Deep learning model-based airway pressure waveform data prediction method and system
Through deep learning models, especially LSTM and Transformer models, real-time acquisition and personalization of fine-tuning of airway pressure waveform data is solved, and high-precision and flexible prediction of airway pressure waveform data is achieved.
Patent Information
- Application Number
- CN202510277293.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-10
- Publication Date
- 2025-07-11
AI Technical Summary
In the prior art, the prediction of airway pressure waveform data mainly relies on fixed mathematical models and cannot process complex, nonlinear data, resulting in low prediction accuracy, especially in the face of irregular waveform data and dynamic breathing cycles.
Deep learning models, especially LSTM and Transformer models, are used to obtain the patient's airway pressure waveform data in real time, extract key parameters such as change rate, peak value and valley value, and perform personalized fine-tuning to generate airway pressure waveform data at the set time point or time period in the future.
It improves the prediction accuracy and applicability of airway pressure waveform data, can dynamically adapt to the physiological characteristics of different patients, supports flexible predictions at time points and time periods, and improves the flexibility and accuracy of prediction.
Smart Images
Figure CN120296382A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of artificial intelligence technology, and particularly relates to a prediction method, system, medium and device for airway pressure waveform data based on a deep learning model. Background Art
[0002] Medical ventilators (including medical non-invasive ventilators, medical invasive ventilators, etc.) are key devices in modern medicine for assisting or completely replacing a patient's spontaneous breathing. Medical ventilators monitor parameters such as airway pressure, flow rate, and volume in real time through various sensors, thereby generating waveform data within a certain period. These waveform data are of great significance for doctors to judge the patient's breathing condition and adjust the treatment plan. Among them, airway pressure waveform data is a key reference index for evaluating the working state of the ventilator and the patient's airway resistance. Therefore, the prediction of airway pressure waveform data is particularly important.
[0003] In traditional solutions, the prediction of airway pressure waveform data mainly relies on fixed mathematical models. Although it can reflect the future trend of waveform data to a certain extent, due to its inherent linear assumption and limited feature extraction ability, it cannot handle complex and non-linear data. In addition, when facing irregular waveform data, these methods have low prediction accuracy and cannot be applied to the changing patterns in the dynamic breathing cycle, with low applicability. In existing solutions, linear regression or random forest models are mainly used to predict respiratory waveform data. However, these solutions are still limited by the complexity of respiratory waveform data and are difficult to handle non-linear, long-time-span, and dynamically changing data, resulting in low prediction accuracy. Summary of the Invention
[0004] The purpose of this application is to overcome the defects existing in the prior art.
[0005] To achieve the above purpose, this application proposes a prediction method, system, medium and device for airway pressure waveform data based on a deep learning model.
[0006] In the first aspect, the prediction method for airway pressure waveform data based on a deep learning model proposed by this application includes:
[0007] Acquisition step: Real-time acquire the first airway pressure waveform data generated after a current patient uses a medical ventilator;
[0008] Input step: Input the first airway pressure waveform data into a trained deep learning model;
[0009] Generation step: The trained deep learning model extracts the key parameters corresponding to the first airway pressure waveform data and generates the second airway pressure waveform data at a future set time point or within a set time period according to the key parameters, where the key parameters at least include the change rate, peak value, and valley value.
[0010] In some examples, the deep learning model includes a first model and a second model.
[0011] In some examples, the prediction method further includes:
[0012] According to the size of the set time period, select the first model or the second model to execute the generation step.
[0013] In some examples, the prediction method further includes:
[0014] In the case where the set time period is less than the set threshold, select the first model to execute the generation step, where the first model is an LSTM model;
[0015] In the case where the set time period is not less than the set threshold, select the second model to execute the generation step, where the second model is a Transformer model.
[0016] In some examples, the training process of the trained deep learning model includes:
[0017] Extract the change rate, peak value, and valley value corresponding to the third airway pressure waveform data generated by multiple historical patients after using a medical ventilator respectively, and generate a first training dataset;
[0018] Input the first training dataset into the deep learning model to train the deep learning model, and obtain an initially trained deep learning model;
[0019] Obtain the fourth airway pressure waveform data generated by the current patient after using a medical ventilator within a historical set time period and extract the change rate, peak value, and valley value corresponding to the third airway pressure waveform data, and generate a second training dataset;
[0020] Input the second training dataset into the initial deep learning model to train the initial deep learning model, and obtain a trained deep learning model.
[0021] In some examples, after inputting the second training dataset into the initial deep learning model to train the initial deep learning model and obtaining a trained deep learning model, the prediction method further includes:
[0022] Obtain the airway pressure waveform data predicted by the trained deep learning model in real time;
[0023] Compare the predicted airway pressure waveform prediction data with the actual airway pressure waveform data to obtain a comparison result;
[0024] Optimize the trained deep learning model according to the comparison result.
[0025] In some examples, before inputting the first airway pressure waveform data into the trained deep learning model, the method further includes:
[0026] Eliminate the outliers in the first airway pressure waveform data;
[0027] Fill in the missing values in the first airway pressure waveform data after eliminating the outliers;
[0028] Normalize the first airway pressure waveform data after filling in the missing values;
[0029] Use a fixed-length sliding window to segment the airway pressure waveform data, where the size of the sliding window matches the size of the target prediction time period.
[0030] In a second aspect, the prediction system for airway pressure waveform data based on a deep learning model provided by the present application includes:
[0031] An acquisition module, configured to acquire in real time the first airway pressure waveform data generated after a current patient uses a medical ventilator;
[0032] An input module, configured to input the first airway pressure waveform data into the trained deep learning model;
[0033] A generation module, configured to extract key parameters corresponding to the first airway pressure waveform data and generate second airway pressure waveform data at a future set time point or within a set time period according to the key parameters, where the key parameters at least include a change rate, a peak value, and a valley value.
[0034] In a third aspect, the present application further provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the computer program, the method described in the first aspect above is implemented.
[0035] In a fourth aspect, the present application further provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is caused to execute the method described in the first aspect above.
[0036] Compared with the prior art, the advantages of the present application are:
[0037] (1) By adopting deep learning models (LSTM model and Transformer model) for intelligent prediction, based on multiple key parameters such as the change rate, peak value, and valley value corresponding to the airway pressure waveform data, the accuracy of prediction is improved;
[0038] (2) By collecting the individual data of the current patient in real time and using this data to perform personalized fine-tuning on the deep learning model, the prediction model can dynamically adapt to the physiological characteristics of different patients, with strong applicability. This personalized fine-tuning method can further improve the accuracy of prediction;
[0039] (3) According to actual needs, automatically select the LSTM model and the Transformer model for prediction, and support time point prediction or time period prediction, with high flexibility and strong applicability. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 The following shows a schematic flowchart of the airway pressure waveform data prediction method based on a deep learning model provided by an embodiment of the present application.
[0041] Figure 2 The following shows a block diagram of the structure of the airway pressure waveform data prediction system based on a deep learning model provided by an embodiment of the present disclosure.
[0042] Figure 3 The following shows a schematic block diagram of an electronic device as a classical computing device according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0044] Embodiment 1
[0045] As Figure 1 shown, the prediction method of airway pressure waveform data based on a deep learning model provided by an embodiment of the present invention includes the following steps:
[0046] S101, Acquisition step: Real-time acquisition of the first airway pressure waveform data generated after the current patient uses a medical ventilator.
[0047] Specifically, the medical ventilator monitors the current airway pressure waveform data of the patient in real time through its built-in gas pressure sensor.
[0048] S102, input step: input the first airway pressure waveform data into a trained deep learning model.
[0049] In some examples, the deep learning model includes a first model and a second model.
[0050] In some examples, the training process of the trained deep learning model includes:
[0051] The change rate, peak value and valley value corresponding to the third airway pressure waveform data generated by multiple patients using medical ventilators in the history are extracted respectively to generate a first training data set, wherein the quality of the training data is ensured by extracting multiple key features of the airway pressure waveform data, so as to better train the model and improve the prediction accuracy of the model.
[0052] Inputting the first training data set into a deep learning model to train the deep learning model, thereby obtaining a trained initial deep learning model;
[0053] The fourth airway pressure waveform data generated by the current patient after using the medical ventilator within the historical set time period is obtained, and the change rate, peak value and valley value corresponding to the third airway pressure waveform data are extracted to generate a second training data set. Specifically, the current patient is a special case patient. Based on the airway pressure waveform data of the current patient, the model is fine-tuned to adapt to different individuals, which can further improve the prediction accuracy.
[0054] The second training data set is input into the initial deep learning model to train the initial deep learning model to obtain a trained deep learning model.
[0055] In some examples, before inputting the first airway pressure waveform data into the trained deep learning model, the method further includes:
[0056] Eliminating abnormal values in the first airway pressure waveform data;
[0057] Filling missing values in the first airway pressure waveform data after removing outliers, wherein by removing outliers and filling missing values, noise interference is reduced to ensure high-quality training data;
[0058] The first airway pressure waveform data after filling the missing values is normalized, wherein the airway pressure data is normalized or standardized, and the values of each data are usually adjusted to between [0,1] to better adapt to the deep learning model. Specifically, the normalization formula is as follows:
[0059] x norm =(xx min ) / x max -x min (1)
[0061] In formula (1), x is the time series data, x max is the data with the largest value in the time series data, x min is the data with the smallest value in the time series data, x norm is the normalized data.
[0062] The airway pressure waveform data is segmented using a sliding window of fixed length, where the size of the sliding window matches the length of the target prediction time period. For example, when predicting the pressure waveform for the next 30 seconds, the length of the sliding window is set to 30 seconds. The user can also adjust the window size according to the needs to capture a sufficient amount of airway pressure waveform data as training data.
[0063] Specifically, the LSTM model has good performance in predicting time series data. Especially for the waveform data changes in a short time period, it can well capture the dependency relationship and has high prediction accuracy. Its structure is as follows:
[0064] Input layer: The preprocessed airway pressure waveform data is input into the model.
[0065] Recurrent layer: Captures the dependency relationship of time series data through a recursive structure.
[0066] Fully connected layer: Maps the output of the recurrent layer to the predicted airway pressure value.
[0067] The Transformer model can effectively capture the dependency relationship of long time series data, so as to be used for the prediction of more complex and long sequence airway pressure waveform data with high prediction accuracy. Its structural characteristics include:
[0068] Embedding layer: Embeds the input data into a vector of fixed dimension, including the historical airway pressure data and patient individual information.
[0069] Position encoding layer: Injects the position information in the time series data into the model through position encoding to make up for the lack of built-in time order in the Transformer model.
[0070] Self-attention mechanism layer: Used to capture the global dependency relationship between different time points in the time series.
[0071] Specifically, the prediction formula of the LSTM model is:
[0072]
[0073] In formula (2), is the predicted future airway pressure value, and Δt can be customarily set according to the user's needs.
[0074] Equation (2) is interpreted as: the input of the model is the airway pressure values in the past T seconds, and the output is the predicted airway pressure values in a future period of time.
[0075] The formula corresponding to the self-attention mechanism layer of the Transformer model is:
[0076]
[0077] In Equation (3), Q, K, and V represent the query matrix, the key matrix, and the value matrix respectively, and d k is the dimension. This formula can effectively capture the dependencies in time series data and improve the prediction accuracy.
[0078] The formula corresponding to the position encoding layer of the Transformer model is:
[0079]
[0080] In Equation (4), pos is the encoding position, d is the embedding dimension. In the present invention, d is selected as the empirical value 128, i is the index dimension, and i in the position encoding changes with the change of d and does not need to be specially set. As long as d is determined, i will be automatically calculated according to the formula.
[0081] S103, Generation step: The trained deep learning model extracts the key parameters corresponding to the first airway pressure waveform data and generates the second airway pressure waveform data at a future set time point or within a set time period according to the key parameters, where the key parameters at least include: the change rate, the peak value, and the valley value.
[0082] In some examples, the prediction method further includes: selecting the first model or the second model to perform the above generation step according to the size of the set time period.
[0083] In some instances, the prediction method further includes:
[0084] In the case where the set time period is less than the set threshold, select the first model to perform the generation step, where the first model is an LSTM model;
[0085] In the case where the set time period is not less than the set threshold, select the second model to perform the generation step, where the second model is a Transformer model.
[0086] Specifically, when the set time period is less than 2 minutes, the LSTM model is selected to execute the generation step. When the set time period is not less than 2 minutes, the Transformer model is selected to execute the generation step. According to actual requirements, the LSTM model or the Transformer model is automatically selected for prediction, and time point prediction or time period prediction is supported, with high flexibility and strong applicability.
[0087] In some examples, after inputting the second training data set into the initial deep learning model to train the initial deep learning model to obtain a trained deep learning model, the method further includes:
[0088] Obtaining in real time the airway pressure waveform data predicted by the trained deep learning model;
[0089] Comparing the predicted airway pressure waveform prediction data with the actual airway pressure waveform data to obtain a comparison result;
[0090] Optimizing the trained deep learning model according to the comparison result.
[0091] Specifically, an enhanced learning mechanism is adopted to further optimize the model by providing real-time feedback on the difference between the actual airway pressure waveform data and the predicted airway pressure waveform data, continuously improving the prediction accuracy.
[0092] Embodiment 2
[0093] As Figure 2 shown, the airway pressure waveform data prediction system based on a deep learning model provided by an embodiment of the present invention includes:
[0094] An acquisition module configured to acquire in real time first airway pressure waveform data generated after a current patient uses a medical ventilator;
[0095] An input module configured to input the first airway pressure waveform data into a trained deep learning model;
[0096] A generation module configured to, based on the first airway pressure waveform data, extract key parameters corresponding to the first airway pressure waveform data and, according to the key parameters, generate second airway pressure waveform data at a future set time point or within a set time period, where the key parameters at least include: a change rate, a peak value, and a valley value.
[0097] Embodiment 3
[0098] Figure 3The block diagram of the hardware structure of an embodiment of the electronic device provided by the present invention is shown. The electronic device includes a processor 601 and a memory 602 storing computer program instructions. When the processor executes the computer program instructions, the methods disclosed in the embodiments of the present invention above are implemented.
[0099] Specifically, the above-mentioned processor 601 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention.
[0100] The memory 602 may include a mass storage for data or instructions. By way of example and not limitation, the memory 602 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disc, a magneto-optical disc, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. In a suitable case, the memory 602 may include a removable or non-removable (or fixed) medium. In a suitable case, the memory 602 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 602 is a non-volatile solid state memory.
[0101] The memory may include a read only memory (ROM), a random access memory (RAM), a magnetic disk storage media device, an optical storage media device, a flash memory device, an electrical, optical, or other physical / tangible memory storage device. Thus, in general, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described in reference to the method according to one aspect of the present disclosure.
[0102] In some examples, as Figure 3 shown, the electronic device may further include a communication interface 603 and a bus 610. Among them, as Figure 3 shown, the processor 601, the memory 602, and the communication interface 603 are connected through the bus 610 and complete communication with each other. The electronic device in the embodiments of the present invention may be a local server or other computing devices, or may be a cloud server.
[0103] The communication interface 603 is mainly used to implement communication between the various modules, devices, units, and / or devices in the embodiments of the present invention.
[0104] The bus 610 includes hardware, software, or both, and couples the components of the online data flow metering device to each other. By way of example and not limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-E) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, the bus 610 may include one or more buses. Although embodiments of the present invention describe and illustrate specific buses, the present invention contemplates any suitable bus or interconnect.
[0105] The present invention also provides a computer program product, which includes computer program instructions that, when executed by a processor, implement the method provided by the embodiments of the present invention. Such computer program product may be, for example, a software installation package, a plug-in compatible with a related software system, etc.
[0106] It should be clear that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order between steps after understanding the spirit of the present invention.
[0107] In addition, the "first", "second", and similar terms used in the present disclosure do not denote any order, quantity, or importance, but are only used to distinguish different parts.
[0108] As described above, the foregoing are only specific embodiments of the present invention. Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, modules, and units described above may refer to the processes in the foregoing method embodiments and will not be elaborated here. It should be understood that the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or substitutions within the technical scope disclosed by the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.
Claims
1. A prediction method for airway pressure waveform data based on a deep learning model, comprising: An acquisition step: acquiring in real time first airway pressure waveform data generated after a current patient uses a medical ventilator; An input step: inputting the first airway pressure waveform data into a trained deep learning model; A generation step: the trained deep learning model extracts key parameters corresponding to the first airway pressure waveform data and generates second airway pressure waveform data at a future set time point or within a set time period according to the key parameters, wherein the key parameters at least include a change rate, a peak value, and a valley value.
2. The prediction method according to claim 1, wherein The deep learning model includes a first model and a second model different from the first model.
3. The prediction method according to claim 2, wherein It further includes: Selecting the first model or the second model to execute the generation step according to the size of the set time period.
4. The airway pressure waveform data prediction method based on a deep learning model according to claim 3, characterized in that, It further includes: In the case where the set time period is less than a set threshold, selecting the first model to execute the generation step, wherein the first model is an LSTM model; In the case where the set time period is not less than the set threshold, selecting the second model to execute the generation step, wherein the second model is a Transformer model.
5. The prediction method according to claim 1, wherein The training process of the trained deep learning model includes: Respectively extracting the change rate, peak value, and valley value corresponding to third airway pressure waveform data generated after multiple historical patients use a medical ventilator to generate a first training data set; Inputting the first training data set into the deep learning model to train the deep learning model to obtain a trained initial deep learning model; Acquiring fourth airway pressure waveform data generated after the current patient uses a medical ventilator within a historical set time period and extracting the change rate, peak value, and valley value corresponding to the fourth airway pressure waveform data to generate a second training data set; Inputting the second training data set into the initial learning model to train the initial deep learning model to obtain a trained deep learning model.
6. The prediction method according to claim 4, wherein After inputting the second training data set into the initial deep learning model to train the initial deep learning model to obtain a trained deep learning model, the method further includes: Acquiring in real time the airway pressure waveform data predicted by the trained deep learning model; Comparing the predicted airway pressure waveform prediction data with the actual airway pressure waveform data to obtain a comparison result; Optimizing the trained deep learning model according to the comparison result.
7. The prediction method according to claim 1, wherein Before inputting the first airway pressure waveform data into the trained deep learning model, the prediction method further includes: Removing outliers from the first airway pressure waveform data; Filling in missing values in the first airway pressure waveform data after removing outliers; Normalizing the first airway pressure waveform data after filling in missing values; Using a sliding window with a fixed length to segment the airway pressure waveform data, wherein the size of the sliding window matches the size of the target prediction time period.
8. A prediction system for airway pressure waveform data based on a deep learning model, comprising: An acquisition module, configured to acquire in real time first airway pressure waveform data generated after a current patient uses a medical ventilator; An input module, configured to input the first airway pressure waveform data into a trained deep learning model; A generation module, configured to extract key parameters corresponding to the first airway pressure waveform data and generate second airway pressure waveform data at a future set time point or within a set time period according to the key parameters, wherein the key parameters at least include a change rate, a peak value, and a valley value.
9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and the computer program is used to execute the method described in any one of the above claims 1-7.
10. An electronic device, characterized in that, The electronic device includes: A processor; A memory for storing executable instructions of the processor; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of the above claims 1-7.