Perioperative data calculation method, terminal device, and computer-readable storage medium

Abstract

This application is applicable to the field of data processing technology, and provides a perioperative data calculation method, a terminal device, and a computer-readable storage medium, including: obtaining the time-series data of the first modality during the operation of the target disease to obtain the first time-series data; extracting the first time-series features of the first time-series data; predicting the second time-series features of the second time-series data according to the first time-series features; wherein, the second time-series data is the time-series data of the second modality during the operation of the target disease. In the embodiments of this application, the time-series features of the time-series data of one modality are predicted to obtain the time-series features of the time-series data of another modality, so that the time-series data of different modalities are finely aligned at the feature level, taking into account the association of different modality data at the feature level, and effectively improving the accuracy and robustness of data completion.

Description

Technical Field

[0001] This application belongs to the technical field of data processing, and particularly relates to a perioperative data calculation method, a terminal device, and a computer-readable storage medium. Background Art

[0002] With the continuous progress of medical technology and the increasing clinical needs, multimodal medical models play an important role in medical research and clinical practice. Multimodal medical models have the potential to process complex data and provide accurate diagnosis and treatment plans. They can integrate various medical data, including images, genes, electronic medical records, etc., providing strong support for disease diagnosis, risk prediction, and personalized treatment.

[0003] Due to technical limitations, disease progression, and other reasons, some modal data may not be available or have high heterogeneity and isomerism, affecting the accuracy and reliability of subsequent processing such as disease diagnosis and risk prediction. Therefore, how to optimize the use of missing data sets without sacrificing the performance of multimodal medical models has become an important research direction. Summary of the Invention

[0004] The embodiments of this application provide a perioperative data calculation method, a terminal device, and a computer-readable storage medium, which can improve the accuracy and robustness of data completion.

[0005] In a first aspect, the embodiments of this application provide a perioperative data calculation method, including:

[0006] Obtain the time-series data of the first modality during the operation of the target disease to obtain the first time-series data;

[0007] Extract the first time-series features of the first time-series data;

[0008] Predict the second time-series features of the second time-series data according to the first time-series features; wherein, the second time-series data is the time-series data of the second modality during the operation of the target disease.

[0009] In the embodiments of this application, the time-series features of the time-series data of one modality are used to predict the time-series features of the time-series data of another modality, enabling fine alignment of the time-series data of different modalities at the feature level, considering the correlation of different modal data at the feature level, and effectively improving the accuracy and robustness of data completion.

[0010] In a possible implementation manner of the first aspect, the predicting the second time-series features of the second time-series data according to the first time-series features includes:

[0011] Perform a timing analysis based on the first timing feature to generate a third timing feature; wherein, the timing analysis is used to obtain the overall trend and variation law of the input data;

[0012] Perform a frequency-domain analysis based on the first timing feature to generate a fourth timing feature; wherein, the frequency-domain analysis is used to obtain the frequency-domain features of the input data and transform the obtained frequency-domain features into the time domain;

[0013] Obtain the second timing feature based on the third timing feature and the fourth timing feature.

[0014] In the above manner, the polynomial regression is used to model the intrinsic overall trend and variation law of the timing feature sequence, providing a macroscopic guidance for subsequent processing. Through the frequency-domain analysis, the model can more deeply understand the semantic meaning of the time series, thereby generating more accurate and detailed timing features. The combination of the results of the timing analysis and the frequency-domain analysis helps to accurately mine the timing features.

[0015] In a possible implementation manner of the first aspect, the method further includes:

[0016] Obtain first sample data and second sample data during the operation of the target disease; wherein, the first sample data is the timing data of the first modality, and the second sample data is the timing data of the second modality;

[0017] Perform a fusion process on the first sample data and the second sample data to obtain fused timing data;

[0018] Train a first model according to the fused timing data to obtain the trained first model; wherein, the trained first model is used to extract the first timing feature of the first timing data.

[0019] In the above training manner, by fusing the timing data of different modalities, the trained model can learn the internal correlation between the timing data of different modalities, thereby helping to improve the model accuracy.

[0020] In a possible implementation manner of the first aspect, the method further includes:

[0021] Obtain third sample data and fourth sample data during the operation of the target disease; wherein, the third sample data is the timing data of the first modality, and the fourth sample data is the timing data of the second modality;

[0022] Train a second model according to the third sample data and the fourth sample data to obtain the trained second model; wherein, the trained second model is used to predict the second timing feature of the second timing data according to the first timing feature.

[0023] In a possible implementation of the first aspect, training the second model according to the third sample data and the fourth sample data to obtain the trained second model includes:

[0024] Performing multi-dimensional feature extraction in the frequency domain on the third sample data and the fourth sample data to obtain multi-dimensional frequency domain features;

[0025] Completing the third sample data and the fourth sample data according to the multi-dimensional frequency domain features to obtain the preliminarily completed third sample data and fourth sample data;

[0026] Training the second model according to the preliminarily completed third sample data and fourth sample data to obtain the trained second model.

[0027] In the above method, the multi-dimensional frequency domain features enrich the data representation ability and are conducive to providing multi-dimensional and multi-level feature information; performing preliminary completion according to the multi-dimensional frequency domain features makes the sample data more complete and accurate, thus providing reliable data support for subsequent model training.

[0028] In a possible implementation of the first aspect, the steps of training the second model include:

[0029] Inputting the third sample data into the trained first model to obtain the actual time series features of the third sample data;

[0030] Inputting the actual time series features of the third sample data into the second model to obtain the predicted time series features of the fourth sample data;

[0031] Inputting the fourth sample data into the trained first model to obtain the actual time series features of the fourth sample data;

[0032] Calculating the model loss value according to the actual time series features of the third sample data, the actual time series features and the predicted time series features of the fourth sample data;

[0033] If the model loss value is less than the preset value, determining the current second model as the trained second model;

[0034] If the model loss value is greater than the preset value, adjusting the model parameters of the second model according to the model loss value to obtain the adjusted second model, and continuing to train the adjusted second model until the trained second model is obtained.

[0035] In a possible implementation of the first aspect, calculating the model loss value according to the actual timing characteristics of the third sample data, the actual timing characteristics of the fourth sample data, and the predicted timing characteristics includes:

[0036] Obtaining the predicted occurrence probability of the target disease according to the predicted timing characteristics of the fourth sample data;

[0037] Calculating a first loss value according to the predicted occurrence probability and the true occurrence probability corresponding to the fourth sample data;

[0038] Calculating a second loss value according to the actual timing characteristics and the predicted timing characteristics of the fourth sample data;

[0039] Predicting the predicted timing characteristics of the third sample data according to the actual timing characteristics of the fourth sample data;

[0040] Calculating a third loss value according to the actual timing characteristics and the predicted timing characteristics of the third sample data;

[0041] Calculating the model loss value according to the first loss value, the second loss value, and the third loss value.

[0042] In the above calculation method of the model loss value, the first loss value is an important indicator to measure the difference between the predicted occurrence probability and the true occurrence probability of the target disease, forming an indirect constraint on the model, which is conducive to improving the accuracy of data reconstruction; the second loss value can improve the accuracy of the model in reconstructing the timing characteristics during the forward propagation process to retain the integrity of the original information; the third loss value further constrains and optimizes the model during the backward reconstruction process, improving the stability and consistency of the model during the backward derivation process; the three loss values are organically combined to jointly form the cornerstone of the model optimization process, and can constrain the model from multiple aspects, thereby helping to improve the accuracy of data completion.

[0043] In a possible implementation of the first aspect, predicting the predicted timing characteristics of the third sample data according to the actual timing characteristics of the fourth sample data includes:

[0044] Inputting the actual timing characteristics of the fourth sample data into a third model to output the predicted timing characteristics of the third sample data;

[0045] Wherein, the third model includes a plurality of cascaded residual blocks, each of the residual blocks includes a dimensionality reduction network layer and a dimensionality increase network layer, and the output end of the dimensionality reduction network layer is respectively connected to the input end of the dimensionality increase network layer in the same residual block and the output end of the dimensionality reduction network layer in the next residual block.

[0046] Through the above-mentioned third model, the intermediate features of each residual block are fused with the intermediate features of the next residual block through skip connections. Through the encoding and decoding processes of layer-by-layer refinement and adaptive fusion, higher-level and more abstract features in the input mining data are extracted. Through this third model, not only is the model's understanding ability of complex time-series data deepened, but also the accuracy and interpretability of data completion are significantly improved. In addition, by reversely constraining the model through the third module, it is beneficial to improve the consistency and integrity of data during the data completion process.

[0047] In a second aspect, an embodiment of the present application provides a perioperative data calculation device, including:

[0048] An acquisition unit, configured to acquire time-series data of a first modality during the surgical procedure of a target disease to obtain first time-series data;

[0049] A first model, configured to extract first time-series features of the first time-series data;

[0050] A second model, configured to predict second time-series features of second time-series data according to the first time-series features; wherein, the second time-series data is time-series data of a second modality during the surgical procedure of the target disease.

[0051] In a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the perioperative data calculation method according to any one of the above-mentioned first aspects.

[0052] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, it implements the perioperative data calculation method according to any one of the above-mentioned first aspects.

[0053] In a fifth aspect, an embodiment of the present application provides a computer program product, which when running on a terminal device causes the terminal device to execute the perioperative data calculation method according to any one of the above-mentioned first aspects.

[0054] It can be understood that the beneficial effects of the above-mentioned second aspect to the fifth aspect can refer to the relevant descriptions in the above-mentioned first aspect, and will not be repeated here. Description of the Drawings

[0055] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0056] Figure 1 It is a schematic flowchart of the perioperative data calculation method provided by the embodiment of the present application;

[0057] Figure 2 It is a schematic diagram of the second model provided by the embodiment of the present application;

[0058] Figure 3 It is a schematic diagram of the perioperative data calculation model provided by the embodiment of the present application;

[0059] Figure 4 It is a schematic diagram of the training framework provided by the embodiment of the present application;

[0060] Figure 5 It is a schematic diagram of the third model provided by the embodiment of the present application;

[0061] Figure 6 It is a structural block diagram of the perioperative data calculation device provided by the embodiment of the present application;

[0062] Figure 7 It is a schematic structural diagram of the terminal device provided by the embodiment of the present application. Detailed implementation manners

[0063] In the following description, for the purpose of illustration rather than limitation, specific details such as specific system structures and technologies are proposed to thoroughly understand the embodiments of the present application. However, those skilled in the art should clearly understand that the present application can also be implemented in other embodiments without these specific details. In other cases, the detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present application.

[0064] It should be understood that when used in the specification of the present application and the appended claims, the term "comprising" indicates the presence of the described features, wholes, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0065] It should also be understood that the term " / and" as used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0066] As used in the specification and appended claims of this application, the term "if" may be construed as "when" or "once" or "in response to determining" or "in response to detecting" depending on the context. Similarly, the phrase "if determined" or "if [the described condition or event] is detected" may be construed as meaning "once determined" or "in response to determining" or "once [the described condition or event] is detected" or "in response to detecting [the described condition or event]" depending on the context.

[0067] In addition, in the description of the specification and appended claims of this application, the terms "first", "second", "third", etc. are only used for distinguishing descriptions and cannot be construed as indicating or implying relative importance.

[0068] Reference to "one embodiment" or "some embodiments" or the like described in the specification of this application means that a specific feature, structure, or characteristic described in connection with that embodiment is included in one or more embodiments of this application. Thus, statements such as "in one embodiment", "in some embodiments", "in other some embodiments", "in still other embodiments", etc. that appear in different places in this specification do not necessarily all refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways.

[0069] With the continuous progress of medical technology and the increasing clinical needs, multi-modal medical models play an important role in medical research and clinical practice. Multi-modal medical models have the potential to process complex data and provide accurate diagnosis and treatment plans. They can integrate various medical data, including images, genes, electronic medical records, etc., and provide strong support for disease diagnosis, risk prediction, and personalized treatment.

[0070] Due to technical limitations, disease progression, and other reasons, some modal data may not be available or have high hetero-heterogeneity, which affects the accuracy and reliability of subsequent processing such as disease diagnosis and risk prediction. Therefore, how to optimize the use of missing data sets without sacrificing the performance of multi-modal medical models has become an important research direction.

[0071] Based on this, the embodiments of this application provide a perioperative data calculation method. In the embodiments of this application, the temporal characteristics of the temporal data of one modality are predicted to obtain the temporal characteristics of the temporal data of another modality, so that the temporal data of different modalities are finely aligned at the feature level, taking into account the association of different modal data at the feature level, and effectively improving the accuracy and robustness of data completion.

[0072] See Figure 1, which is a schematic flowchart of the perioperative data calculation method provided by an embodiment of the present application. As an example rather than a limitation, the method may include the following steps:

[0073] S101, obtain the time-series data of the first modality during the operation of the target disease to obtain the first time-series data.

[0074] In the embodiment of the present application, the time-series data refers to the data arranged in chronological order. The time-series data may be medical data such as images, genes, and electronic medical records involved in the operation of the target disease. Among them, the medical indicators represented by the time-series data of different modalities are different. For example, when the target disease is a cardiovascular disease, the time-series data of the first modality may be heart rate data, and the time-series data of the second modality may be mean arterial pressure data.

[0075] It can be understood that in the embodiment of the present application, the time-series characteristics of one modality of time-series data are used to predict the time-series characteristics of another modality of time-series data. For example, the time-series characteristics of the heart rate data during cardiovascular surgery can be used to predict the time-series characteristics of the mean arterial pressure data during cardiovascular surgery.

[0076] S102, extract the first time-series characteristics of the first time-series data.

[0077] Optionally, the first time-series data may be input into the trained first model to output the first time-series characteristics of the first time-series data.

[0078] In the above method, since the first model is pre-trained, using the first model to extract time-series characteristics can not only improve the accuracy of feature extraction, but also improve the efficiency of feature extraction.

[0079] In one implementation, the pre-training process of the first model may include the following steps:

[0080] Obtain the first sample data and the second sample data during the operation of the target disease; wherein, the first sample data is the time-series data of the first modality, and the second sample data is the time-series data of the second modality; perform fusion processing on the first sample data and the second sample data to obtain fused time-series data; train the first model according to the fused time-series data to obtain the trained first model.

[0081] Optionally, before the fusion processing, the time-series data of different modalities may be normalized. For example, the first sample data and the second sample data are respectively normalized to a normal distribution. Normal distribution normalization, also known as Z-score normalization or dimensionless processing, is a commonly used data preprocessing method. Its main purpose is to convert the original data into a distribution form with a mean of 0 and a standard deviation of 1, so that data with different dimensions or magnitudes can be compared and further analyzed. The formula for normal distribution normalization is:

[0082] ;

[0083] Among them, is the original data, is the mean of the original data, is the standard deviation of the original data. The mean represents the average level or central tendency of the data. The standard deviation represents the degree of dispersion of the data distribution, that is, the average distance between the data points and the mean.

[0084] Through normalization processing, a new data set can be obtained. Each data point in this data set represents the deviation degree of the original data point relative to the mean, and the unit is the standard deviation. In this way, regardless of the dimension or magnitude of the original data, the normalized data can be compared and analyzed on the same scale.

[0085] Optionally, a semantic-aware focused attention mechanism (CAFA) can be used for fusion processing. The specific process includes the following: Through the formula split uniformly into M data groups, and perform time-domain attention calculation on each data group as a unit, so as to perform global interaction and modeling on the time-series data during cardiovascular surgery; Through the formula capture the long-term dependencies in each data group, which can better understand the dynamic changes of the time-series data and provide more accurate and comprehensive information for subsequent modal coding and feature extraction; Through the formula perform multi-dimensional QKV decomposition on the information output by LSTM, and through the formula convert the QKV vectors into context-aware hidden representations (or called feature representations), and these representations can protect the semantic information of the input sequence; Then through the formula connect the calculated hidden representations into a feature vector (i.e., fused time-series data).

[0086] Among them, the splie() function is used to split the input data into multiple data groups. LSTM() is the function of the Long Short-Term Memory network. In this network, by introducing memory units and a series of gating mechanisms (such as input gate, forget gate, and output gate), the flow and retention of information in the sequence are effectively controlled, so as to capture dependencies in long time series. The softmax() function is used to convert a real-valued vector into a probability distribution function. The concat() function is used to connect multiple sequences.

[0087] Exemplarily, during the pre-training process of the first model, the fused time-series data is input into the first model to output predicted time-series features; the cross-entropy loss between the predicted time-series features and the true time-series features is calculated to obtain the loss value of the first model; if the loss value of the first model is less than the preset value, the current first model is recorded as the trained first model; if the loss value of the first model is greater than or equal to the preset value, the model parameters of the first model are adjusted according to the loss value of the first model, and the adjusted first model is continuously trained until the trained first model is obtained.

[0088] S103. Predict the second time-series features of the second time-series data according to the first time-series features.

[0089] Wherein, the second time-series data is the time-series data of the second modality during the surgical procedure of the target disease.

[0090] In one implementation, S103 may include:

[0091] Perform time-series analysis on the first time-series features to generate third time-series features; wherein, the time-series analysis is used to obtain the overall trend and variation law of the input data;

[0092] Perform frequency-domain analysis on the first time-series features to generate fourth time-series features; wherein, the frequency-domain analysis is used to obtain the frequency-domain features of the input data and transform the obtained frequency-domain features into the time domain;

[0093] Obtain the second time-series features according to the third time-series features and the fourth time-series features.

[0094] Optionally, time-series analysis can be performed through the formula wherein, represents the input data at the t-th time step of the i-th decoder, D is the total number of decoders. is the polynomial space matrix. is the third time-series feature.

[0095] In this method, polynomial regression is used to model the internal overall trend and variation law of the time-series feature sequence, providing a macroscopic guidance for subsequent processing.

[0096] Optionally, frequency-domain analysis can be performed through the following formula:

[0097]

[0098] wherein, FFT() is the Fourier transform, used to convert time-domain data into frequency-domain data. argTopK represents taking the K data with larger values in the sequence. IFFT() is the inverse Fourier transform, used to convert frequency-domain data into time-domain data.

[0099] In the above method, it is equivalent to converting the time-series data into the frequency domain, obtaining the part with a larger spectral amplitude from the frequency-domain data, and converting it back to the time domain. This enables the model to more deeply understand the semantic meaning of the time series, thereby generating more accurate and detailed time-series features.

[0100] Optionally, the method for obtaining the second time-series feature based on the third time-series feature and the fourth time-series feature may be: performing time-series feature diffusion processing on the first time-series feature to obtain a time-series diffusion feature; performing time-series analysis on the time-series diffusion feature to obtain a third time-series feature; performing frequency-domain analysis on the time-series diffusion feature to obtain a fourth time-series feature; and weighting the time-series diffusion feature, the third time-series feature, and the fourth time-series feature to obtain the second time-series feature.

[0101] Among them, the time-series feature diffusion processing may adopt the encoder in the transformer model.

[0102] Optionally, multiple decoders may be set, and each decoder performs the above-mentioned time-series analysis and frequency-domain analysis to obtain the second time-series feature output by each decoder; then, the second time-series features output by the multiple decoders are averaged to output the final second time-series feature.

[0103] In one implementation, S103 may be implemented through a second model. Specifically, the first time-series feature is input into the trained second model, and the second time-series feature of the predicted second time-series data is output.

[0104] Exemplarily, see Figure 2 , which is a schematic diagram of the second model provided by the embodiments of the present application. By way of example and not limitation, as Figure 2 shown, the second model may include D decoders, and each decoder includes a time-series feature diffusion module, a global perception module, and a multi-frequency aggregation module. Among them, the feature diffusion module is used to perform time-series feature diffusion processing on the input data. The global perception module is used to perform time-series analysis on the input data. The multi-frequency aggregation module is used to perform frequency-domain analysis on the input data.

[0105] The above-mentioned second model is a generative model based on a diffusion structure, which can realize high-quality reconstruction from the time-series features of one modality to the time-series features of another modality, effectively improving the accuracy of data completion.

[0106] Exemplarily, see Figure 3 , which is a schematic diagram of the perioperative data calculation model provided by the embodiments of the present application. By way of example and not limitation, as Figure 3As shown in the figure, the perioperative data calculation model may include a trained first model and a trained second model. The above S101-S102 can be implemented by the perioperative data calculation model. Specifically: input the first time-series data into the perioperative data calculation model, the trained first model extracts the first time-series features of the first time-series data, and inputs the first time-series features into the second model; the trained second model predicts the second time-series features of the second time-series data according to the first time-series features, that is, the perioperative data calculation model outputs the second time-series features.

[0107] In the embodiments of the present application, the time-series features of one modality of time-series data are predicted to obtain the time-series features of another modality of time-series data, so that the time-series data of different modalities are finely aligned at the feature level, considering the correlation of different modality data at the feature level, effectively improving the accuracy and robustness of data completion.

[0108] It should be noted that the first model in the perioperative data calculation model can be pre-trained separately. Use the pre-trained first model and the untrained second model to construct the perioperative data calculation model, and then train the perioperative data calculation model, that is, train the second model in the perioperative data calculation model to obtain the trained perioperative data calculation model. In practical applications, data completion is achieved through the trained perioperative data calculation model.

[0109] The training method of the perioperative data calculation model is introduced below.

[0110] S201, obtain the third sample data and the fourth sample data during the operation of the target disease.

[0111] Among them, the third sample data is the time-series data of the first modality, and the fourth sample data is the time-series data of the second modality.

[0112] S202, train the second model according to the third sample data and the fourth sample data to obtain the trained second model.

[0113] In the embodiments of the present application, in order to train the perioperative data calculation model, a training framework is built.

[0114] See Figure 4 , which is a schematic diagram of the training framework provided by the embodiments of the present application. As an example but not a limitation, as Figure 4 shown, the training framework may include a first model, a second model, a third model, and a prediction module.

[0115] Based on Figure 4 the training framework shown, the training process in S202 may include:

[0116] Input the third sample datax a Input the trained first model to obtain the actual time series features of the third sample data h ;

[0117] Input the actual time series features of the third sample data h into the second model to obtain the fourth sample data x b and its predicted time series features h’ ;

[0118] Input the fourth sample data x b into the trained first model to obtain the actual time series features of the fourth sample data h’’ ;

[0119] Calculate the model loss value according to the actual time series features of the third sample data h , the actual time series features of the fourth sample data h’ and the predicted time series features h’’ ;

[0120] If the model loss value is less than the preset value, determine the current second model as the trained second model;

[0121] If the model loss value is greater than the preset value, adjust the model parameters of the second model according to the model loss value to obtain the adjusted second model, and continue to train the adjusted second model until the trained second model is obtained.

[0122] In one embodiment, the calculation method of the model loss value may include:

[0123] Obtain the predicted occurrence probability of the target disease according to the predicted time series features of the fourth sample data h’ ; s’ ;

[0124] Calculate the first loss value according to the predicted occurrence probability s’ and the true occurrence probability corresponding to the fourth sample data s ; L cls ;

[0125] Calculate the second loss value according to the actual time series features of the fourth sample data h’’ and the predicted time series features h’ ; L f ;

[0126] Predict the predicted time series features of the third sample data according to the actual time series features of the fourth sample data h’’ ; h’’’ ;

[0127] According to the actual timing characteristics of the third sample data h and the predicted timing characteristics h’’’ calculate the third loss value L b ;

[0128] According to the first loss value L cls 、the second loss value L f and the third loss value L b calculate the model loss value.

[0129] Based on Figure 4 the training framework shown, in one implementation, the predicted timing characteristics of the fourth sample data h’ can be input into the prediction module, and the predicted occurrence probability of the target disease is output s’ . The true occurrence probability corresponding to the fourth sample data s can be manually labeled or the occurrence probability obtained according to expert prior knowledge.

[0130] It should be noted that the prediction module can be pre-trained. Optionally, the output of the prediction module can be a probability value between 0 and 1, indicating the occurrence probability of the target disease after surgery. Calculate the first loss value L cls can be to calculate the difference value between the predicted occurrence probability s’ and the true occurrence probability s , such as calculating through cross-entropy loss or binary cross-entropy loss, etc.

[0131] Optionally, the first loss value L cls 、the second loss value L f and the third loss value L b can be weighted to obtain the model loss value.

[0132] It can be understood that the first loss value can characterize the accuracy of the occurrence probability prediction based on the predicted timing characteristics of the fourth sample data, the second loss value can characterize the difference degree between the actual timing characteristics and the predicted timing characteristics of the fourth sample data, and the third loss value can characterize the difference degree between the actual timing characteristics and the predicted timing characteristics of the third sample data.

[0133] In the calculation method of the above model loss value, the first loss value is an important indicator for measuring the difference between the predicted occurrence probability and the true occurrence probability of the target disease, forming an indirect constraint on the model, which is conducive to improving the accuracy of data reconstruction; the second loss value can improve the accuracy of the model in reconstructing the temporal features during the forward propagation process to retain the integrity of the original information; the third loss value further constrains and optimizes the model during the backward reconstruction process, improving the stability and consistency of the model during the backward derivation process; the three loss values are organically combined to jointly form the cornerstone of the model optimization process, and can constrain the model from multiple aspects, thereby helping to improve the accuracy of data completion.

[0134] Based on Figure 4 the training framework shown, in one implementation, the actual temporal features of the fourth sample data h’’ can be input into the third model, and the predicted temporal features of the third sample data are output h’’’。

[0135] Optionally, the third model may include a plurality of cascaded residual blocks, each of the residual blocks includes a dimensionality reduction network layer and a dimensionality increase network layer, and the output end of the dimensionality reduction network layer is respectively connected to the input end of the dimensionality increase network layer in the same residual block and the output end of the dimensionality reduction network layer in the next residual block.

[0136] Exemplarily, referring to Figure 5 , it is a schematic diagram of the third model provided by the embodiments of the present application. By way of example and not limitation, as Figure 5 shown, the third model includes 3 residual blocks.

[0137] Among them, the output end of the dimensionality reduction network layer in the first residual block is connected to the input end of the dimensionality increase network layer in the first residual block and the output end of the dimensionality reduction network layer in the second residual block. The output end of the dimensionality reduction network layer in the second residual block is connected to the input end of the dimensionality increase network layer in the second residual block and the output end of the dimensionality reduction network layer in the third residual block.

[0138] Among them, the input data of the first residual block and the output data of the dimensionality increase network layer in the first residual block are weighted to generate the output data of the first residual block, which is used as the input data of the second residual block. The input data of the second residual block and the output data of the dimensionality increase network layer in the second residual block are weighted to generate the output data of the second residual block, which is used as the input data of the third residual block. The input data of the third residual block and the output data of the dimensionality increase network layer in the third residual block are weighted to generate the output data of the third residual block, which is used as the output data of the third model.

[0139] It should be noted that Figure 5Only the case of 3 residual blocks is shown, and more or fewer residual blocks can be cascaded in practical applications. In the embodiments of the present application, the number of residual blocks included in the third model is not specifically limited.

[0140] Through the above-mentioned third model, the intermediate features of each residual block are fused with the intermediate features of the next residual block through skip connections. Through the encoding and decoding process of layer-by-layer refinement and adaptive fusion, higher-level and more abstract features in the input mining data are extracted. Through this third model, not only the understanding ability of the model for complex time-series data is deepened, but also the accuracy and interpretability of data completion are significantly improved. In addition, by reverse-constraining the model through the third module, it is beneficial to improve the consistency and integrity of data during the data completion process.

[0141] In one embodiment, before training the second model, the sample data can also be subjected to preliminary completion processing. Specifically, it includes:

[0142] Performing multi-dimensional feature extraction in the frequency domain on the third sample data and the fourth sample data to obtain multi-dimensional frequency domain features;

[0143] Completing the third sample data and the fourth sample data according to the multi-dimensional frequency domain features to obtain the preliminarily completed third sample data and fourth sample data;

[0144] Training the second model according to the preliminarily completed third sample data and fourth sample data to obtain the trained second model.

[0145] In one implementation, the process of preliminary completion processing can be realized through the following formula:

[0146]

[0147]

[0148]

[0149]

[0150] Among them, MSVD() is used for dimensionality increase mapping. split() is used to split the input data into multiple groups of data. FFT() is used to perform Fourier transform on the input data to convert the input data into frequency domain data. is a multi-layer perceptron (MLP) in the frequency domain. IFFT() is used to perform inverse Fourier transform on the input data to convert the input data into time domain data. concat() is used to merge multiple groups of data into one-dimensional data.

[0151] Optionally, after concat, the missing parts of the sample data obtained from the one-dimensional data are completed to the corresponding positions of the sample data according to the corresponding relationship of the sampling time, so as to achieve preliminary completion of the sample data.

[0152] In the above preliminary completion process, the multi-dimensional frequency domain features are split into multiple groups of data, enriching the data representation capability; then, the multi-layer perceptron is used to extract features from the multiple groups of data, which is conducive to providing multi-dimensional and multi-level feature information; through the above preliminary completion process, the sample data can be effectively preliminarily completed, making the sample data more complete and accurate, thereby providing reliable data support for subsequent model training.

[0153] In one implementation, the above-mentioned MSVD() dimensionality-up mapping process can be implemented by the following formula:

[0154] ;

[0155] Among them, x n is the data of the nth sampling point in the sample data. Shuffle() is used to rearrange the elements in the array in random order. m is the rearranged data. SVD() is used to perform singular value decomposition on the input data. m is the data after SVD decomposition. n The data after dimensionality increase.

[0156] The above-mentioned dimensionality-increasing mapping process is helpful to mine the internal correlation of the data, greatly enriches the data representation capability, and provides a reliable data basis for subsequent data completion.

[0157] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0158] Corresponding to the perioperative data calculation method described in the above embodiment, Figure 6 This is a structural block diagram of the perioperative data calculation device provided in an embodiment of the present application. For the sake of convenience of explanation, only the parts related to the embodiment of the present application are shown.

[0159] Reference Figure 6 , the device 6 comprises:

[0160] The acquisition unit 61 is used to acquire time series data of a first modality during a surgery for a target disease to obtain first time series data.

[0161] The first model 62 is used to extract the first time series features of the first time series data.

[0162] A second model 63, configured to predict a second temporal feature of second temporal data according to the first temporal feature; wherein, the second temporal data is temporal data of a second modality during a surgical procedure for the target disease.

[0163] Optionally, the second model 63 is further configured to:

[0164] Perform temporal analysis according to the first temporal feature to generate a third temporal feature; wherein, the temporal analysis is used to obtain the overall trend and variation law of the input data;

[0165] Perform frequency domain analysis according to the first temporal feature to generate a fourth temporal feature; wherein, the frequency domain analysis is used to obtain the frequency domain feature of the input data and transform the obtained frequency domain feature into the time domain;

[0166] Obtain the second temporal feature according to the third temporal feature and the fourth temporal feature.

[0167] Optionally, the apparatus 6 further includes a training unit 64, configured to:

[0168] Obtain first sample data and second sample data during a surgical procedure for the target disease; wherein, the first sample data is temporal data of the first modality, and the second sample data is temporal data of the second modality;

[0169] Perform fusion processing on the first sample data and the second sample data to obtain fused temporal data;

[0170] Train the first model according to the fused temporal data to obtain the trained first model; wherein, the trained first model is used to extract the first temporal feature of the first temporal data.

[0171] Optionally, the training unit 64 is further configured to:

[0172] Obtain third sample data and fourth sample data during a surgical procedure for the target disease; wherein, the third sample data is temporal data of the first modality, and the fourth sample data is temporal data of the second modality;

[0173] Train the second model according to the third sample data and the fourth sample data to obtain the trained second model; wherein, the trained second model is used to predict the second temporal feature of the second temporal data according to the first temporal feature.

[0174] Optionally, the training unit 64 is further configured to:

[0175] Perform multi-dimensional feature extraction in the frequency domain on the third sample data and the fourth sample data to obtain multi-dimensional frequency domain features;

[0176] Complete the third sample data and the fourth sample data according to the multi-dimensional frequency domain characteristics to obtain the initially completed third sample data and fourth sample data;

[0177] Train the second model according to the initially completed third sample data and fourth sample data to obtain the trained second model.

[0178] Optionally, the training unit 64 is further configured to:

[0179] Input the third sample data into the trained first model to obtain the actual time series characteristics of the third sample data;

[0180] Input the actual time series characteristics of the third sample data into the second model to obtain the predicted time series characteristics of the fourth sample data;

[0181] Input the fourth sample data into the trained first model to obtain the actual time series characteristics of the fourth sample data;

[0182] Calculate the model loss value according to the actual time series characteristics, the actual time series characteristics and the predicted time series characteristics of the fourth sample data of the third sample data;

[0183] If the model loss value is less than the preset value, determine the current second model as the trained second model;

[0184] If the model loss value is greater than the preset value, adjust the model parameters of the second model according to the model loss value to obtain the adjusted second model, and continue to train the adjusted second model until the trained second model is obtained.

[0185] Optionally, the training unit 64 is further configured to:

[0186] Obtain the predicted occurrence probability of the target disease according to the predicted time series characteristics of the fourth sample data;

[0187] Calculate the first loss value according to the predicted occurrence probability and the true occurrence probability corresponding to the fourth sample data;

[0188] Calculate the second loss value according to the actual time series characteristics and the predicted time series characteristics of the fourth sample data;

[0189] Predict the predicted time series characteristics of the third sample data according to the actual time series characteristics of the fourth sample data;

[0190] Calculate the third loss value according to the actual time series characteristics and the predicted time series characteristics of the third sample data;

[0191] Calculate the model loss value according to the first loss value, the second loss value, and the third loss value.

[0192] Optionally, the training unit 64 is further configured to:

[0193] Input the actual timing feature of the fourth sample data into a third model, and output the predicted timing feature of the third sample data;

[0194] Wherein, the third model includes a plurality of cascaded residual blocks, each of the residual blocks includes a dimensionality reduction network layer and a dimensionality increase network layer, and the output end of the dimensionality reduction network layer is respectively connected to the input end of the dimensionality increase network layer in the same residual block and the output end of the dimensionality reduction network layer in the next residual block.

[0195] It should be noted that, for the information interaction, execution process, etc. between the above-mentioned devices / units, since they are based on the same concept as the method embodiments of the present application, for their specific functions and the technical effects brought, please refer to the method embodiment part for details, and will not be elaborated here.

[0196] In addition, Figure 6 The perioperative data calculation device shown may be a software unit, a hardware unit, or a unit combining software and hardware built into an existing terminal device, may also be integrated into the terminal device as an independent attachment, or may exist as an independent terminal device.

[0197] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of the present application. The specific working processes of the units and modules in the above system can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated here.

[0198] Figure 7 is a schematic structural diagram of a terminal device provided by an embodiment of the present application. As Figure 7 shown, the terminal device 7 in this embodiment includes: at least one processor 70 ( Figure 7Only one (only one is shown) processor, a memory 71, and a computer program 72 stored in the memory 71 and executable on the at least one processor 70 are shown. When the processor 70 executes the computer program 72, the steps in any of the above-described perioperative data calculation method embodiments are implemented.

[0199] The terminal device may be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that Figure 7 This is only an example of the terminal device 7 and does not constitute a limitation on the terminal device 7. It may include more or fewer components than shown in the figure, or combine some components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0200] The processor 70 may be a central processing unit (CPU). The processor 70 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0201] In some embodiments, the memory 71 may be an internal storage unit of the terminal device 7, such as the hard disk or memory of the terminal device 7. In other embodiments, the memory 71 may also be an external storage device of the terminal device 7, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 7. Further, the memory 71 may also include both the internal storage unit and the external storage device of the terminal device 7. The memory 71 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program, etc. The memory 71 may also be used to temporarily store data that has been output or will be output.

[0202] An embodiment of the present application also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments can be implemented.

[0203] An embodiment of the present application provides a computer program product, and when the computer program product runs on a terminal device, the terminal device can implement the steps in the above method embodiments when executed.

[0204] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above method embodiments of the present application, a computer program can be used to instruct relevant hardware to complete. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps in the above method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the device / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disc, etc. In some jurisdictions, according to legislation and patent practice, the computer-readable medium cannot be an electrical carrier signal and a telecommunication signal.

[0205] In the above embodiments, the descriptions of the respective embodiments have their own focuses. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0206] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0207] In the embodiments provided in the present application, it should be understood that the disclosed device / terminal device and method can be implemented in other ways. For example, the device / terminal device embodiments described above are merely illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in electrical, mechanical or other forms.

[0208] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0209] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A perioperative data calculation method, characterized in that: include: Acquire time series data of a first modality during a surgical procedure for a target disease to obtain first time series data; wherein the time series data is medical data involved in the surgical procedure for the target disease, and the medical data includes at least one of the following: images, genes, and electronic medical records; Extracting a first time series feature of the first time series data; Predicting a second time series feature of second time series data according to the first time series feature; wherein the second time series data is time series data of a second modality during surgery for the target disease; The method further comprises: Acquire first sample data and second sample data during a surgical procedure for the target disease; wherein the first sample data is time series data of the first modality, and the second sample data is time series data of the second modality; Performing fusion processing on the first sample data and the second sample data to obtain fused time series data; Training a first model according to the fused time series data to obtain the trained first model; wherein the trained first model is used to extract first time series features of the first time series data; The method further comprises: Acquire third sample data and fourth sample data during the surgical procedure of the target disease; wherein the third sample data is the time series data of the first modality, and the fourth sample data is the time series data of the second modality; Training a second model according to the third sample data and the fourth sample data to obtain the trained second model; wherein the trained second model is used to predict the second time series feature of the second time series data according to the first time series feature; The step of training the second model comprises: Inputting the third sample data into the trained first model to obtain actual time series features of the third sample data; Inputting the actual time series characteristics of the third sample data into the second model to obtain the predicted time series characteristics of the fourth sample data; Inputting the fourth sample data into the trained first model to obtain actual time series features of the fourth sample data; Calculate the model loss value according to the actual time series characteristics of the third sample data, the actual time series characteristics of the fourth sample data, and the predicted time series characteristics; If the model loss value is less than a preset value, determining the current second model as the trained second model; If the model loss value is greater than a preset value, the model parameters of the second model are adjusted according to the model loss value to obtain the adjusted second model, and the adjusted second model is continued to be trained until the trained second model is obtained.

2. The perioperative data calculation method according to claim 1, characterized in that: The predicting a second time series feature of second time series data according to the first time series feature includes: Performing time series analysis according to the first time series feature to generate a third time series feature; wherein the time series analysis is used to obtain the overall trend and change law of the input data; Performing frequency domain analysis according to the first time series feature to generate a fourth time series feature; wherein the frequency domain analysis is used to obtain frequency domain features of input data and transform the obtained frequency domain features into the time domain; The second timing characteristic is obtained according to the third timing characteristic and the fourth timing characteristic.

3. The perioperative data calculation method according to claim 1, characterized in that: The step of training the second model according to the third sample data and the fourth sample data to obtain the trained second model includes: Performing multi-dimensional feature extraction on the third sample data and the fourth sample data in the frequency domain to obtain multi-dimensional frequency domain features; Completing the third sample data and the fourth sample data according to the multi-dimensional frequency domain features to obtain the third sample data and the fourth sample data after preliminary completion; The second model is trained according to the third sample data and the fourth sample data after preliminary completion to obtain the trained second model.

4. The perioperative data calculation method according to claim 1, characterized in that: The calculating the model loss value according to the actual time series feature of the third sample data, the actual time series feature of the fourth sample data and the predicted time series feature includes: Obtaining a predicted probability of occurrence of the target disease according to the predicted time series characteristics of the fourth sample data; Calculate a first loss value according to the predicted occurrence probability and the actual occurrence probability corresponding to the fourth sample data; Calculate a second loss value according to the actual time series characteristics and the predicted time series characteristics of the fourth sample data; Predicting the predicted time series characteristics of the third sample data according to the actual time series characteristics of the fourth sample data; Calculate a third loss value according to the actual time series characteristics and the predicted time series characteristics of the third sample data; The model loss value is calculated according to the first loss value, the second loss value and the third loss value.

5. The perioperative data calculation method according to claim 4, characterized in that: The predicting the predicted time series feature of the third sample data according to the actual time series feature of the fourth sample data includes: Inputting the actual time series features of the fourth sample data into the third model, and outputting the predicted time series features of the third sample data; Among them, the third model includes multiple cascaded residual blocks, each of the residual blocks includes a dimensionality reduction network layer and a dimensionality increase network layer, and the output end of the dimensionality reduction network layer is respectively connected to the input end of the dimensionality increase network layer in the same residual block and the output end of the dimensionality reduction network layer in the next residual block.

6. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images