Anesthesia machine ventilation trigger identification method, system, and anesthesia machine
By analyzing the shape of the ventilation curve using a convolutional neural network, the problem of false triggering or non-triggering caused by sensor errors and circuit leakage in traditional anesthesia machines is solved, achieving more precise and safer ventilation control.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEYER MEDICAL CO LTD
- Filing Date
- 2024-12-19
- Publication Date
- 2026-06-23
AI Technical Summary
Traditional anesthesia machines are easily affected by sensor errors or circuit leaks when determining ventilation triggering, leading to false triggering or failure to trigger, and thus failing to provide accurate and reliable ventilation support in complex environments.
Convolutional neural networks are used to analyze the shape of ventilation curves. By collecting airway pressure and flow waveform data, converting them into matrix data, and inputting them into the convolutional neural network, the output probability value is compared with a threshold to determine the ventilation triggering condition, thus avoiding reliance on sensor absolute values.
It achieves more accurate ventilation trigger judgment, solves the problem of false triggering or non-triggering caused by sensor error or circuit leakage in traditional methods, and ensures more accurate and safer anesthesia ventilation control.
Smart Images
Figure CN119680071B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical device technology, and in particular to a ventilation trigger recognition method, system, and anesthesia machine for anesthesia. Background Technology
[0002] Triggered ventilation recognition is a crucial function in the ventilation control system of anesthesia machines. This function is particularly important in trigger-based ventilation modes such as pressure support (PS) and synchronized intermittent forced ventilation (SIMV), especially when the user has spontaneous breathing ability. When a user initiates an inspiratory effort, airway pressure decreases, accompanied by a change in inspiratory flow rate. The anesthesia machine can determine whether triggering conditions are met by monitoring changes in airway pressure or inspiratory flow rate. Once the airway pressure drops significantly or the inspiratory flow rate reaches a certain threshold, the device will determine that the user has initiated an inspiratory action, thereby activating ventilation support for the user.
[0003] Traditional anesthesia machines typically follow these steps to determine whether a trigger has occurred: (1) Determination of the end of exhalation: During exhalation, the anesthesia machine first needs to determine whether exhalation has ended. Generally, the device considers the end of exhalation to be marked by stable airway pressure and expiratory flow. Only when these two parameters stabilize will the system activate the trigger determination function. Specifically, end-expiratory pressure is considered the sign of the end of exhalation; when pressure and flow reach a stable state, the exhalation action is considered essentially complete. (2) Real-time determination of trigger conditions: After exhalation, the anesthesia machine will determine whether the trigger conditions are met by real-time monitoring of airway pressure or flow. The specific determination criteria are as follows:
[0004] Pressure trigger: The device records the end-expiratory pressure value. When the airway pressure is lower than the set end-expiratory pressure value, it is considered that the user has initiated an inhalation effort, and the trigger condition is met.
[0005] Flow trigger: When the inhalation flow exceeds the preset threshold, the system determines that the user has inhaled and the trigger condition is met.
[0006] However, in actual use, anesthesia machines often experience false triggering or failure to trigger when determining whether the triggering conditions are met. For example, when a leak occurs in the ventilation circuit, the airway pressure may continue to drop after expiration. Since traditional systems rely on the stability of airway pressure to determine whether expiration has ended, a leak may cause the system to misjudge that expiration has not ended, thus preventing the triggering function from being activated. Another example is false triggering or failure to trigger due to flow sensor malfunction or misalignment. In some cases, the flow sensor may malfunction or become misaligned, causing it to incorrectly measure flow when no actual flow is present. This can happen during the true end of expiration, when the flow sensor may incorrectly show that there is still flow in the airway, causing the system to believe that the user is still exhaling and preventing the triggering function from being activated. Conversely, if the flow sensor falsely detects inspiratory flow, even if the user has not initiated inspiration, the system will incorrectly interpret it as an inspiration, thus triggering a false trigger.
[0007] These issues demonstrate that traditional triggering methods may fail to provide accurate and reliable ventilation support in complex or unusual usage environments. Therefore, there is an urgent need to improve existing technologies to ensure accurate assessment of the user's inspiratory effort under various conditions, thereby achieving safer and more effective anesthetic ventilation control. Summary of the Invention
[0008] The purpose of this invention is to overcome the above-mentioned defects of the prior art, thereby providing a ventilation trigger recognition method, system and anesthesia machine for anesthesia machines.
[0009] To solve the above-mentioned technical problems, the ventilation trigger recognition method for an anesthesia machine provided by the present invention includes:
[0010] Step 1: Collect airway pressure waveform data and flow waveform data from the anesthesia machine;
[0011] Step 2: Convert the collected airway pressure waveform data and flow waveform data into matrix data;
[0012] Step 3: Input the matrix data into the trained convolutional neural network, and the convolutional neural network outputs probability values based on the input matrix data;
[0013] Step 4: Compare the probability value with the threshold and determine whether the ventilation triggering condition is met; if the triggering condition is met, the anesthesia machine starts ventilation support.
[0014] As an improvement to the above method, in step 1, airway pressure waveform data and flow waveform data are collected every 100ms.
[0015] As an improvement to the above method, the matrix data in step 2 is a 20×20 matrix.
[0016] As an improvement to the above method, the trained convolutional neural network in step 3 includes two convolutional layers, two sampling layers, and one fully connected layer.
[0017] As an improvement to the above method, the kernel size of the convolutional layer is 5×5.
[0018] As an improvement to the above method, the convolutional neural network module is trained using preprocessed ventilation trigger case waveform data, which includes: capacitive pressure trigger waveform data, capacitive flow trigger waveform data, pressure-controlled pressure trigger waveform data, and pressure-controlled flow trigger waveform data.
[0019] As an improvement to the above method, the preprocessing step of the trigger case waveform data includes: determining and retaining the waveform data before ventilation triggering in the trigger case waveform data, and setting the waveform data after ventilation triggering to zero.
[0020] To achieve another objective of the present invention, the present invention also provides a ventilation trigger recognition system for an anesthesia machine, used to execute the above-described ventilation trigger recognition method for anesthesia machines, comprising:
[0021] The data acquisition module is used to acquire airway pressure waveform data and flow waveform data of the anesthesia machine in real time;
[0022] The data conversion module is used to convert the collected airway pressure waveform data and flow waveform data into matrix data;
[0023] Convolutional neural networks output probability values based on input matrix data;
[0024] The judgment module is used to compare the probability value with the threshold and determine whether the ventilation triggering condition is met; if the triggering condition is met, the anesthesia machine starts ventilation support.
[0025] As an improvement to the above system, the system further includes: a training module, wherein the training module is used to preprocess the ventilation trigger case waveform data and use the preprocessed ventilation trigger case waveform data for training; the trigger case waveform data includes: capacitive pressure trigger waveform data, capacitive flow trigger waveform data, pressure-controlled pressure trigger waveform data, and pressure-controlled flow trigger waveform data; the preprocessing steps of the trigger case waveform data include: determining and retaining the waveform data before ventilation triggering in the trigger case waveform data, and setting the waveform data after ventilation triggering to zero.
[0026] To achieve another objective of the present invention, the present invention also provides an anesthesia machine, including the ventilation trigger recognition system of the anesthesia machine described above.
[0027] Compared with the prior art, the advantages of the present invention are that the ventilation trigger recognition method, system and anesthesia machine of the present invention, by using convolutional neural network to analyze the shape of the ventilation curve rather than simply relying on the absolute value of the sensor, can more accurately determine the triggering conditions, effectively solving the problem of false triggering or non-triggering caused by sensor error or circuit leakage in traditional methods, thereby achieving more precise and safer anesthesia ventilation control. Attached Figure Description
[0028] Figure 1 This is a schematic diagram of the ventilation trigger recognition method for an anesthesia machine provided in Embodiment 1 of the present invention;
[0029] Figure 2 This is a schematic diagram of a convolutional neural network;
[0030] Figure 3(a) is a schematic diagram of the capacitive pressure trigger waveform data;
[0031] Figure 3(b) is a schematic diagram of the waveform data triggered by the capacitive flow control;
[0032] Figure 3(c) is a schematic diagram of the pressure trigger waveform data of the pressure control system;
[0033] Figure 3(d) is a schematic diagram of the pressure-controlled flow trigger waveform data;
[0034] Figure 4 This is a schematic diagram of the ventilation trigger judgment process. Detailed Implementation
[0035] The technical solutions provided by the present invention will be further illustrated below with reference to the embodiments.
[0036] To overcome the problem of false triggering or non-triggering that may be caused by relying on the absolute value of sensors in traditional methods, this invention uses a convolutional neural network (CNN) to determine whether the triggering conditions are met by recognizing the shape of the ventilation curve, thereby effectively solving problems such as loop leakage, pressure sensor offset, and flow sensor offset.
[0037] Example 1
[0038] This embodiment provides a ventilation trigger recognition method for anesthesia machines. For example... Figure 1 As shown, it includes the following steps:
[0039] 1. Construction and training of convolutional neural networks
[0040] like Figure 2 As shown, the convolutional neural network (CNN) in this method consists of the following layers:
[0041] Two convolutional layers: used for feature extraction.
[0042] Two sampling layers: used for data compression.
[0043] A fully connected layer: used for the final classification decision.
[0044] The input data size is set to 20×20, and the convolution kernel size is 5×5. This configuration can effectively extract and compress data features while ensuring a moderate data sample size and reasonable training speed.
[0045] 2. Training of Convolutional Neural Networks
[0046] The convolutional neural network is trained using clinical data. The training process mainly includes the following four types of trigger cases: pressure-controlled trigger waveform data as shown in Figure 3(a), flow-controlled trigger waveform data as shown in Figure 3(b), pressure-controlled trigger waveform data as shown in Figure 3(c), and flow-controlled trigger waveform data as shown in Figure 3(d).
[0047] During training, the convolutional neural network needs to learn the features before triggering. To avoid interference from the waveform after triggering in feature extraction, the ventilation waveform after triggering is manually removed during training. Specifically, the region of inspiratory effort (i.e., the waveform portion before triggering) is manually determined, and the feature values of the waveform portion after triggering are uniformly set to zero, thereby preventing subsequent waveforms from interfering with the device's learning process.
[0048] 3. Triggering judgment steps
[0049] Figure 4 The trigger determination process is illustrated. First, real-time pressure and flow waveform data are acquired. Each data sampling point lasts 100ms, meaning 10 points are collected per second, for a total of 20 points. The data size is 20×20, sufficient to describe the inhalation characteristics before triggering. This method allows for effective data acquisition with a moderate memory footprint, not affecting normal system operation. The number of sampling points can be increased when system memory is sufficient. Next, the pressure and flow waveforms are filled with data. The real-time acquired pressure and flow data are filled into the 20×20 matrix, with the remaining blank areas filled with zeros, completing the data extraction by the convolutional neural network. This processing method ensures that the system can effectively extract features from the curve shape and make correct trigger determinations.
[0050] Example 2
[0051] This embodiment provides a ventilation trigger recognition system for an anesthesia machine, used to execute the ventilation trigger recognition method for anesthesia machines in Embodiment 1, including:
[0052] The data acquisition module is used to acquire airway pressure waveform data and flow waveform data of the anesthesia machine in real time;
[0053] The data conversion module is used to convert the collected airway pressure waveform data and flow waveform data into matrix data;
[0054] Convolutional neural networks output probability values based on input matrix data;
[0055] The judgment module is used to compare the probability value with the threshold and determine whether the ventilation triggering condition is met; if the triggering condition is met, the anesthesia machine starts ventilation support.
[0056] Preferably, the system further includes a training module, wherein the training module is used to preprocess the ventilation trigger case waveform data and use the preprocessed ventilation trigger case waveform data for training; the trigger case waveform data includes: capacity-controlled pressure trigger waveform data, capacity-controlled flow trigger waveform data, pressure-controlled pressure trigger waveform data, and pressure-controlled flow trigger waveform data; the preprocessing steps of the trigger case waveform data include: determining and retaining the waveform data before ventilation triggering in the trigger case waveform data, and setting the waveform data after ventilation triggering to zero.
[0057] Example 3
[0058] This embodiment provides an anesthesia machine, including the ventilation trigger recognition system of the anesthesia machine in Embodiment 2.
[0059] As can be seen from the above detailed description of the present invention, the present invention uses a convolutional neural network to analyze the shape of the ventilation curve rather than simply relying on the absolute value of the sensor, which can more accurately determine the triggering conditions and effectively solve the problems of false triggering or non-triggering caused by sensor errors or circuit leakage in traditional methods, thereby achieving more precise and safer anesthesia ventilation control.
[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A ventilation trigger recognition system for an anesthesia machine, comprising: The data acquisition module is used to acquire airway pressure waveform data and flow waveform data of the anesthesia machine in real time; The data conversion module is used to convert the collected airway pressure waveform data and flow waveform data into matrix data; Convolutional neural networks output probability values based on input matrix data; The judgment module is used to compare the probability value with the threshold and determine whether the ventilation triggering condition is met. If the triggering conditions are met, the anesthesia machine will initiate ventilation support; and The training module is used to preprocess the ventilation trigger case waveform data and use the preprocessed ventilation trigger case waveform data for training. The working process of the ventilation trigger recognition system of the anesthesia machine includes: Step 1: Collect airway pressure waveform data and flow waveform data from the anesthesia machine; Step 2: Convert the collected airway pressure waveform data and flow waveform data into matrix data; Step 3: Input the matrix data into the trained convolutional neural network, and the convolutional neural network outputs probability values based on the input matrix data; Step 4: Compare the probability value with the threshold and determine whether the ventilation triggering condition is met; if the triggering condition is met, the anesthesia machine starts ventilation support. The convolutional neural network is trained using preprocessed ventilation trigger case waveform data, which includes: capacity-controlled pressure trigger waveform data, capacity-controlled flow trigger waveform data, pressure-controlled pressure trigger waveform data, and pressure-controlled flow trigger waveform data. The preprocessing steps for the waveform data of the trigger case include: determining and retaining the waveform data before ventilation triggering in the waveform data of the trigger case, and setting the waveform data after ventilation triggering to zero.
2. The ventilation trigger recognition system for an anesthesia machine according to claim 1, characterized in that, In step 1, airway pressure waveform data and flow waveform data are collected every 100ms.
3. The ventilation trigger recognition system for an anesthesia machine according to claim 1, characterized in that, The matrix data in step 2 is a 20×20 matrix.
4. The ventilation trigger recognition system for an anesthesia machine according to claim 1, characterized in that, The trained convolutional neural network in step 3 includes two convolutional layers, two sampling layers, and one fully connected layer.
5. The ventilation trigger recognition system for an anesthesia machine according to claim 4, characterized in that, The kernel size of the convolutional layer is 5×5.
6. An anesthesia machine, characterized in that, Includes the ventilation trigger recognition system for the anesthesia machine as described in claim 1.