Training device for hypotension prediction model and hypotension prediction device

Abstract

This invention discloses a training device and a hypotension prediction device for a hypotension prediction model. The device includes: a sample generation module for preprocessing blood pressure parameters and waveform data from a database to obtain training samples; the waveform data includes photoplethysmography (PPG) waves and electrocardiograms; a sample partitioning module for partitioning the training samples according to the target advance prediction time to obtain positive and negative hypotension prediction samples; and a model training module for training a basic time-series convolutional network based on the positive and negative samples, and determining the trained basic time-series convolutional network as the hypotension prediction model corresponding to the target advance prediction time. This invention can improve the accuracy of hypotension prediction.

Description

Technical Field

[0001] This invention relates to the field of health prediction technology, and in particular to a training device for a hypotension prediction model and a hypotension prediction device. Background Technology

[0002] During surgical anesthesia, close monitoring of multiple physiological parameters is essential to ensure patient safety. These parameters collectively constitute a comprehensive assessment of the patient's vital signs. Among them, blood pressure, as a core indicator reflecting hemodynamic stability and organ perfusion status, is crucial for stability. Intraoperative hypotension (IOH), typically defined as a sustained mean arterial pressure (MAP) below 65 mmHg, is one of the most common and serious complications during anesthesia. Numerous clinical studies have confirmed that IOH is closely associated with severe adverse outcomes such as postoperative acute kidney injury, myocardial injury, and even long-term mortality. Related data show that the incidence of IOH in general surgery is extremely high; in some studies, up to 88% of patients experience at least one IOH event. Its frequency and duration are key indicators of clinical harm. Therefore, effective prediction of IOH has become a core issue in improving perioperative patient safety.

[0003] Traditional intraoperative blood pressure management heavily relies on the anesthesiologist's clinical experience. Physicians directly observe blood pressure readings using intermittent or continuous blood pressure monitoring devices and judge the patient's physiological state based on their prior knowledge. This method is inherently retrospective, allowing only passive intervention after blood pressure has decreased, lacking the ability to proactively predict risks. With technological advancements, researchers have begun to explore using statistical methods or basic machine learning models (such as logistic regression) to analyze blood pressure data. However, these early computational methods are typically limited by two major bottlenecks: first, their performance heavily depends on tedious manual feature engineering, a process that is not only time-consuming and labor-intensive, but also often results in features that fail to capture complex and non-linear physiological dynamics; second, the data sources for analysis are often limited to single blood pressure parameters, failing to provide a comprehensive assessment of the patient's condition. Summary of the Invention

[0004] This invention provides a training device for a hypotension prediction model and a hypotension prediction device to improve the accuracy of hypotension prediction.

[0005] According to one aspect of the present invention, a method for training a hypotension prediction model is provided, comprising: The blood pressure parameters and waveform data in the database are preprocessed to obtain training samples; the waveform data includes photoplethysmography (PPG) and electrocardiogram (ECG). The training samples are divided according to the target prediction time to obtain positive and negative samples for low blood pressure prediction. The basic time-series convolutional network is trained based on the positive and negative samples, and the trained basic time-series convolutional network is determined as the hypotension prediction model corresponding to the target advance prediction time.

[0006] According to another aspect of the present invention, a method for predicting low blood pressure is provided, comprising: Preprocessing is performed on the blood pressure parameters and waveform data to be detected to obtain predictive features; Based on the expected advance prediction time, a target hypotension prediction model corresponding to the expected advance prediction time is selected from at least two candidate hypotension prediction models.

[0007] The predicted features are input into the target hypotension prediction model to obtain the probability of hypotension occurring within the expected advance prediction time output by the target hypotension prediction model. The hypotension prediction model is obtained by training the hypotension prediction model according to any of the methods described in this invention.

[0008] According to another aspect of the present invention, a training apparatus for a hypotension prediction model is provided, comprising: The sample generation module is used to preprocess blood pressure parameters and waveform data in the database to obtain training samples; the waveform data includes photoplethysmography (PPG) and electrocardiogram (ECG). The sample partitioning module is used to partition the training samples according to the target advance prediction time to obtain positive and negative samples for low blood pressure prediction. The model training module is used to train the basic time series convolutional network based on the positive and negative samples, and to determine the trained basic time series convolutional network as the hypotension prediction model corresponding to the target advance prediction time.

[0009] According to another aspect of the present invention, a hypotension prediction device is provided, comprising: The feature extraction module is used to preprocess the blood pressure parameters and waveform data to be detected to obtain predictive features; The model selection module is used to select a target hypotension prediction model from at least two candidate hypotension prediction models that corresponds to the expected advance prediction time.

[0010] The hypotension prediction module is used to input the prediction features into the target hypotension prediction model to obtain the probability of hypotension occurring within the expected advance prediction time output by the target hypotension prediction model. The hypotension prediction model is obtained by training the hypotension prediction model training device described in any embodiment of the present invention.

[0011] According to another aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements a training method for a hypotension prediction model or a hypotension prediction method according to any embodiment of the present invention.

[0012] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform a training method or a hypotension prediction method for a hypotension prediction model according to any embodiment of the present invention.

[0013] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the training method or hypotension prediction method of the hypotension prediction model according to any embodiment of the present invention.

[0014] This invention integrates traditional blood pressure parameters with dynamic physiological waveform data to capture complex physiological precursors that are difficult to detect using traditional methods, indicating that hypotension is about to occur. At the same time, it trains corresponding hypotension prediction models for different advance prediction time requirements. The combination of the two significantly improves the accuracy of hypotension prediction.

[0015] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the structure of a training device for a hypotension prediction model according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a hypotension prediction device according to another embodiment of the present invention. Figure 3 This is a flowchart of a training method for a hypotension prediction model according to another embodiment of the present invention; Figure 4 This is a flowchart of a hypotension prediction method provided according to another embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device that implements an embodiment of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0019] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] Figure 1 This is a schematic diagram of a training device for a hypotension prediction model according to an embodiment of the present invention. This embodiment is applicable to situations requiring advance prediction of the risk of hypotension occurring during surgery. The device can be implemented in hardware and / or software and can be configured in an electronic device with corresponding data processing capabilities. Figure 1 As shown, the device includes: The sample generation module 110 is used to preprocess the blood pressure parameters and waveform data in the database to obtain training samples.

[0021] The sample segmentation module 120 is used to segment the training samples according to the target advance prediction time to obtain positive and negative samples for low blood pressure prediction.

[0022] The model training module 130 is used to train the basic time series convolutional network based on the positive and negative samples, and to determine the trained basic time series convolutional network as the hypotension prediction model corresponding to the target advance prediction time.

[0023] The waveform data includes photoplethysmography (PLETH) and electrocardiogram (ECG).

[0024] Specifically, the database contains multiple patient surgical data entries, each recording the patient's blood pressure parameters and waveform data during the surgery. Each surgical data entry undergoes preprocessing, including: outlier handling: filtering blood pressure parameters based on clinically recognized physiological threshold ranges to remove outliers exceeding these ranges; signal integrity checking: identifying and removing flat segments in the waveform data caused by noise interference, artifacts, or sensor malfunctions by evaluating waveform amplitude and analyzing zero crossover rate; and hypotension event labeling: using a mean arterial pressure <65 mmHg as a basis for defining hypotension events, labeling each surgical data entry window by window to generate hypotension event tags. After preprocessing, each surgical data entry is used as a training sample. To facilitate data input into the model, each training sample can be further divided into multiple fixed time windows (e.g., 30 seconds) of sample segments.

[0025] To support hypotension prediction at different advance prediction time points (e.g., 5, 10, and 15 minutes in advance), this invention selects to train a separate hypotension prediction model for each advance prediction time. After determining the target advance prediction time point from multiple advance prediction time points, the training samples are divided according to the target advance prediction time point to determine the positive and negative samples applicable to the target advance prediction time point.

[0026] The basic time-series convolutional network is trained using positive and negative samples until it passes a test. The tested basic time-series convolutional network is then identified as the hypotension prediction model corresponding to the target advance prediction time. The model is trained using a preset optimizer with a binary cross-entropy loss function and a maximum training epoch of 1000. The model gradually converges on the training set, indicating that it can accurately perform the risk prediction task for hypotension events. Each advance prediction time point is sequentially used as the target advance prediction time point, and the above steps are repeated to obtain the hypotension prediction model corresponding to each advance prediction time point.

[0027] This invention integrates traditional blood pressure parameters with dynamic physiological waveform data to capture complex physiological precursors that are difficult to detect using traditional methods, indicating that hypotension is about to occur. At the same time, it trains corresponding hypotension prediction models for different advance prediction time requirements. The combination of the two significantly improves the accuracy of hypotension prediction.

[0028] Based on the above embodiments, optionally, the sample segmentation module 120 includes: The first sample partitioning unit is used to identify positive samples from the training samples where low blood pressure occurs, within the time frame before the target prediction occurs. The second sample partitioning unit is used to identify sample segments in the training samples where hypotension did not occur as negative samples.

[0029] Specifically, using the occurrence time of a hypotension event as a reference, for training samples where hypotension occurs, the sample segments within the target advance prediction time before the hypotension event are labeled as positive samples. For training samples where hypotension does not occur, all sample segments of the training sample are designated as negative samples. By constructing positive and negative labels for different advance prediction time points, training and validation for multiple prediction advance time points can be achieved without changing the model structure, simply by using different labels.

[0030] Based on the above embodiments, optionally, the basic time series convolutional network includes a preliminary feature extraction module, a residual connection module, a feature reprocessing module, and an integrated classification output module; The preliminary feature extraction module includes a first convolutional block and a second convolutional block, used for initial feature extraction; The residual connection module includes residual blocks composed of transform networks, which are used to enhance feature representation and transfer network gradients. The feature reprocessing module includes a third convolutional block and a fourth convolutional block, used for feature refinement and channel compression; The integrated classification output module includes an adaptive average pooling layer and a fully connected layer for outputting the probability of low blood pressure.

[0031] Specifically, the basic time-series convolutional network constructed in this invention is a time-series convolutional network (TS-CNN). Its overall architecture consists of four core parts connected in series: (1) a preliminary feature extraction module, which consists of two concatenated convolutional blocks; (2) a residual connection module, which is used to deepen the network and optimize gradient propagation; (3) a subsequent feature processing module, which consists of two other concatenated convolutional blocks, responsible for feature refinement and dimensionality reduction; and (5) a global integration and classification output module, which consists of a global feature integration module and a classification output module, used to generate the final prediction result.

[0032] Preliminary feature extraction module: contains two concatenated initial convolutional blocks, namely the first convolutional block and the second convolutional block. The input of the first convolutional block is positive and negative samples, and the output of the first convolutional block is directly used as the input of the second convolutional block, so as to achieve the gradual deepening of feature representation. At the same time, both convolutional blocks are processed based on the same multi-channel input tensor. Through the layer-by-layer convolutional operation, the temporal features from local to high order are captured step by step. The internal processing flow of each convolutional block is as follows: (1) One-dimensional convolution: a one-dimensional convolutional kernel is used to slide the input sequence along the time axis to extract local temporal patterns. For the input sequence s[t] and the convolutional kernel k[τ] of length Lk, its output o[t] can be expressed as o[t]=τ=0Lk-1s[t-τ]·k[τ]. The first convolutional block uses 65 convolutional kernels, and the second convolutional block uses 128 convolutional kernels. The kernel size is 3 and the stride is 1. (2) Batch normalization: Normalizes the output after convolution to accelerate model convergence and improve training stability. (3) ReLU activation: Introduces nonlinearity to enable the network to learn more complex features. Its function is f(x)=max(0,x). (5) Max pooling: Performs downsampling to retain the most significant features and reduce data dimensionality. The pooling kernel size is 2 and the stride is 2. (5) Dropout: Randomly deactivates some neurons with a probability of 0.2 as a regularization method to prevent overfitting.

[0033] Residual Connection Module: After initial feature extraction, the model integrates a residual block. The core of this module is a shortcut connection that directly adds the module's input x (i.e., the output of the first convolutional block) to the output F(x) of the internal transform layer network, obtaining the final output H(x) = F(x) + x. This design effectively alleviates the vanishing gradient problem in deep networks, allowing the network to learn deeper feature representations. The internal transform network F(x) consists of two convolutional layers (128 kernels) and corresponding batch normalization and ReLU activation functions. Similarly, the two convolutional layers in the structure are connected in series, meaning the output of the previous convolutional layer serves as the input to the next two. The residual path retains the module input and fuses it element-wise with the output after convolutional transformation, thereby enhancing feature representation and efficiently transferring network gradients.

[0034] Subsequent Feature Processing Module: After the residual block, the feature map output by the residual block enters a subsequent processing module consisting of two more cascaded convolutional blocks (the third and fourth convolutional blocks). Structurally similar to the initial feature extraction module, this module's core function shifts to feature refinement and channel compression. Specifically, these two convolutional blocks sequentially refine key local temporal patterns and perform progressive dimensionality reduction: the third convolutional block uses 256 convolutional kernels, while the fourth convolutional block reduces this to 128. This design enhances features relevant to the prediction target and ensures that the refined high-level features can be input into the subsequent global pooling layer in a more compact form, preparing for the final global information integration.

[0035] Global Feature Integration Module: Based on the feature map output from the fourth convolutional block, the model introduces an adaptive average pooling layer for global integration. This layer converts the variable-length feature map output from the preceding modules into a fixed-dimensional feature vector. This layer averages all values ​​within each channel along the time dimension, compressing the input feature map (dimension C × time step T) into a fixed-length vector that is only related to the number of channels. This operation achieves efficient aggregation of global temporal information while avoiding the additional parameter overhead caused by changes in sequence length, thus improving the model's generalization ability.

[0036] Classification Output Module: This module consists of a fully connected layer and a final activation function: (1) Fully Connected Layer: The fixed-dimensional feature vector output by the adaptive average pooling layer is linearly transformed through the fully connected layer to achieve global integration of features and map them to the final output dimension, thereby obtaining a predicted score that represents the probability of future hypotension. Then, a Dropout layer (probability p=0.5) is connected to reduce the risk of overfitting. (2) Sigmoid Activation Function: The output layer uses the Sigmoid activation function to perform nonlinear mapping on the predicted score output by the fully connected layer, compressing it to a probability value in the [0,1] interval. This probability is used to represent the possibility of a hypotension event in the future, and 0.5 is used as the judgment threshold: when the probability is greater than 0.5, a hypotension warning is output; otherwise, it is judged that no hypotension has occurred.

[0037] Optionally, based on the above embodiments, the apparatus further includes: The parameter conversion module is used to interpolate or fill the blood pressure parameters in each positive or negative sample to obtain a long-term sequence with the same length as the waveform data in the sample. The parameter replacement module is used to replace the blood pressure parameters in the sample with the long-term series.

[0038] Specifically, to address the difficulty in integrating blood pressure parameters (mean arterial pressure SBP, mean arterial pressure MBP) with dynamic waveform data (photoplethysmography and electrocardiogram), this invention performs multimodal signal integration before the data enters the model. This mainly includes sequence length equalization: interpolating / filling the mean arterial pressure and mean arterial pressure (both are generally discretely sampled numerical signals) into a sequence of equal length, making its length consistent with the waveform sequence of the photoplethysmography and electrocardiogram, and replacing the original mean arterial pressure and mean arterial pressure with the newly obtained long-time sequence before inputting it into the model.

[0039] Based on the above embodiments, optionally, the blood pressure parameters include mean arterial pressure and mean arterial pressure; the mean arterial pressure, systolic pressure, photoplethysmography (PPG) and electrocardiogram (ECG) each correspond to an input tensor of the basic time series convolutional network.

[0040] Specifically, the input tensor is organized by channel: the input tensor of the model is constructed with signal type as an independent channel, and finally forms a multi-channel (mean arterial pressure + systolic blood pressure + photoplethysmography pulse wave + electrocardiogram) time series input, realizing the structural unification and fusion of physiological signals of different modalities.

[0041] For example, to evaluate the predictive performance of the trained model under different configurations and the contribution of each input parameter, a master experiment and an ablation experiment were conducted. In this experiment, the prediction task was a binary classification, namely "hypotension occurred" (positive sample) and "hypotension did not occur" (negative sample). The model's performance on the test set will be listed below. We compared the performance of the complete model integrating four physiological parameters with the baseline model using only two traditional blood pressure parameters (mean arterial pressure, mean arterial pressure), and the results are shown in Table 1 below: Table 1. Performance of different parameter combination models at different advance prediction times. To investigate the contributions of PLETH and ECG signals, ECG or PLETH was removed from the complete four-parameter model, respectively, to test the changes in model performance. The results are shown in Table 2 below: Table 2 In the main experiment with a 10-minute prediction window, the four-parameter model proposed in this invention performed excellently, with an AUROC of 0.8734, which is significantly higher than the baseline model using only traditional blood pressure parameters (0.8567), demonstrating the superiority of multimodal fusion.

[0042] In the ablation experiments, the performance changes of the model further revealed its internal mechanism. When the electrocardiogram (ECG) was removed from the complete model, the AUROC dropped significantly from 0.8734 to 0.8629; the effect of removing the photoplethysmography (PPG) wave was even greater, indicating that both of the newly added physiological signals made significant contributions to the prediction. Particularly noteworthy is the strong positive synergistic effect of PPG wave and ECG, where the combined performance improvement far exceeds the sum of their individual contributions. This demonstrates that the model of this invention is efficient and robust, capable of accurately identifying the risk of hypotension by deeply integrating information from different physiological dimensions.

[0043] Figure 2 This is a schematic diagram of a training device for a hypotension prediction model according to an embodiment of the present invention. This embodiment is applicable to situations requiring advance prediction of the risk of hypotension occurring during surgery. The device can be implemented in hardware and / or software and can be configured in an electronic device with corresponding data processing capabilities. Figure 2 As shown, the device includes: Feature extraction module 210 is used to preprocess the blood pressure parameter to be detected and the waveform data to be detected to obtain predicted features; The model selection module 220 is used to select a target hypotension prediction model corresponding to the expected advance prediction time from at least two candidate hypotension prediction models based on the expected advance prediction time.

[0044] Hypotension prediction module 230 is used to input the prediction features into the target hypotension prediction model to obtain the probability of hypotension occurring within the expected advance prediction time output by the target hypotension prediction model. The hypotension prediction model is obtained by training the hypotension prediction model training device described in any embodiment of the present invention.

[0045] Specifically, when hypotension prediction is required, on the one hand, blood pressure parameters and waveform data of the patient over a period of time are collected as the blood pressure parameters and waveform data to be detected, and these are preprocessed to obtain the predictive features to be input into the model. On the other hand, based on the expected advance prediction time selected by medical staff, a target hypotension prediction model corresponding to the expected advance prediction time is selected from at least two candidate hypotension prediction models. The predictive features are input into the target hypotension prediction model, which analyzes the predictive features and outputs the probability of hypotension occurrence within the expected advance prediction time. The embodiments of the present invention combine blood pressure parameters and waveform data when predicting hypotension, and select a target hypotension prediction model that meets the prediction requirements to predict the probability of hypotension occurrence, which greatly improves the accuracy of hypotension prediction.

[0046] Figure 3This is a flowchart of a training method for a hypotension prediction model according to another embodiment of the present invention, the method comprising: S310. Preprocess the blood pressure parameters and waveform data in the database to obtain training samples.

[0047] S320. The training samples are divided according to the target advance prediction time to obtain positive and negative samples for low blood pressure prediction.

[0048] S330. Train the basic time series convolutional network based on the positive and negative samples, and determine the trained basic time series convolutional network as the hypotension prediction model corresponding to the target advance prediction time.

[0049] The waveform data includes photoplethysmography (PPG) and electrocardiogram (ECG).

[0050] The specific training process of the hypotension prediction model can be found in other embodiments of the present invention, and will not be repeated here.

[0051] This invention integrates traditional blood pressure parameters with dynamic physiological waveform data to capture complex physiological precursors that are difficult to detect using traditional methods, indicating that hypotension is about to occur. At the same time, it trains corresponding hypotension prediction models for different advance prediction time requirements. The combination of the two significantly improves the accuracy of hypotension prediction.

[0052] Figure 4 This is a flowchart of a hypotension prediction method according to another embodiment of the present invention, the method comprising: S410. Preprocess the blood pressure parameters and waveform data to be detected to obtain predictive features.

[0053] S420. Select the target hypotension prediction model corresponding to the expected advance prediction time from at least two candidate hypotension prediction models based on the expected advance prediction time.

[0054] S430. Input the predicted features into the target hypotension prediction model to obtain the probability of hypotension occurring within the expected advance prediction time output by the target hypotension prediction model.

[0055] The hypotension prediction model is trained using the hypotension prediction model training method described in any embodiment of the present invention.

[0056] The specific process for predicting low blood pressure is described in other embodiments of the present invention, and will not be repeated here.

[0057] The embodiments of the present invention combine blood pressure parameters and waveform data when predicting hypotension, and select a target hypotension prediction model that meets the prediction requirements to predict the probability of hypotension occurrence, which greatly improves the accuracy of hypotension prediction.

[0058] Figure 5 A schematic diagram of an electronic device 50 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0059] like Figure 5 As shown, the electronic device 50 includes at least one processor 51 and a memory, such as a read-only memory (ROM) 52 or a random access memory (RAM) 53, communicatively connected to the at least one processor 51. The memory stores computer programs executable by the at least one processor. The processor 51 can perform various appropriate actions and processes based on the computer program stored in the ROM 52 or loaded from storage unit 58 into the RAM 53. The RAM 53 can also store various programs and data required for the operation of the electronic device 50. The processor 51, ROM 52, and RAM 53 are interconnected via a bus 54. An input / output (I / O) interface 55 is also connected to the bus 54.

[0060] Multiple components in electronic device 50 are connected to I / O interface 55, including: input unit 56, such as keyboard, mouse, etc.; output unit 57, such as various types of monitors, speakers, etc.; storage unit 58, such as disk, optical disk, etc.; and communication unit 59, such as network card, modem, wireless transceiver, etc. Communication unit 59 allows electronic device 50 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0061] Processor 51 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 51 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 51 performs the various methods and processes described above, such as training methods for hypotension prediction models or hypotension prediction methods.

[0062] In some embodiments, the training method for the hypotension prediction model or the hypotension prediction method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 58. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 50 via ROM 52 and / or communication unit 59. When the computer program is loaded into RAM 53 and executed by processor 51, one or more steps of the training method for the hypotension prediction model or the hypotension prediction method described above may be performed. Alternatively, in other embodiments, processor 51 may be configured to perform the training method for the hypotension prediction model or the hypotension prediction method by any other suitable means (e.g., by means of firmware).

[0063] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0064] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0065] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0066] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0067] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0068] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0069] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0070] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A training device for a hypotension prediction model, characterized in that, The device includes: The sample generation module is used to preprocess blood pressure parameters and waveform data in the database to obtain training samples; the waveform data includes photoplethysmography (PPG) and electrocardiogram (ECG). The sample partitioning module is used to partition the training samples according to the target prediction time to obtain positive and negative samples for low blood pressure prediction. The model training module is used to train the basic time series convolutional network based on the positive and negative samples, and to determine the trained basic time series convolutional network as the hypotension prediction model corresponding to the target advance prediction time.

2. The apparatus according to claim 1, characterized in that, The sample partitioning module includes: The first sample partitioning unit is used to identify positive samples from the training samples where low blood pressure occurs, within the time frame before the target prediction occurs. The second sample partitioning unit is used to identify sample segments in the training samples where hypotension did not occur as negative samples.

3. The apparatus according to claim 1, characterized in that, The basic time-series convolutional network includes a preliminary feature extraction module, a residual connection module, a feature reprocessing module, and an integrated classification output module; The preliminary feature extraction module includes a first convolutional block and a second convolutional block, used for initial feature extraction; The residual connection module includes residual blocks composed of transform networks, which are used to enhance feature representation and transfer network gradients. The feature reprocessing module includes a third convolutional block and a fourth convolutional block, used for feature refinement and channel compression; The integrated classification output module includes an adaptive average pooling layer and a fully connected layer for outputting the probability of low blood pressure.

4. The apparatus according to claim 1, characterized in that, The device further includes: The parameter conversion module is used to interpolate or fill the blood pressure parameters in each positive or negative sample to obtain a long-term sequence with the same length as the waveform data in the sample. The parameter replacement module is used to replace the blood pressure parameters in the sample with the long-term series.

5. The apparatus according to claim 1, characterized in that, The blood pressure parameters include mean arterial pressure (MAP) and mean arterial pressure (MAP). The MAP, systolic blood pressure, photoplethysmography (PPG), and electrocardiogram (ECG) each correspond to an input tensor of a basic time-series convolutional network.

6. A hypotension prediction device, characterized in that, The device includes: The feature extraction module is used to preprocess the blood pressure parameters and waveform data to be detected to obtain predictive features; The model selection module is used to select a target hypotension prediction model corresponding to the expected advance prediction time from at least two candidate hypotension prediction models based on the expected advance prediction time. The hypotension prediction module is used to input the prediction features into the target hypotension prediction model to obtain the probability of hypotension occurring within the expected advance prediction time output by the target hypotension prediction model. The hypotension prediction model is obtained by training the hypotension prediction model according to claim 1.

7. A training method for a hypotension prediction model, characterized in that, The method includes: The blood pressure parameters and waveform data in the database are preprocessed to obtain training samples; the waveform data includes photoplethysmography (PPG) and electrocardiogram (ECG). The training samples are divided according to the target prediction time to obtain positive and negative samples for low blood pressure prediction. The basic time-series convolutional network is trained based on the positive and negative samples, and the trained basic time-series convolutional network is determined as the hypotension prediction model corresponding to the target advance prediction time.

8. A method for predicting hypotension, characterized in that, The method includes: Preprocessing is performed on the blood pressure parameters and waveform data to be detected to obtain predictive features; Select the target hypotension prediction model from at least two candidate hypotension prediction models based on the expected advance prediction time. The predicted features are input into the target hypotension prediction model to obtain the probability of hypotension occurring within the expected advance prediction time output by the target hypotension prediction model. The hypotension prediction model is trained using the hypotension prediction model training method described in claim 7.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the training method of the hypotension prediction model of claim 7 or the hypotension prediction method of claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the training method of the hypotension prediction model of claim 7 or the hypotension prediction method of claim 8.

Images