Fetal heart uterine contraction automatic interpretation algorithm based on multi-domain representation and ensemble learning
Through multi-domain characterization and ensemble learning methods, screening and fusing the features in the fetal heart contraction map, the shortcomings of fetal ECG signal feature extraction and long-term dependency capture in the prior art are solved, and more accurate and reliable automatic interpretation of fetal heart contraction maps are achieved.
Patent Information
- Application Number
- CN202411863442.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-04-11
AI Technical Summary
The existing automatic interpretation algorithm for fetal heart contraction charts fails to effectively extract and analyze the frequency and time-frequency domain characteristics in fetal electrocardiogram signals, and fails to deeply capture the long-term dependence relationships in the signal, resulting in insufficient accuracy and reliability of interpretation.
The multi-domain characterization and ensemble learning method is adopted to filter out features that are helpful for CTG interpretation through a two-stage feature selection algorithm (variance threshold screening and differential evolution algorithm), and use Stacking ensemble learning algorithm to fuse these features with the basic model to construct the final ensemble model, and at the same time, it realizes global and local visual interpretation based on Kernel SHAP technology.
It effectively improves the accuracy and interpretability of automatic interpretation of CTG signals, ensures more comprehensive signal feature extraction, provides more accurate and reliable diagnostic information, and provides an important reference for clinical decision-making.
Smart Images
Figure CN119807656A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical signal processing, and in particular to an automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and integrated learning. Background Art
[0002] Fetal monitoring refers to the use of a variety of monitoring methods to monitor the growth and development of the fetus in the mother's uterus in real time and accurately and assess its health status. Electronic fetal heart monitoring is a core monitoring method widely used in the field of perinatal monitoring. It significantly improves the safety of monitoring through non-invasive technology and is used to describe the status of continuous fetal heart rate (FHR) and uterine contraction (UC). Cardiotocography (CTG) is a widely trusted and widely used electronic fetal heart monitoring technology. As a widely used non-invasive prenatal diagnostic method, cardiotocography (CTG) is essential for monitoring the health of the fetus in the uterus before and after delivery. It is a core tool for clinicians and midwives to obtain fetal status information directly through visual observation during the diagnosis and treatment process. However, the interpretation of CTG is not absolutely objective. It is highly dependent on established interpretation guidelines and the doctor's rich clinical experience. This subjectivity increases the complexity of interpretation to a certain extent, and may lead to ambiguity in interpretation and potential misdiagnosis, posing a potential threat to the health of pregnant women and fetuses. In order to make up for this limitation and improve the accuracy of fetal health status judgment, it is particularly important to develop an intelligent auxiliary judgment tool based on fetal heart rate (FHR). However, there are technical difficulties, mainly reflected in:
[0003] During fetal heart rate monitoring, there are many irresistible interferences in the fetal heart rate signal acquisition process, such as sensor abnormalities, maternal movement, fetal activity, etc., which will cause the fetal heart rate signal to be frequently partially lost. Signal loss is particularly serious for premature births, mothers with high body mass index, multiple pregnancies, and the second stage of labor, which makes the automated analysis of fetal heart rate signals extremely challenging. Current algorithms fail to effectively extract and analyze the frequency domain and time-frequency domain features in fetal ECG signals, which are crucial for accurately judging the periodicity and non-stationarity of the signals; the algorithms fail to deeply capture and express the long-term dependencies in the signals, which is also critical for understanding the long-term trend of fetal heart rate changes and predicting potential risks. Summary of the invention
[0004] The purpose of the present invention is to provide an automatic interpretation algorithm for fetal heart and uterine contraction chart based on multi-domain representation and integrated learning, adopting the idea of multi-representation and multi-model integration to fully extract relevant features that are helpful for CTG interpretation, promote the development of CTG signal analysis field, and provide clinicians with more accurate and reliable auxiliary diagnostic tools.
[0005] According to one aspect of the present disclosure, the following technical solution is provided: an automatic interpretation algorithm of fetal heart rate and uterine contraction chart based on multi-domain representation and integrated learning, comprising:
[0006] The feature selection is performed through a two-stage feature selection algorithm. First, the feature set is screened by variance threshold;
[0007] Then input it into the differential evolution algorithm for secondary feature selection;
[0008] Finally, the Stacking ensemble learning algorithm is used to effectively integrate the above-mentioned screening features with the base model, construct the final integrated model, and realize global and local visual interpretation based on the Kernel SHAP technology.
[0009] According to the automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and integrated learning of at least one embodiment of the present disclosure, the steps of variance threshold screening are as follows: first, the variance of each feature is calculated and the variance is retained; then, the features are filtered, the retained feature variances are traversed, the features whose variances exceed the set threshold are retained, and the features whose variances are lower than the set threshold are discarded; finally, a feature set that meets the set threshold is output.
[0010] According to at least one embodiment of the fetal heart rate and uterine contraction graph automatic interpretation algorithm based on multi-domain representation and integrated learning of the present disclosure, the differential evolution algorithm includes initialization, mutation, crossover and selection steps, and performs a heuristic random search algorithm based on group differences. It introduces a differential mutation model that utilizes individual differences in the current group to construct mutant individuals, which is used to solve the overall optimal solution in the multidimensional space.
[0011] According to the automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and integrated learning of at least one embodiment of the present disclosure, the initialization step is specifically as follows:
[0012] Assume that there are individuals in the solution space, that is, the population size is, each individual is a dimension vector, and the initial population is randomly generated:
[0013]
[0014] Among them, i is the i-th individual, j is the j-th component, and rand(0,1) represents a random number in the interval [0,1]. represents the lower bound of the jth component, represents the upper bound of the j-th component.
[0015] According to the automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and integrated learning of at least one embodiment of the present disclosure, the variation step is specifically as follows:
[0016] The differential evolution algorithm generates new mutant individuals based on the difference between two random vectors. For the i-th target individual vector X r1 (t), new mutation individual vector V i (t) is shown in the following formula:
[0017] V i (t) = X r1 (t)+F×(X r2 (t)-X r3 (t))#
[0018] Among them, t is the iteration step length, X r1 (t), X r2 (t), X r3 (t) are three individuals selected randomly, and r1≠r2≠r3, and F is the variation factor.
[0019] According to at least one embodiment of the present disclosure, the automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and integrated learning, the crossover step is as follows:
[0020] A crossover operation is performed on each individual and its generated mutation vector, that is, the elements in the target vector are combined with the elements in the new mutation vector to generate a verification vector. The method of the crossover operation is:
[0021]
[0022] Among them, CR∈[0,1] is the crossover probability, j rand is a random component to ensure that at least one dimension of the experimental individual vector after crossover is provided by the mutant individual vector.
[0023] According to the automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and integrated learning of at least one embodiment of the present disclosure, the selection step is as follows:
[0024] The differential evolution algorithm uses a greedy algorithm to select an individual as the next generation from the target individual and the test individuals generated by mutation and crossover operations based on the value of the fitness function;
[0025]
[0026] During differential evolution, mutation, crossover, and selection operations are performed sequentially until the maximum number of iterations is reached.
[0027] Technical effects and advantages of the present invention:
[0028] The present invention can realize automatic interpretation of CTG signals through a two-stage feature selection algorithm, which effectively makes up for the shortcomings of manual feature extraction algorithms in capturing long-term span dependencies.
[0029] Automatic interpretation not only improves analysis efficiency, but also ensures more comprehensive signal feature extraction, facilitates subsequent analysis and research of CTG signals, provides doctors with more accurate and reliable diagnostic information, and provides important reference for clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] The accompanying drawings illustrate exemplary embodiments of the present disclosure and together with the description serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.
[0031] Figure 1 It is a flow chart of an automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and integrated learning of the present invention.
[0032] Figure 2 It is the normal CTG and corresponding stress map of the present invention.
[0033] Figure 3 It is the abnormal CTG and its corresponding stress diagram of the present invention.
[0034] Figure 4 It is a comparative experimental diagram of the SHAP decision diagrams of different samples of the present invention. DETAILED DESCRIPTION
[0035] For descriptive purposes, the present disclosure may use spatially relative terms such as "under," "beneath," "under," "above," "on," "higher," and "side (e.g., as in "sidewall")," to describe the relationship of one component to another (other) component as shown in the accompanying drawings. The spatially relative terms are intended to encompass different orientations of the device in use, operation, and / or manufacture in addition to the orientation depicted in the accompanying drawings. For example, if the device in the accompanying drawings is turned over, components described as "under" or "beneath" other components or features would then be positioned "above" the other components or features. Thus, the exemplary term "under" can encompass both the "above" and "below" orientations. Furthermore, the device may be otherwise oriented (e.g., rotated 90 degrees or at other orientations), and the spatially relative descriptors used herein should be interpreted accordingly.
[0036] like Figure 1 As shown, the present invention discloses an automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and integrated learning, including:
[0037] S1, feature selection is performed through a two-stage feature selection algorithm, firstly, the feature set is screened by variance threshold;
[0038] S2, then input it into the differential evolution algorithm for secondary feature selection;
[0039] S3. Finally, the Stacking ensemble learning algorithm is used to effectively integrate the above-mentioned screening features with the base model, construct the final integrated model, and realize global and local visual interpretation based on Kernel SHAP technology.
[0040] In this embodiment, the steps of variance threshold screening are as follows: first, the variance of each feature is calculated and retained; then, the features are filtered, the retained feature variances are traversed, the features whose variances exceed the set threshold are retained, and the features whose variances are lower than the set threshold are discarded. Based on the sensitive analysis of the feature variances, the features are screened more carefully. This method aims to remove features with smaller variances and retain features with larger variances to improve the performance of the model, and finally output a feature set that meets the set threshold.
[0041] The specific implementation of the algorithm is as follows: Preserve the variance, where n is the number of data points for each feature, x i is the eigenvalue corresponding to each data point, is the mean value of the feature; this method considers the difference of each data point in the dataset from the mean.
[0042] In this embodiment, the differential evolution algorithm is used as the feature selection algorithm of the second stage to further optimize the feature subset and improve the model performance. The differential evolution algorithm includes initialization, mutation, crossover and selection steps, and performs a heuristic random search algorithm based on group differences. It introduces a differential mutation model that uses individual differences in the current group to construct mutant individuals, which is used to solve the overall optimal solution in the multidimensional space.
[0043] In this embodiment, the initialization steps are specifically as follows:
[0044] Assume that there are individuals in the solution space, that is, the population size is, each individual is a dimension vector, and the initial population is randomly generated:
[0045]
[0046] Among them, i is the i-th individual, j is the j-th component, and rand(0,1) represents a random number in the interval [0,1]. represents the lower bound of the jth component, represents the upper bound of the j-th component.
[0047] In this embodiment, the mutation steps are specifically as follows:
[0048] The differential evolution algorithm generates new mutant individuals based on the difference between two random vectors. For the i-th target individual vector X r1(t), new mutation individual vector V i (t) is shown in the following formula:
[0049] V i (t) = X r1 (t)+F×(X r2 (t)-X r3 (t))#
[0050] Among them, t is the iteration step length, X r1 (t), X r2 (t), X r3 (t) are three individuals selected randomly, and r1≠r2≠r3, and F is the variation factor.
[0051] In this embodiment, the crossover steps are as follows:
[0052] A crossover operation is performed on each individual and its generated mutation vector, that is, the elements in the target vector are combined with the elements in the new mutation vector to generate a verification vector. The method of the crossover operation is:
[0053]
[0054] Among them, CR∈[0,1] is the crossover probability, j rand is a random component to ensure that at least one dimension of the experimental individual vector after crossover is provided by the mutant individual vector.
[0055] In this embodiment, the selection steps are as follows:
[0056] The differential evolution algorithm uses a greedy algorithm to select an individual as the next generation from the target individual and the test individuals generated by mutation and crossover operations based on the value of the fitness function;
[0057]
[0058] During differential evolution, mutation, crossover, and selection operations are performed sequentially until the maximum number of iterations is reached. The algorithm explores different regions in the search space by vector differencing, applies crossover to exchange information between the mutated solution and the current solution, and performs selection operations to directly forward to the expected region containing a more suitable solution.
[0059] The present invention maximizes the advantages of different characterization methods to improve the accuracy and interpretability of CTG interpretation, and the results are shown in Figures 2, 3, and 4.
[0060] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments / methods or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments / methods or examples described in this specification and the features of different embodiments / methods or examples, unless they are contradictory.
[0061] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.
[0062] Those skilled in the art should understand that the above embodiments are only for the purpose of clearly illustrating the present disclosure, and are not intended to limit the scope of the present disclosure. For those skilled in the art, other changes or modifications may be made based on the above disclosure, and these changes or modifications are still within the scope of the present disclosure.
Claims
1. An automatic interpretation algorithm for fetal heart and uterine contraction chart based on multi-domain representation and ensemble learning, characterized in that: include: The feature selection is performed through a two-stage feature selection algorithm. First, the feature set is screened by variance threshold; Then input it into the differential evolution algorithm for secondary feature selection; Finally, the Stacking ensemble learning algorithm is used to effectively integrate the above-mentioned screening features with the base model, construct the final integrated model, and realize global and local visual interpretation based on the Kernel SHAP technology.
2. The automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and ensemble learning according to claim 1 is characterized in that: The steps of variance threshold screening are as follows: first, the variance of each feature is calculated and retained; then, the features are filtered, the retained feature variances are traversed, the features whose variances exceed the set threshold are retained, and the features whose variances are lower than the set threshold are discarded; finally, a feature set that meets the set threshold is output.
3. The automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and ensemble learning according to claim 1 is characterized in that: The differential evolution algorithm includes initialization, mutation, crossover and selection steps, performs a heuristic random search algorithm based on group differences, introduces a differential mutation model that uses individual differences in the current group to construct mutant individuals, and is used to solve the overall optimal solution in multidimensional space.
4. The automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and ensemble learning according to claim 3 is characterized in that: The initialization steps are as follows: Assume that there are individuals in the solution space, that is, the population size is, each individual is a dimension vector, and the initial population is randomly generated: Among them, i is the i-th individual, j is the j-th component, and rand(0,1) represents a random number in the interval [0,1]. represents the lower bound of the jth component, represents the upper bound of the j-th component.
5. The automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and ensemble learning according to claim 3 is characterized in that: The mutation steps are as follows: The differential evolution algorithm generates new mutant individuals based on the difference between two random vectors. For the i-th target individual vector X r1 (t), new mutation individual vector V i (t) is shown in the following formula: V i (t)=X r1 (t)+F×(X r2 (t)-X r3 (t))# Among them, t is the iteration step length, X r1 (t), X r2 (t), X r3 (t) are three individuals selected randomly, and r1≠r2≠r3, F is the variation factor.
6. The automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and ensemble learning according to claim 3 is characterized in that: The crossover steps are as follows: A crossover operation is performed on each individual and its generated mutation vector, that is, the elements in the target vector are combined with the elements in the new mutation vector to generate a verification vector. The method of the crossover operation is: Among them, CR∈[0,1] is the crossover probability, j rand is a random component to ensure that at least one dimension of the experimental individual vector after crossover is provided by the mutant individual vector.
7. The automatic interpretation algorithm of fetal heart and uterine contraction chart based on multi-domain representation and ensemble learning according to claim 3 is characterized in that: The selection steps are as follows: The differential evolution algorithm uses a greedy algorithm to select an individual as the next generation from the target individual and the test individuals generated by mutation and crossover operations based on the value of the fitness function; During differential evolution, mutation, crossover, and selection operations are performed sequentially until the maximum number of iterations is reached.