An unsupervised partial adversarial adaptation method and system for non-identical label spaces

Abstract

The present invention relates to the field of electrocardiogram analysis technology, and discloses an unsupervised partial adversarial adaptive method and system for non-identical label spaces, the method comprising: constructing a deep learning diagnostic model, a feature extractor, a feature classifier, and a fully connected feature processor; constructing an entropy minimization loss function based on the source domain classification loss; calculating a selection factor, and adjusting the weight of the source domain data in the diagnostic model training according to the selection factor; constructing a final loss function during the diagnostic model training process, and optimizing the feature extractor, feature classifier, and fully connected feature processor by minimizing the final loss function. Through the above method, an unsupervised partial adversarial adaptive system for non-identical label spaces is realized. By introducing the EL loss function, the selection factor, and the final loss function to perform adversarial training on the model, the cross-domain migration performance of the deep learning model can be effectively improved, and the diagnostic performance of the model in non-identical label spaces can be guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of electrocardiogram analysis, and in particular to an unsupervised partial adversarial adaptive method and system for non-identical label spaces. Background Art

[0002] Timely and intelligent diagnosis of arrhythmias is crucial for the control of cardiovascular disease. According to the World Health Organization (WHO), cardiovascular disease is the leading cause of death worldwide, accounting for approximately one-third of all deaths each year. Arrhythmias are the most common cardiovascular disease. As a non-invasive diagnostic tool, the electrocardiogram (ECG) is one of the most effective tools for doctors to identify arrhythmias. In fact, most arrhythmias can be diagnosed using an ECG alone. However, analyzing every heartbeat is necessary to detect cardiac abnormalities, a cumbersome and time-consuming process. Interpreting ECG recordings is also challenging and is typically reserved for cardiologists. Furthermore, different patients may experience different patterns for the same disease.

[0003] Deep learning methods have been widely used in the diagnosis of arrhythmias. However, due to the difficulty of sample labeling, the number of samples available for classification model learning is very limited. Furthermore, existing deep learning models still face the following challenges in the field of ECG diagnosis: (i) Training a good deep network requires a large amount of labeled data. In clinical applications, collecting sufficient labeled ECG recordings is often time-consuming and expensive. (ii) The differences between ECG signals are small, which seriously affects the generalization ability of the diagnostic model. In addition, different factors such as age and acquisition equipment can also affect the diagnostic performance of the model. Therefore, in a novel scenario with only many unlabeled ECG recordings, the diagnostic performance without labeled data will be significantly reduced. (iii) Deep domain adaptation, a common transfer learning method, has recently received widespread attention in addressing cross-domain problems. However, these studies have only focused on multi-center datasets with the same label space. Due to the diverse types of arrhythmias, the label spaces of different datasets in real applications are often different. Faced with the different label spaces between datasets in different domains, existing cross-domain arrhythmia diagnosis methods have not been widely studied. Therefore, it is necessary to design a method to optimize model training so that the model can maintain the best performance in different label spaces. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the above-mentioned prior art and to provide an unsupervised partial adversarial adaptation method and system for non-identical label spaces, so as to achieve the migration of pre-trained models to new datasets and improve the diagnostic performance of the models.

[0005] On the one hand, an unsupervised partial adversarial adaptation method in a non-identical label space is provided, comprising the following steps:

[0006] S1: Collect a clinical electrocardiogram dataset in a certain label space, build a deep learning diagnostic model, and simultaneously build a feature extractor, feature classifier, and fully connected feature processor based on the diagnostic model;

[0007] S2: Input the source domain electrocardiogram into the feature classifier to obtain the source domain classification loss, and feed the data in the unlabeled dataset into the feature classifier to construct an entropy minimization loss function;

[0008] S3: Calculate a selection factor based on the source data set input into the diagnostic model and the newly input data set, and adjust the weight of the source domain data in the diagnostic model training according to the selection factor;

[0009] S4: Constructing a final loss function in the diagnostic model training process, and optimizing the feature extractor, feature classifier, and fully connected feature processor by minimizing the final loss function.

[0010] Furthermore, in step S1, constructing a feature extractor, a feature classifier, and a fully connected feature processor based on the diagnostic model further includes:

[0011] Selecting the neural network structure used by the feature extractor and the feature classifier according to the actual processing task and data characteristics;

[0012] The fully connected feature processor is constructed from fully connected layers.

[0013] Preferably, the classification process based on the feature extractor and the fully connected feature processor specifically includes:

[0014] S100: ECG data is input into the feature extractor, and the output features are flattened by the fully connected feature processor to convert the multi-dimensional feature vector into a one-dimensional vector;

[0015] S200: The flattened features are input to the subsequent fully connected layer for nonlinear processing. The fully connected layer introduces nonlinear factors to enhance the nonlinear feature expression capability of the features.

[0016] S300: The features after nonlinear processing are sent to a binary classifier, and the binary classifier selects a suitable algorithm for classification according to specific task requirements.

[0017] More preferably, a reverse layer is included between the feature extractor and the fully connected feature processor:

[0018] In the forward classification process, the reverse layer does not play a role;

[0019] During the back-propagation process, the reverse layer reverses the back-propagated gradient, and the feature extractor receives gradient information different from normal back-propagation, forcing it to learn feature representations that confuse the fully connected feature processor.

[0020] Furthermore, in step S2, inputting the source domain electrocardiogram into the feature classifier to obtain the source domain classification loss further includes:

[0021] The source domain classification loss CL formula measures the classification loss of the source domain ECG feature in the feature classifier by calculating the sum of the logarithmic probabilities of the sample being predicted to belong to a certain class and not to belong to a certain class. The formula is as follows:

[0022]

[0023] in, Indicates that the feature classifier is used for input data The processed output belongs to the label probability prediction value, The closer to the true label, the smaller the loss value.

[0024] Furthermore, in step S2, feeding the data in the unlabeled dataset to the feature classifier to construct an entropy minimization loss function further includes:

[0025] First, the log-likelihood value corresponding to the probability that the sample belongs to a certain category is obtained according to the source domain classification loss formula , among which, if , ,think ;like , ,think ;

[0026] The entropy minimization principle is used to measure the classification effect of the feature classifier on unlabeled data, so as to construct the entropy minimization loss function EL. The probability distribution is more concentrated, the larger the loss function value is, the When the probability distribution is relatively uniform, the loss function value is small, and the formula is as follows:

[0027] .

[0028] Furthermore, in step S3, calculating the selection factor based on the source data set input to the diagnostic model and the newly input data set further includes:

[0029] The selection factor is used to reduce the number of label classes in the forward classification process. The contribution of the corresponding source domain data. At the same time, in the back propagation process, the obtained contribution is in turn applied to the source domain dataset samples to reduce the label class The influence of the source domain data on data calibration during the training process, the selection factor calculation process includes the following formulas in sequence:

[0030]

[0031]

[0032]

[0033]

[0034] in, Represents input data, is a feature classifier, Represents the output result after the data is input into the feature classifier, represents the source dataset, represents a new dataset, Indicates the source dataset category, Represents the new dataset category, Represents the selection factor.

[0035] Furthermore, in step S4, constructing the final loss function in the diagnostic model training process further includes:

[0036] The final loss function consists of four parts, involving different data sets, loss functions and parameters, and the formula is as follows:

[0037] ,

[0038] in, , λ represents the weight coefficient, is the feature classifier, is the feature extractor, is the fully connected feature processor, represents the entropy minimization loss function of the feature classifier, represents the entropy minimization loss function of the fully connected feature processor, represents the selection factor, Represents samples taken from the source dataset , Indicates taking samples from the new dataset , Indicates the source dataset category, Represents the new dataset category, Indicates that the feature classifier is used for input data The processed output belongs to the label probability prediction value.

[0039] On the other hand, an unsupervised partial adversarial adaptive system in a non-identical label space is provided, including:

[0040] A model building module is used to collect a clinical electrocardiogram dataset in a certain label space, build a deep learning diagnostic model, and simultaneously build a feature extractor, feature classifier, and fully connected feature processor based on the diagnostic model;

[0041] A classification loss optimization module is used to input the source domain electrocardiogram into the feature classifier to obtain the source domain classification loss, and feed the data in the unlabeled data set into the feature classifier to construct an entropy minimization loss function;

[0042] A selection factor calibration module is used to calculate a selection factor based on the source data set input into the diagnostic model and the newly input data set, and adjust the weight of the source domain data in the diagnostic model training according to the selection factor;

[0043] The model training module is used to construct a final loss function in the diagnostic model training process and optimize the feature extractor, feature classifier and fully connected feature processor by minimizing the final loss function.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] This paper optimizes the loss of the feature extractor, feature classifier, and fully connected feature processor of the diagnostic model, introduces the EL loss function, selection factor, and final loss function, and effectively improves the cross-domain migration performance of the deep learning model. When a dataset in a completely new label space is input, the high performance of the diagnostic model is still maintained.

[0046] The present invention selects the neural network structure adopted by the feature extractor and the feature classifier according to the actual processing task and data characteristics, and is flexibly applicable to commonly used neural network models in the world. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0048] Figure 1 This is a flow chart of an unsupervised partial adversarial adaptive method for non-identical label spaces of the present invention;

[0049] Figure 2 This is a classification flow chart based on a feature extractor and a fully connected feature processor of the present invention. DETAILED DESCRIPTION

[0050] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0051] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0052] The following describes the specific implementation of the present invention with reference to the accompanying drawings and embodiments.

[0053] like Figure 1 As shown, this embodiment provides an unsupervised partial adversarial adaptation method for non-identical label spaces, and the technical solution includes the following steps:

[0054] S1: Collect a clinical ECG dataset in a certain label space, build a deep learning diagnostic model, and build a feature extractor based on the diagnostic model. , feature classifier and fully connected feature processors ;

[0055] S2: Input the source domain electrocardiogram into the feature classifier Obtain the source domain classification loss CL and feed the data in the unlabeled dataset to the feature classifier Construct entropy minimization loss function EL;

[0056] S3: Calculate a selection factor based on the source data set input into the diagnostic model and the newly input data set, and adjust the weight of the source domain data in the diagnostic model training according to the selection factor;

[0057] S4: Constructing the final loss function in the diagnosis model training process, and optimizing the feature extractor by minimizing the final loss function , feature classifier and fully connected feature processors .

[0058] In step S1, a feature extractor based on the diagnostic model is constructed. , feature classifier and fully connected feature processors Further including:

[0059] Here and Any commonly used structure in the neural network framework (such as CNN, LSTM) can be used without restriction, and the feature extractor can be selected according to the actual processing task and data characteristics. and the feature classifier The neural network structure used;

[0060] Secondly, the fully connected feature processor It is constructed from fully connected layers.

[0061] Specifically, based on the feature extractor and the fully connected feature processor The classification process is as follows Figure 2 As shown, specifically including:

[0062] S100: ECG data is input into the feature extractor , the output features are passed through the fully connected feature processor Perform flattening to convert the multi-dimensional feature vector into a one-dimensional vector;

[0063] S200: The flattened features are input to the subsequent fully connected layer for nonlinear processing. The fully connected layer introduces nonlinear factors to enhance the nonlinear feature expression capability of the features.

[0064] S300: The features after nonlinear processing are sent to a binary classifier, and the binary classifier selects a suitable algorithm for classification according to specific task requirements.

[0065] In this embodiment, the feature extractor and the fully connected feature processor Include a reverse layer in between:

[0066] In the forward classification process, the reverse layer does not play a role;

[0067] During the back propagation process, the reverse layer reverses the back propagated gradient, and the feature extractor Receives gradient information different from normal back propagation, forcing it to learn features that confuse the fully connected feature processor feature representation.

[0068] In step S2, the source domain electrocardiogram is input into the feature classifier Obtaining the source domain classification loss CL further includes:

[0069] The source domain classification loss CL formula measures the classification loss of the source domain ECG feature in the feature classifier by calculating the sum of the logarithmic probabilities of the sample being predicted to belong to a certain class and not to belong to a certain class. The formula is as follows:

[0070]

[0071] in, Represents the feature classifier For input data The processed output belongs to the label probability prediction value, The closer to the true label, the smaller the loss value.

[0072] Next, in order to further optimize the feature classifier Parameters to make it have better generalization ability, feeding the data in the unlabeled dataset to , using the entropy minimization principle, Effectively separate the target domain data to make The classification of is clearer. We use this to construct the entropy minimization loss function EL.

[0073] Specifically, in this embodiment, we first obtain the log-likelihood value corresponding to the probability that the sample belongs to a certain category according to the source domain classification loss formula: , among which, if , ,think ;like , ,think ;

[0074] Then, the entropy minimization principle is used to measure the classification effect of the feature classifier on the unlabeled data, so as to construct the entropy minimization loss function EL. The probability distribution is more concentrated, the larger the loss function value is, the When the probability distribution is relatively uniform, the loss function value is small, and the formula is as follows:

[0075] .

[0076] In step S3, calculating the selection factor based on the source data set input into the diagnostic model and the newly input data set further includes:

[0077] The selection factor is used to reduce the number of label classes in the forward classification process. The contribution of the corresponding source domain data. At the same time, in the back propagation process, the obtained contribution is in turn applied to the source domain dataset samples to reduce the label class The influence of the source domain data on data calibration during the training process, the selection factor calculation process includes the following formulas in sequence:

[0078]

[0079]

[0080]

[0081]

[0082] in, Represents input data, is the feature classifier, Represents the output result after the data is input into the feature classifier, represents the source dataset, represents a new dataset, Indicates the source dataset category, Represents the new dataset category, Represents the selection factor.

[0083] Finally, we construct the final loss function in S4 based on all the above steps, including:

[0084] The final loss function consists of four parts, involving different data sets, loss functions and parameters, and the formula is as follows:

[0085] ,

[0086] in, , λ represents the weight coefficient, is the feature classifier, is the feature extractor, is the fully connected feature processor, represents the entropy minimization loss function of the feature classifier, represents the entropy minimization loss function of the fully connected feature processor, represents the selection factor, Represents samples taken from the source dataset , Indicates taking samples from the new dataset , Indicates the source dataset category, Represents the new dataset category, Indicates that the feature classifier is used for input data The processed output belongs to the label probability prediction value.

[0087] This embodiment provides an unsupervised partial adversarial adaptive system in a non-identical label space, including:

[0088] A model building module is used to collect a clinical electrocardiogram dataset in a certain label space, build a deep learning diagnostic model, and simultaneously build a feature extractor, feature classifier, and fully connected feature processor based on the diagnostic model;

[0089] A classification loss optimization module is used to input the source domain electrocardiogram into the feature classifier to obtain the source domain classification loss, and feed the data in the unlabeled data set into the feature classifier to construct an entropy minimization loss function;

[0090] A selection factor calibration module is used to calculate a selection factor based on the source data set input into the diagnostic model and the newly input data set, and adjust the weight of the source domain data in the diagnostic model training according to the selection factor;

[0091] The model training module is used to construct a final loss function in the diagnostic model training process and optimize the feature extractor, feature classifier and fully connected feature processor by minimizing the final loss function.

[0092] The functional implementation of each module in the above-mentioned unsupervised partial adversarial adaptive system for non-identical label spaces corresponds to the steps of the unsupervised partial adversarial adaptive method for non-identical label spaces in Example 1, and their functions and implementation processes will not be repeated here one by one.

[0093] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that improvements and modifications that do not depart from the principles of the present invention, which are apparent to those skilled in the art, should also be considered within the scope of protection of the present invention.

[0094] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. An unsupervised partial adversarial adaptation method in non-identical label spaces, characterized by: The steps include: S1: Collect a clinical electrocardiogram dataset in a certain label space, build a deep learning diagnostic model, and simultaneously build a feature extractor, feature classifier, and fully connected feature processor based on the diagnostic model; S2: Input the source domain electrocardiogram into the feature classifier to obtain the source domain classification loss, and feed the data in the unlabeled dataset into the feature classifier to construct an entropy minimization loss function; S3: Calculate a selection factor based on the source data set input into the diagnostic model and the newly input data set, and adjust the weight of the source domain data in the diagnostic model training according to the selection factor; S4: constructing a final loss function in the diagnostic model training process, and optimizing the feature extractor, feature classifier, and fully connected feature processor by minimizing the final loss function; In step S2, inputting the source domain electrocardiogram into the feature classifier to obtain the source domain classification loss further includes: The source domain classification loss CL formula measures the classification loss of the source domain ECG feature in the feature classifier by calculating the sum of the logarithmic probabilities of the sample being predicted to belong to a certain class and not to belong to a certain class. The formula is as follows: in, Indicates that the feature classifier is used for input data The processed output belongs to the label probability prediction value, The closer to the true label, the smaller the loss value; Secondly, feeding the data in the unlabeled dataset to the feature classifier to construct an entropy minimization loss function further includes: First, the log-likelihood value corresponding to the probability that the sample belongs to a certain category is obtained according to the source domain classification loss CL formula , among which, if , ,think ;like , ,think ; The entropy minimization principle is used to measure the classification effect of the feature classifier on unlabeled data, so as to construct the entropy minimization loss function EL. The probability distribution is more concentrated, the larger the loss function value is, the more likely it is that When the probability distribution is relatively uniform, the loss function value is small, and the formula is as follows: In step S3, calculating the selection factor based on the source data set input into the diagnostic model and the newly input data set further includes: The selection factor is used to reduce the number of label classes in the forward classification process. The contribution of the corresponding source domain data. At the same time, in the back propagation process, the obtained contribution is in turn applied to the source domain dataset samples to reduce the label class The influence of the source domain data on data calibration during the training process, the selection factor calculation process includes the following formulas in sequence: in, Represents input data, is the feature classifier, Represents the output result after the data is input into the feature classifier, represents the source dataset, represents a new dataset, Indicates the source dataset category, Represents the new dataset category, represents the selection factor; In step S4, constructing the final loss function in the diagnostic model training process further includes: The final loss function consists of four parts, involving different data sets, loss functions and parameters, and the formula is as follows: in, , λ represents the weight coefficient, is the feature classifier, is the feature extractor, is the fully connected feature processor, represents the entropy minimization loss function of the feature classifier, represents the entropy minimization loss function of the fully connected feature processor, represents the selection factor, Represents samples taken from the source dataset , Indicates taking samples from the new dataset , Indicates the source dataset category, Represents the new dataset category, Indicates that the feature classifier is used for input data The processed output belongs to the label probability prediction value.

2. The unsupervised partial adversarial adaptive method for non-identical label spaces according to claim 1, characterized in that: In step S1, constructing a feature extractor, a feature classifier, and a fully connected feature processor based on the diagnostic model further includes: Selecting the neural network structure used by the feature extractor and the feature classifier according to the actual processing task and data characteristics; The fully connected feature processor is constructed from fully connected layers.

3. The unsupervised partial adversarial adaptive method for non-identical label spaces according to claim 2, characterized in that: The classification process based on the feature extractor and the fully connected feature processor specifically includes: S100: ECG data is input into the feature extractor, and the output features are flattened by the fully connected feature processor to convert the multi-dimensional feature vector into a one-dimensional vector; S200: The flattened features are input to the subsequent fully connected layer for nonlinear processing. The fully connected layer introduces nonlinear factors to enhance the nonlinear feature expression capability of the features. S300: The features after nonlinear processing are sent to a binary classifier, and the binary classifier selects a suitable algorithm for classification according to specific task requirements.

4. The unsupervised partial adversarial adaptive method for non-identical label spaces according to claim 3, characterized in that: A reverse layer is included between the feature extractor and the fully connected feature processor: In the forward classification process, the reverse layer does not play a role; During the back-propagation process, the reverse layer reverses the back-propagated gradient, and the feature extractor receives gradient information different from normal back-propagation, forcing it to learn feature representations that confuse the fully connected feature processor.

5. An unsupervised partial adversarial adaptive system in a non-identical label space, characterized in that: include: A model building module is used to collect a clinical electrocardiogram dataset in a certain label space, build a deep learning diagnostic model, and simultaneously build a feature extractor, feature classifier, and fully connected feature processor based on the diagnostic model; A classification loss optimization module is used to input the source domain electrocardiogram into the feature classifier to obtain the source domain classification loss, and feed the data in the unlabeled data set into the feature classifier to construct an entropy minimization loss function; A selection factor calibration module is used to calculate a selection factor based on the source data set input into the diagnostic model and the newly input data set, and adjust the weight of the source domain data in the diagnostic model training according to the selection factor; A model training module, configured to construct a final loss function during the diagnostic model training process, and optimize the feature extractor, feature classifier, and fully connected feature processor by minimizing the final loss function; Wherein, inputting the source domain electrocardiogram into the feature classifier to obtain the source domain classification loss further includes: The source domain classification loss CL formula measures the classification loss of the source domain ECG feature in the feature classifier by calculating the sum of the logarithmic probabilities of the sample being predicted to belong to a certain class and not to belong to a certain class. The formula is as follows: in, Indicates that the feature classifier is used for input data The processed output belongs to the label probability prediction value, The closer to the true label, the smaller the loss value; Secondly, feeding the data in the unlabeled dataset to the feature classifier to construct an entropy minimization loss function further includes: First, the log-likelihood value corresponding to the probability that the sample belongs to a certain category is obtained according to the source domain classification loss CL formula , among which, if , ,think ;like , ,think ; The entropy minimization principle is used to measure the classification effect of the feature classifier on unlabeled data, so as to construct the entropy minimization loss function EL. The probability distribution is more concentrated, the larger the loss function value is, the more likely it is that When the probability distribution is relatively uniform, the loss function value is small, and the formula is as follows: ； The selection factors calculated based on the source data set input into the diagnostic model and the newly input data set further include: The selection factor is used to reduce the number of label classes in the forward classification process. The contribution of the corresponding source domain data. At the same time, in the back propagation process, the obtained contribution is in turn applied to the source domain dataset samples to reduce the label class The influence of the source domain data on data calibration during the training process, the selection factor calculation process includes the following formulas in sequence: in, Represents input data, is a feature classifier, Represents the output result after the data is input into the feature classifier, represents the source dataset, represents a new dataset, Indicates the source dataset category, Represents the new dataset category, represents the selection factor; Constructing the final loss function in the diagnostic model training process further includes: The final loss function consists of four parts, involving different data sets, loss functions and parameters, and the formula is as follows: in, , λ represents the weight coefficient, is the feature classifier, is the feature extractor, is the fully connected feature processor, represents the entropy minimization loss function of the feature classifier, represents the entropy minimization loss function of the fully connected feature processor, represents the selection factor, Represents samples taken from the source dataset , Indicates taking samples from the new dataset , Indicates the source dataset category, Represents the new dataset category, Indicates that the feature classifier is used for input data The processed output belongs to the label probability prediction value.

Images