer cells.

The test set stores the sequence of blood cell analysis data to be tested.

The data sequence is the blood cell analysis data sequence including: white blood cell count, lymphocyte absolute value, lymphocyte percentage, neutrophil absolute value, neutrophil percentage, monocyte absolute value, monocyte percentage, eosinophil absolute value value, percent eosinophils, absolute basophils, percent basophils, red blood cell count, hemoglobin, mean red blood cell volume, mean red blood cell hemoglobin content, mean red blood cell hemoglobin concentration, red blood cell distribution width (CV), platelets Count, platelet distribution width, platelet distribution volume, mean platelet volume.

Step (2): As shown in Figure 2, feature selection is performed on blood cell analysis, the data set is input into k feature selection algorithms, respectively, the top p attributes are selected as the features of data set A, and the classifier is input to carry out Training, output model evaluation metrics (G-Mean, AUC) and recall rate (Recall).

In this embodiment, the model here is the aforementioned classifier.

Because minority samples are the focus of attention, according to the specific analysis of specific problems, the thresholds of the evaluation indicators G-Mean and AUC of the calculation model are re-specified, and weighted G-Mean and weighted AUC are proposed, which are denoted as WG-Mean and WAUC; WG-Mean and WAUC are calculated by the calculation method of , and compared with the original situation, the p features with the best results are selected. The larger the WG-Mean and WAUC calculated here, the better the result. The following inputs will be used This p feature.

Because in this case, the minority class samples are the focus, the re-specification of this threshold is to improve the ratio of the recall rate in the calculation.

In this embodiment, G-Mean and AUC are two indicators for comprehensive evaluation of the classifier.

The calculation formula after re-given threshold:

WAUC=Sensitivity×0.7+Specificity×0.3;

In this embodiment, the data of the original situation is the data that has not been balanced for positive and negative samples.

In the specific implementation, the top 8 attributes are selected as the basic features through the results obtained by the algorithm. Through analysis, the selected features include: platelet distribution width, lymphocyte percentage, absolute lymphocyte value, neutrophil percentage, mean platelet volume, red blood cell count, hemoglobin, and hematocrit.

Referring to Fig. 3, an oversampling algorithm with better results is selected. Steps (3) and (4): Based on the oversampling algorithm, the data set achieves data balance at the data level, and the selected 8 features are input into different oversampling algorithms to obtain a positive and negative sample ratio of 1. : 1 training set P1, P2...Pn, input the training set P1, P2...Pn to the classification algorithm respectively, and then use the test set N for testing, output the model evaluation index (G-Mean, AUC) and recall rate ( Recall); calculate WG-Mean and WAUC, and select the oversampling algorithm i corresponding to the best result to obtain the training set Pi.

In this example, the data set is divided into a training set and a test set, the training set is used to train the model, and the test set is used to test the model; first, the training set is input into the feature selection algorithm and the result of the original data set classification is compared For comparison, select the p features with the best results; then use the oversampling algorithm to obtain a training set with a positive and negative sample ratio of 1:1.

Step (5): Referring to Figure 4, by adjusting the ratio of positive and negative samples, in order to obtain a higher recall rate and a better model evaluation index, the training set M is input into the oversampling algorithm to obtain the training set Pi, and gradually Increase the ratio of positive and negative samples to 1.1:1, 1.2:1, until 2:1, and even give 5:1, 10:1 output Recall, WG-Mean and WAUC, and select the positive result with the best result Negative class sample proportion.

Selecting the proportion of positive and negative samples with the best results can make the model evaluation index reach the best.

In the embodiment of the present disclosure, blood cell analysis in routine examinations in hospitals is used for analysis and prediction, in order to replace other expensive and time-consuming diagnostic methods. It has certain innovation in application.

The technology used in the embodiments of the present disclosure breaks through the weakness that clinical medical researchers do not understand machine learning, and breaks the traditional conventional P-test analysis method.

The embodiments of the present disclosure redefine the threshold of the evaluation index of the calculation model in combination with the specific practical significance.

The embodiments of the present disclosure use an oversampling algorithm to try to increase the proportion of positive samples, so as to obtain better model evaluation indicators and a recall rate of a small number of positive samples.

Example 2

Embodiments of the present specification provide a non-equilibrium learning-based cell distant metastasis classification system, including:

The training set acquisition unit is configured to: obtain several data sequences with distant metastasis of a certain cell and several data sequences without distant metastasis of a certain cell, and form a training set;

The feature selection unit is configured to: input the training set into k feature selection algorithms, respectively select the top p attributes as the features of the training set, input the classifier for training, compare the classification results, and select the result. the best p features;

The oversampling unit is configured to: make the training set achieve data balance at the data level based on the oversampling algorithm, input the training set processed by the feature selection algorithm into n oversampling algorithms, and obtain a positive and negative sample ratio of 1: 1 training set;

The optimal oversampling algorithm obtaining unit is configured to: input the training set with the positive and negative class sample ratio of 1:1 into the classification algorithm, and then use the data sequence of the test set for testing, and select the training set with the best evaluation result. Pi's oversampling algorithm i;

The optimal positive and negative sample ratio acquisition unit is configured to: by adjusting the ratio of positive and negative samples, input the training set to the oversampling algorithm to obtain the training set Pi, and gradually increase the ratio of positive and negative samples to a set ratio, The optimal proportion of positive and negative class samples for classification evaluation.

In another embodiment, during the specific implementation of the above-mentioned system, a server, a data input device and a data display may be used, and the data input device may be used to input the analysis data of blood cells to the server, or the blood cell data stored in the memory may be processed by invoking. After the server performs the above-mentioned processing on the data, the specific result and the relevant data in the data processing process are displayed by the display.

The server includes a training set collection unit, a feature selection unit, an oversampling unit, an optimal oversampling algorithm obtaining unit and an optimal positive and negative class sample ratio obtaining unit.

For the specific implementation process of the above unit, reference may be made to the specific process in Embodiment 1, which will not be described in detail here.

Example three

An embodiment of the present disclosure discloses a computer device, including 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 program, unbalanced learning is implemented The cell distant metastases classification steps.

In this embodiment, for specific steps, refer to the detailed process of Embodiment 1, which will not be described in detail here.

It should be noted that although several modules or sub-modules of the apparatus are mentioned in the detailed description above, this division is merely exemplary and not mandatory. Indeed, in accordance with embodiments of the present disclosure, the features and functions of two or more modules described above may be embodied in one module. Conversely, the features and functions of one module described above can be further divided into multiple modules to be embodied.

Example 4

An embodiment of the present disclosure discloses a computer-readable storage medium on which a computer program is stored, characterized in that, when the program is executed by a processor, the step of classifying distant metastasis of cells based on unbalanced learning is implemented.

In this embodiment, for specific steps, refer to the detailed process of Embodiment 1, which will not be described in detail here.

In this embodiment, the computer program product may comprise a computer-readable storage medium having computer-readable program instructions loaded thereon for carrying out various aspects of the present disclosure. A computer-readable storage medium may be a tangible device that can hold and store instructions for use by the instruction execution device.

It is to be understood that, in the description of this specification, referring to the description of the terms "an embodiment", "another embodiment", "other embodiment", or "the first embodiment to the Nth embodiment" etc. means A particular feature, structure, material, or characteristic described in connection with this embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials and characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

The above descriptions are only preferred embodiments of the present disclosure, and are not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included within the protection scope of the present disclosure.

Claims (9)

1. The abnormal cell distant metastasis classification method based on unbalanced learning is characterized by comprising the following steps:

obtaining a plurality of data sequences with certain cells having distant metastasis and a plurality of data sequences without certain cells having distant metastasis, and forming a training set;

respectively inputting the training set into k feature selection algorithms, respectively selecting p attributes ranked in the front as features of the training set, inputting the attributes into a classifier for training, comparing classification results, and selecting p features with the best results;

enabling the training set to achieve data balance on a data level based on an oversampling algorithm, and inputting the training set processed by the feature selection algorithm into n oversampling algorithms to obtain a training set with a positive-negative sample ratio of 1: 1;

respectively inputting the training sets with the positive and negative sample ratio of 1:1 into a classification algorithm, testing by using the data sequence of the test set, and selecting the training set with the optimal evaluation result_PiThe oversampling algorithm i of (1);

inputting the training set into the obtained training set by adjusting the proportion of the positive and negative samples_PiThe over-sampling algorithm gradually increases the proportion of the positive and negative samples to a set proportion, and selects the proportion of the positive and negative samples with the optimal classification evaluation.

2. The method of classifying abnormal cell distant metastasis based on unbalanced learning according to claim 1, wherein the data sequence comprises: leukocyte count, lymphocyte absolute value, lymphocyte percentage, neutrophil absolute value, neutrophil percentage, monocyte absolute value, monocyte percentage, eosinophil absolute value, eosinophil percentage, basophil absolute value, basophil percentage, erythrocyte count, hemoglobin, erythrocyte mean volume, erythrocyte mean hemoglobin content, erythrocyte mean hemoglobin concentration, erythrocyte distribution width, platelet count, platelet distribution width, platelet distribution volume, and platelet mean volume.

3. The method of claim 1, wherein the selecting the p features with the best results comprises: width of platelet distribution, lymphocyte percentage, absolute value of lymphocytes, neutrophil percentage, mean volume of platelets, red blood cell count, hemoglobin, and hematocrit.

4. The abnormal cell distant metastasis classification method based on unbalanced learning as claimed in claim 1, wherein the data of the training set is subjected to data screening before feature selection, and the data integrity is judged, and the samples containing the missing data are deleted.

5. A cell distant metastasis classification system based on unbalanced learning is characterized by comprising:

a training set acquisition unit configured to: obtaining a plurality of data sequences with certain cells having distant metastasis and a plurality of data sequences without certain cells having distant metastasis, and forming a training set;

a feature selection unit configured to: respectively inputting the training set into k feature selection algorithms, respectively selecting p attributes ranked in the front as features of the training set, inputting the attributes into a classifier for training, comparing classification results, and selecting p features with the best results;

an oversampling unit configured to: enabling the training set to achieve data balance on a data level based on an oversampling algorithm, and inputting the training set processed by the feature selection algorithm into n oversampling algorithms to obtain a training set with a positive-negative sample ratio of 1: 1;

an optimal oversampling algorithm obtaining unit configured to: respectively inputting the training sets with the positive and negative sample ratio of 1:1 into a classification algorithm, testing by using the data sequence of the test set, and selecting the training set with the optimal evaluation result_PiThe oversampling algorithm i of (1);

an optimal positive and negative sample proportion obtaining unit configured to: inputting the training set into the obtained training set by adjusting the proportion of the positive and negative samples_PiThe over-sampling algorithm gradually increases the proportion of the positive and negative samples to a set proportion, and selects the proportion of the positive and negative samples with the optimal classification evaluation.

6. The system of claim 5, wherein the selecting the p features that yield the best results comprises: width of platelet distribution, lymphocyte percentage, absolute value of lymphocytes, neutrophil percentage, mean volume of platelets, red blood cell count, hemoglobin, and hematocrit.

7. The cell remote transfer classification system based on unbalanced learning is characterized by comprising a server, a data input device and a data display, wherein the data input device is used for inputting analysis data of blood cells into the server or calling the blood cell data stored in a memory in a calling mode, and the display is used for displaying specific results and related data in a data processing process;

the server is configured to include:

a training set acquisition unit configured to: obtaining a plurality of data sequences with certain cells having distant metastasis and a plurality of data sequences without certain cells having distant metastasis, and forming a training set;

a feature selection unit configured to: respectively inputting the training set into k feature selection algorithms, respectively selecting p attributes ranked in the front as features of the training set, inputting the attributes into a classifier for training, comparing classification results, and selecting p features with the best results;

an oversampling unit configured to: enabling the training set to achieve data balance on a data level based on an oversampling algorithm, and inputting the training set processed by the feature selection algorithm into n oversampling algorithms to obtain a training set with a positive-negative sample ratio of 1: 1;

an optimal oversampling algorithm obtaining unit configured to: respectively inputting the training sets with the positive and negative sample ratio of 1:1 into a classification algorithm, testing by using the data sequence of the test set, and selecting the training set with the optimal evaluation result_PiThe oversampling algorithm i of (1);

an optimal positive and negative sample proportion obtaining unit configured to: inputting the training set into the obtained training set by adjusting the proportion of the positive and negative samples_PiThe over-sampling algorithm gradually increases the proportion of the positive and negative samples to a set proportion, and selects the proportion of the positive and negative samples with the optimal classification evaluation.

8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for classifying distant metastasis of abnormal cells based on unbalanced learning according to any one of claims 1 to 4.

9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of a method for classifying distant metastasis of abnormal cells based on unbalanced learning according to any one of claims 1 to 4.

Images