A breathing machine

Abstract

This invention provides a ventilator comprising a high-frequency ventilation module and a conventional-frequency ventilation module driven independently by dual air sources, a coaxial airflow coupling unit, a multimodal monitoring unit, an intelligent main control unit, and a safety warning unit. It constructs a dual-dimensional, multimodal monitoring system for microscopic ventilation and macroscopic oxygenation of the lungs through EIT electrical impedance tomography and blood oxygen monitoring. Intelligent closed-loop adjustment of ventilation parameters is achieved based on a PID+CNN-LSTM hybrid algorithm. The coaxial annular nozzle structure enables interference-free superposition of high-frequency and conventional airflows. A three-level safety warning mechanism and multi-scenario adaptation interfaces are also included. This invention significantly improves ventilation accuracy and safety, reduces the risk of ventilator-associated lung injury, and is suitable for respiratory support in various scenarios such as ICU, emergency care, and airway surgery.

Description

Technical Field

[0001] This invention relates to the field of ventilators, more specifically to a ventilator that combines high-frequency ventilation with conventional ventilation, and even further to a ventilator capable of multimodal monitoring and intelligent closed-loop regulation. Background Technology

[0002] Mechanical ventilation is a core support method for patients with severe respiratory failure. Current ventilation techniques are mainly divided into two categories: conventional frequency ventilation (CMV) and high-frequency jet ventilation (HFJV).

[0003] Conventional ventilation uses a relatively large tidal volume and a low frequency to achieve alveolar ventilation. Commonly used clinical parameters are tidal volume (VT): 4–8 mL / kg ideal body weight (PBW) and respiratory rate: 10–25 breaths / minute. Although this can ensure minute ventilation, for patients with significantly reduced lung compliance (such as ARDS), if the tidal volume is set too high, it can still easily cause alveolar overdistension, increasing the risk of barotrauma and ventilator-associated lung injury (VILI).

[0004] High-frequency jet ventilation (HFJV) achieves intrapulmonary gas exchange through extremely small tidal volumes and high-frequency pulsed gas ejection. Typical parameters are tidal volume (VT): 1–3 mL / kg ideal body weight (PBW), and operating frequency: 120–300 breaths / min (2–5 Hz) for adults and up to 600 breaths / min for children / newborns. Because its tidal volume is close to or lower than the anatomical dead space, it can significantly reduce the risk of alveolar overdistension and reduce lung injury. However, when used alone, it may result in insufficient carbon dioxide clearance or limited oxygenation efficiency, making it difficult to meet the ventilation and oxygenation needs of patients with severe respiratory failure on its own.

[0005] In existing technologies, some devices attempt to superimpose high-frequency and normal-frequency ventilation, but this approach suffers from the following key drawbacks:

[0006] (1) There is a lack of real-time monitoring of local lung ventilation, making it impossible to identify local areas of alveolar collapse / overexpansion. Parameter adjustment relies on experience and has low accuracy.

[0007] (2) The monitoring dimension is singular, relying only on macroscopic indicators such as blood oxygen saturation, without linking to microscopic changes in lung ventilation, and the regulation lag is strong;

[0008] (3) The high-frequency and normal-frequency parameters are fixed combination patterns and cannot be dynamically adapted according to individual patient differences (such as weight, degree of lung injury);

[0009] (4) The air path design is unreasonable, and the high-frequency and normal-frequency airflows interfere with each other, resulting in a decrease in ventilation efficiency. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of existing overlay ventilation devices mentioned above, and thereby provide an open-type overlay ventilation intelligent ventilator. Specifically, it includes a high-frequency ventilation module, a conventional-frequency ventilation module, an airway coupling unit, a multimodal monitoring unit, an intelligent main control unit, and a human-machine interaction and safety warning unit. The multimodal monitoring unit includes an EIT (extra-internal pressure) lung impedance monitoring subunit, a blood oxygen saturation monitoring subunit, an airway parameter monitoring subunit, and an exhaled gas analysis subunit. The airway parameter monitoring subunit incorporates a high-precision pressure sensing subunit, a rapid oxygen sensing subunit, and a carbon dioxide sensing subunit for real-time acquisition of airway peak pressure, plateau pressure, positive end-expiratory pressure, oxygen concentration, and end-expiratory carbon dioxide concentration. The exhaled gas analysis subunit acquires samples through a gas acquisition interface and analyzes changes in the concentrations of carbon dioxide and oxygen in the exhaled gas.

[0011] Furthermore, the gas path coupling unit includes a high-frequency ventilation input interface, a normal-frequency ventilation input interface, a gas flow coupling cavity, a gas acquisition interface, and an open gas path interface, wherein the high-frequency ventilation input interface is connected to the high-frequency ventilation module, and the normal-frequency ventilation input interface is connected to the normal-frequency ventilation module.

[0012] Furthermore, the high-frequency ventilation input interface and the normal-frequency ventilation input interface are simultaneously connected to the airflow coupling cavity.

[0013] Furthermore, the airflow coupling unit has a flow sensor to monitor the flow parameters of the jet airflow in real time; the gas acquisition interface is connected to the airway parameter monitoring subunit and the exhaled gas analysis subunit to collect gas samples from the patient.

[0014] Furthermore, the open airway interface is used to adapt to at least one of the following: jet laryngoscope, 3 / 4-cavity jet tubing, and standard endotracheal tube adapter.

[0015] Furthermore, the intelligent main control unit incorporates a data fusion module, a ventilation status assessment module, and an intelligent adjustment module, enabling synchronous processing of multi-source data, hierarchical assessment of respiratory status, and closed-loop adjustment of ventilation parameters.

[0016] Furthermore, the present invention provides a method for controlling a ventilator, comprising the following steps:

[0017] S1. Data Acquisition: After the system starts, each sub-unit of the multimodal monitoring unit starts data acquisition synchronously. The data fusion module of the multimodal monitoring unit synchronizes the data acquired by each sub-unit in time, removes signal noise, extracts effective feature data, preprocesses the multimodal data, and then transmits it to the ventilation status assessment module of the intelligent main control unit.

[0018] S2. The ventilation status assessment module built into the intelligent main control unit receives the preprocessed multimodal data mentioned above and, based on the preset graded assessment model, completes the real-time assessment of the patient's oxygenation status, ventilation status, and lung injury risk status;

[0019] S3. The intelligent adjustment module built into the intelligent main control unit receives the ventilation status assessment result and generates the optimal ventilation parameter adjustment command based on the PID+CNN-LSTM hybrid algorithm model;

[0020] The PID algorithm is used for real-time closed-loop fine-tuning of parameters to eliminate steady-state error of parameters. The parameters output by the model are used as set values, and the airway parameters monitored in real time are used as feedback values ​​to perform millisecond-level parameter fine-tuning. The results are transmitted to the high-frequency ventilation and normal-frequency ventilation modules through high-speed solenoid valves.

[0021] The CNN-LSTM hybrid model is used for long-term prediction of ventilation status and global optimization of parameters. Based on the ventilation status assessment results, the operating parameters of the high-frequency ventilation module and the normal-frequency ventilation module are dynamically adjusted in a closed loop. The model takes 10 consecutive frames of EIT images and 5 consecutive seconds of multimodal physiological parameters as input and outputs the globally optimal combination of ventilation parameters.

[0022] S4. While performing parameter adjustment based on the PID+CNN-LSTM hybrid algorithm model, continuously collect the adjusted monitoring data to form a closed-loop control of "monitoring-evaluation-adjustment-feedback". At the same time, input the adjustment data into the PID+CNN-LSTM hybrid algorithm model to continuously optimize the algorithm's adaptability and achieve individualized ventilation support for different patients.

[0023] Furthermore, the real-time evaluation in step S2 includes:

[0024] Oxygenation status is assessed based on real-time SpO2 values, which are then classified as normal, mildly deficient, or severely deficient. Specifically, an SpO2 value of 95% or higher is considered normal, between 90% and 94% indicates mild oxygenation deficiency, and less than 90% indicates severe oxygenation deficiency.

[0025] Ventilation status assessment is based on the ventilation uniformity index calculated from the EIT images, which determines whether the ventilation is uniform, slightly uneven, or severely uneven. Specifically, an EIT ventilation uniformity index greater than or equal to 0.6 indicates uniform ventilation, between 0.4 and 0.59 indicates slightly uneven ventilation, and less than 0.4 indicates severely uneven ventilation.

[0026] Risk assessment identifies the risk level of alveolar overdistension and collapse based on plateau pressure values ​​and local impedance changes in EIT images; among which, plateau pressure P... platA lung overexpansion risk is defined as a lung overexpansion when the lung's ventilatory resistance is ≥28 cmH2O and the endoscopic interventional test (EIT) shows local hyperventilation; a lung collapse risk is defined as a lung overexpansion when the EIT shows a relative decrease in local ventilation resistance greater than 20%.

[0027] Furthermore, in step S3, when the SpO2 value is detected to be less than 90% and the ventilation uniformity is less than 0.4, it is assessed as severe hypoxia and severe ventilation heterogeneity. At this time, the high-frequency jet pressure is increased by 3%-8% and the frequency is increased by 5%-10%, while the normal frequency ventilation pressure is increased by 3%-6% and the inspiratory-to-expiratory ratio is adjusted to 1:1.5 to 1:2, thereby simultaneously increasing the oxygen concentration by 3%-5%.

[0028] When SpO2 is monitored to be greater than or equal to 95% and ventilation uniformity is greater than or equal to 0.6, oxygenation is assessed as normal and ventilation is uniform. At this time, the high-frequency jet ventilation frequency is reduced by 5%-10%, while the normofrequency ventilation pressure is reduced by 3%-6%, and the inspiratory-to-expiratory ratio is adjusted to 1:2 to 1:2.5. This reduces ventilation intensity while ensuring oxygenation, thereby minimizing the risk of lung injury.

[0029] When P is detected plat When alveolar hyperdistension is detected at 28 cmH2O or higher and an early warning of alveolar hyperdistension is given, it is assessed as a high risk of alveolar hyperdistension. At this time, the high-frequency jet pressure is reduced by 5%-8%, the conventional ventilation pressure is reduced by 5%-8%, and the positive end-expiratory pressure (PEEP) is reduced by 1 cmH2O to avoid further alveolar damage.

[0030] When a warning of local alveolar collapse is detected and the SpO2 value is normal, it is assessed as a risk of local alveolar collapse. At this time, the frequency of high-frequency pulmonary embolism is increased by 8%-12%, the duration of pulmonary embolism is extended by 5%-10%, the frequency of conventional ventilation is increased by 5%-10%, the inspiratory-to-expiratory ratio is adjusted to 1:2, and the positive end-expiratory pressure (PEEP) is increased by 1 cmH2O to promote the re-expansion of collapsed alveoli.

[0031] Furthermore, the PID algorithm is used for real-time closed-loop fine-tuning of parameters to eliminate steady-state errors of parameters. The parameters output by the model are used as setpoints, and the airway parameters monitored in real time are used as feedback values ​​to perform millisecond-level parameter fine-tuning. The results are transmitted to the high-frequency ventilation and normal-frequency ventilation modules through high-speed solenoid valves.

[0032] The CNN-LSTM hybrid model is used for long-term prediction of ventilation status and global optimization of parameters. Based on the ventilation status assessment results, the operating parameters of the high-frequency ventilation module and the normal-frequency ventilation module are dynamically adjusted in a closed loop. Taking 10 consecutive frames of EIT images and 5 consecutive seconds of multimodal physiological parameters as input, the model outputs the globally optimal combination of ventilation parameters.

[0033] Based on the above-mentioned invention, the ventilator and its control method of the present invention can realize the integrated design of "multimodal monitoring-intelligent closed-loop adjustment" in open ventilation scenarios, solving the core problem of "blindly stacking" traditional equipment; it captures microscopic ventilation changes in the lungs in real time through EIT technology, and dynamically optimizes ventilation parameters by combining machine learning algorithms to adapt to individual differences in lung function of different patients; it significantly reduces the risk of ventilator-associated lung injury through multi-dimensional risk warning and automatic emergency adjustment; and it is compatible with multiple clinical interfaces through low dead space and low resistance airway optimization design, adapting to the respiratory support needs of multiple scenarios such as ICU, emergency care, and airway surgery. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the 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.

[0035] Figure 1 : A schematic diagram of the overall structure of the ventilator in this application;

[0036] Figure 2 : A schematic diagram of the airway coupling unit of the ventilator in this application;

[0037] Figure label:

[0038] 1-Ventilation input interface; 2-Airflow coupling unit; 3-Gas acquisition interface; 4-Open airway interface; 8-Intelligent main control unit; 9-High-frequency ventilation module; 10-Normal frequency ventilation module; 11-Airway coupling unit; 12-Multimodal monitoring unit; 13-EIT pulmonary impedance monitoring subunit; 14-Blood oxygen saturation monitoring subunit; 15-High-precision pressure sensing subunit; 16-Oxygen sensing subunit; 17-Carbon dioxide sensing subunit. Detailed Implementation

[0039] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. It is obvious that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0040] This invention discloses an open-type superimposed ventilation intelligent ventilator, the overall structure of which is as follows: Figure 1As shown, the system includes a high-frequency ventilation module 9, a conventional-frequency ventilation module 10, an airway coupling unit 11, a multimodal monitoring unit 12, an intelligent main control unit 8, and a human-machine interaction and safety warning unit. The high-frequency ventilation module 9 uses a high-frequency solenoid valve as the core actuator, achieving an airflow output of 150 L / min under certain pressure, with a frequency adjustment range of 50-2000 breaths / min and a jet pressure adjustment range of 0.1-2 bar. The conventional-frequency ventilation module 10 integrates a servo motor pressure control system, supporting PSV, PCV, and spontaneous breathing synchronous trigger modes. Its ventilation frequency adjustment range is 1-150 breaths / min, ventilation pressure adjustment range is 0.1-3.5 bar, I:E ratio adjustment range is 5:1-1:5, and tidal volume adjustment range is 2-15 ml / kg. Within this adjustment range, the initial value can be adaptively set according to the patient's weight; the upper limit of airway pressure is less than or equal to 120 cmH2O (the specific value can be customized). Both the high-frequency ventilation module 9 and the conventional-frequency ventilation module 10 employ a pressure-volume dual control mode, with a tidal volume output error of less than or equal to ±5%. They feature built-in high-speed solenoid valves (response time less than or equal to 4ms) to ensure the stability and repeatability of the ventilation airflow. Furthermore, they are equipped with multi-stage oxygen sensors for monitoring, with an oxygen concentration adjustment range of 21%–100% and a dynamic response time of less than or equal to 800ms. They also feature built-in automatic calibration of respiratory mechanics parameters and support one-click adaptation for patients across weight ranges from 3kg to 150kg. The high-frequency ventilation module 9 and the conventional-frequency ventilation module 10 adopt a dual-gas-source independent drive design to avoid mutual airflow interference, and both are connected to a gas path coupling unit at their output ends.

[0041] The structure of the gas path coupling unit 11 is as follows Figure 2 As shown, the system includes a ventilation input interface 1, an airflow coupling unit 2, a gas sampling interface 3, and an open air path interface 4. The high-frequency ventilation input interface 1 is connected to the high-frequency ventilation module 9, and the normal-frequency ventilation input interface is connected to the normal-frequency ventilation module 10. Both are simultaneously connected to the airflow coupling cavity 2. The airflow coupling cavity 2 has a cavity length of 50 mm and an inner diameter of 10 mm, with a smooth arc-shaped flow guide structure on its inner wall. Fluid dynamics simulation and in vitro experiments have verified that the air path structure of this embodiment can attenuate the cross-interference between the high-frequency and normal-frequency ventilation airflows by up to 94%, effectively avoiding airflow backflow and parameter disturbances.

[0042] The airflow coupling chamber 2 incorporates a pressure sensor and flow regulator with an accuracy of ±1%FS, enabling real-time monitoring of the pressure and flow parameters of the coupled airflow. The gas acquisition interface 3 connects to the exhaled gas analysis subunit for collecting patient exhaled gas samples. The open airway interface 4 features a low-resistance, low-dead-space design, adaptable to various clinical interfaces such as jet laryngoscopes, 3 / 4-cavity jet catheters, and standard endotracheal intubation adapters, corresponding to scenarios such as laryngeal and tracheal surgery, ICU invasive ventilation, and emergency resuscitation. The open airway interface 4 meets the needs of multiple scenarios, improving the device's adaptability and versatility.

[0043] The multimodal monitoring unit includes an EIT pulmonary impedance monitoring subunit 13, a blood oxygen saturation monitoring subunit 14, an airway parameter monitoring subunit, and an exhaled gas analysis subunit. The EIT pulmonary impedance monitoring subunit 13 consists of a multi-channel flexible silver fiber electrode array, an impedance signal acquisition unit, and a real-time image reconstruction module. It has a sampling frequency of 50Hz, an impedance measurement range of, for example, 5-500Ω, and an image resolution of 256×256 pixels. It generates an intrapulmonary impedance distribution image every 0.5 seconds, accurately identifying local areas of alveolar collapse and overexpansion, and outputting a ventilation uniformity index and risk warning signals.

[0044] The blood oxygen saturation monitoring subunit 14 uses a fingertip blood oxygen probe, paired with a dual-wavelength signal processor with 660nm red light and 940nm infrared light. The SpO2 measurement range is 70-100% with an accuracy of ±2%, and the pulse rate measurement range is 30-250 beats / minute, updating monitoring data every second. The airway parameter monitoring subunit incorporates a high-precision pressure sensing subunit 15, an oxygen sensing subunit 16 with a response time of less than or equal to 1 second, and a pressure-compensated carbon dioxide sensing subunit 17, acquiring parameters such as airway peak pressure, plateau pressure, positive end-expiratory pressure (PEEP), oxygen concentration, and end-expiratory carbon dioxide concentration in real time. The exhaled gas analysis subunit acquires samples through a gas acquisition interface and analyzes changes in the concentrations of carbon dioxide and oxygen in exhaled gas.

[0045] The intelligent main control unit 8 employs a microcontroller based on the ARM Cortex-M7 core with a main frequency of 480MHz. It is equipped with a 16-bit high-speed ADC acquisition module and multi-channel RS232 and Ethernet communication interfaces, which are electrically connected to the high-frequency ventilation module 9, the normal-frequency ventilation module 10, and the multimodal monitoring unit 12, respectively, enabling millisecond-level data acquisition and command response. The intelligent main control unit 8 integrates a data fusion module, a ventilation status assessment module, and an intelligent adjustment module, achieving synchronous processing of multi-source data, graded assessment of respiratory status, and closed-loop adjustment of ventilation parameters.

[0046] The human-machine interface and safety warning unit is equipped with a 15.6-inch industrial-grade touchscreen display, a physical emergency stop button, a 64GB local SD card storage unit, and expansion interfaces such as Ethernet, HDMI, and USB. The display can show real-time EIT lung ventilation distribution heatmaps, monitoring parameter trend curves, current ventilation parameters, and alarm information. Simultaneously, the human-machine interface and safety warning unit supports manual / automatic mode switching, custom parameter settings, and patient information entry; it can continuously store 72 hours of monitoring data and adjustment logs, and supports export in PDF / CSV format. The safety warning unit is equipped with an audible and visual alarm with a volume ≥75dB, a preset three-level alarm mechanism, and simultaneously executes the corresponding safety protection strategy when an alarm is triggered.

[0047] In this embodiment, all operating parameters of the whole machine comply with standards such as GB 9706.1-2020 (IEC 60601-1:2012) and YY9706.287-2025 (ISO 80601-2-87:2021), and the EMC immunity meets YY 9706.102-2021 (IEC 60601-1-2). The software part has passed IEC 62304 Class C certification, meeting the registration requirements of Class III medical devices.

[0048] Next, the complete workflow of the intelligent closed-loop regulation of the intelligent ventilator of the present invention will be described in detail, specifically in four stages:

[0049] 1. Synchronous data acquisition phase

[0050] After system startup, each subunit of the multimodal monitoring unit 12 synchronously starts data acquisition: the EIT electrode array of the EIT pulmonary impedance monitoring subunit 13 is fitted to the patient's chest, generating an intrapulmonary impedance distribution image every 0.5 seconds to identify areas of abnormal ventilation; the blood oxygen saturation monitoring subunit 14's blood oxygen saturation probe provides SpO2 and pulse rate data every 1 second; the airway parameter monitoring subunit acquires airway pressure, oxygen concentration, and other parameters in real time at a sampling frequency of 100Hz; and the exhaled gas analysis subunit analyzes the CO2 concentration of exhaled gas in real time. The data fusion module of the multimodal monitoring unit 12 synchronizes the time of each acquired data stream, with a synchronization error of less than or equal to 10ms. Simultaneously, it removes signal noise using a Kalman filter algorithm, extracts effective feature data, preprocesses the multimodal data, and then transmits it to the ventilation status assessment module.

[0051] 2. Ventilation status classification assessment stage

[0052] The ventilation status assessment module built into the intelligent main control unit 8 receives the preprocessed multimodal data and, based on a preset grading assessment model, completes real-time assessment of the patient's oxygenation status, ventilation status, and lung injury risk status.

[0053] Oxygenation status assessment: Based on real-time SpO2 values, it is determined to be normal, mildly deficient, or severely deficient. Specifically, an SpO2 value greater than or equal to 95% is considered normal, between 90% and 94% (inclusive) is considered mildly deficient, and less than 90% is considered severely deficient.

[0054] Ventilation status assessment: The ventilation uniformity index is calculated based on the EIT images to determine whether the ventilation is uniform, mildly uneven, or severely uneven. Specifically, an EIT ventilation uniformity index greater than or equal to 0.6 indicates uniform ventilation, a value between 0.4 and 0.59 (inclusive) indicates mildly uneven ventilation, and a value less than 0.4 indicates severely uneven ventilation.

[0055] Risk assessment: Based on plateau pressure values ​​and changes in local impedance on EIT images, the risk level of alveolar overdistension and collapse is identified. Among these, plateau pressure (P0.05) is... plat A lung overexpansion risk is indicated by an H2O level greater than or equal to 28 cmH2O. An EIT reading of more than 20% in multiple frames within the same respiratory cycle indicates a high risk of alveolar collapse.

[0056] 3. Intelligent parameter adjustment stage

[0057] The intelligent control unit 8, with its built-in intelligent adjustment module, receives ventilation status assessment results and generates optimal ventilation parameter adjustment commands based on a PID+CNN-LSTM hybrid algorithm model. Specifically, the PID algorithm is used for real-time closed-loop fine-tuning of parameters, eliminating steady-state errors. Using the model output parameters as setpoints and real-time monitored airway parameters as feedback values, millisecond-level parameter fine-tuning is performed, completing the adjustment within one second. This adjustment is transmitted to the high-frequency ventilation module 9 and the normal-frequency ventilation module 10 via high-speed solenoid valves. The CNN-LSTM hybrid model is used for long-term temporal prediction of ventilation status and global parameter optimization. Based on the ventilation status assessment results, it dynamically adjusts the operating parameters of the high-frequency and normal-frequency ventilation modules in a closed loop. Using 10 consecutive frames of EIT images and 5 consecutive seconds of multimodal physiological parameters as input, it outputs the globally optimal combination of ventilation parameters, with a model inference latency of less than or equal to 10ms. The core adjustment logic is as follows:

[0058] 1. When the SpO2 value is less than 90% and the ventilation uniformity is less than 0.4, it is assessed as severe hypoxia and severe ventilation heterogeneity. At this time, the high-frequency jet pressure is increased by 3%-8% and the frequency is increased by 5%-10%, while the normal frequency ventilation pressure is increased by 3%-6% and the inspiratory-to-expiratory ratio is adjusted to 1:1.5-1:2, and the oxygen concentration is increased by 3%-5% simultaneously.

[0059] 2. When the SpO2 value is greater than or equal to 95% and the ventilation uniformity is greater than or equal to 0.6, it is assessed as normal oxygenation and uniform ventilation. At this time, the high-frequency jet ventilation frequency is reduced by 5%-10%, while the normofrequency ventilation pressure is reduced by 3%-6%, and the inspiratory-to-expiratory ratio is adjusted to 1:2-1:2.5. This reduces the ventilation intensity while ensuring oxygenation, thereby minimizing the risk of lung injury.

[0060] 3. When P is detected plat When the alveolar hyperdistension warning is greater than 28 cmH2O and alveolar hyperdistension is detected, it is assessed as a high risk of alveolar hyperdistension. At this time, the high-frequency jet pressure is reduced by 5%-8%, the conventional ventilation pressure is reduced by 5%-8%, and the positive end-expiratory pressure (PEEP) is reduced by 1 cmH2O to avoid further alveolar damage.

[0061] 4. When a local alveolar collapse warning is detected and the SpO2 value is normal, it is assessed as a risk of local alveolar collapse. At this time, the high-frequency jet ventilation frequency is increased by 8%-12%, the jet duration is extended by 5%-10%, the conventional frequency ventilation frequency is increased by 5%-10%, the inspiratory-to-expiratory ratio is adjusted to 1:2, and the positive end-expiratory pressure (PEEP) is increased by 1 cmH2O to promote the re-expansion of collapsed alveoli.

[0062] While performing parameter adjustments, the system continuously collects monitoring data after adjustment, forming a closed-loop control of "monitoring-evaluation-adjustment-feedback". At the same time, the adjustment data is input into the model to continuously optimize the adaptability of the algorithm and achieve individualized ventilation support for different patients.

[0063] The CNN-LSTM hybrid model employs a dual-branch input structure. The first branch is a CNN feature extraction branch, which takes 10 consecutive frames of EIT lung ventilation images (size 256×256×1) as input, passing through 3 convolutional layers, 2 pooling layers, and 1 Flatten layer to extract spatial features of local lung ventilation. The second branch is an LSTM temporal feature extraction branch, which takes a 5-second sequence of multimodal physiological parameters as input (including SpO2 value, airway peak pressure, plateau pressure, positive end-expiratory pressure PEEP, oxygen concentration, and CO2 concentration, totaling 6 features), passing through 2 LSTM layers to extract temporal variation features of respiratory status. The output features of the two branches are fused through a Concat layer and then fed into 2 fully connected layers, ultimately outputting optimized adjustment values ​​for 7 ventilation parameters: high-frequency ventilation frequency, high-frequency ventilation pressure, conventional ventilation pressure, conventional ventilation frequency, inspiratory-expiratory ratio, positive end-expiratory pressure PEEP, and oxygen concentration.

[0064] The training dataset for the CNN-LSTM hybrid model was derived from multicenter clinical data from 12 tertiary hospitals across China. Inclusion criteria: ① Age 18-75 years; ② Meeting the Berlin definition of ARDS, severe pneumonia, postoperative atelectasis, or other indications requiring mechanical ventilation support; ③ Receiving high-frequency and conventional ventilation therapy. Exclusion criteria: Patients with severe pneumothorax, severe chest wall deformity, or hemodynamic instability.

[0065] Total dataset size: It contains 7.2 million EIT sequence images of 2,860 patients, 120 million multimodal physiological parameter records, and corresponding optimal ventilation parameter combination labels calibrated by clinicians; the dataset is divided into training, validation and test sets in a 7:2:1 ratio, and a 5-fold cross-validation method is used to ensure the generalization ability of the model.

[0066] Data preprocessing: Normalize, denoise, and unify the size of EIT images; standardize physiological parameters using Z-score, remove outliers, and interpolate missing values; align all data with timestamps to ensure the temporal synchronization of input data.

[0067] Training methods:

[0068] For example, an NVIDIA RTX A6000 GPU card can be used for model training, based on the PyTorch 1.12 deep learning framework;

[0069] Hyperparameter settings: Batch size is 32, the optimizer is Adam optimizer, the initial learning rate is 1e-4, the learning rate decay coefficient is 0.95, the number of training epochs is 100, and the early stopping policy is to stop training if the validation set loss does not decrease for 10 consecutive epochs.

[0070] Loss function: A composite loss function is used, where total loss = oxygenation target loss + ventilation uniformity loss + lung injury risk loss + parameter smoothing loss, and:

[0071] Oxygenation target loss: Using mean squared error (MSE), the target value of SpO2 is constrained to be greater than or equal to 95%;

[0072] Ventilation uniformity loss: Using MSE, the ventilation uniformity index is constrained to be greater than or equal to 0.8;

[0073] Lung injury risk loss: Using cross-entropy loss, the plateau pressure is constrained to <28cmH2O to eliminate the risk of alveolar over-expansion and collapse;

[0074] Parameter smoothing loss: L1 regularization is used to constrain the adjustment range of ventilation parameters and avoid patient discomfort caused by drastic parameter fluctuations.

[0075] Model fusion: The trained CNN-LSTM model is fused with the PID algorithm. The PID algorithm uses the parameters output by the model as the set value and the real-time monitored airway parameters and oxygenation parameters as feedback values ​​to perform millisecond-level parameter fine-tuning, taking into account both the global optimization capability of the model and the real-time stability of the PID algorithm.

[0076] Model Validation and Technical Results:

[0077] Validation results based on the test set show that the ventilation parameters output by the CNN-LSTM model of this invention have a 92.7% consistency with the optimal parameters calibrated by experienced clinical respiratory therapists. The model can predict the risk of decreased oxygenation and ventilatory asymmetry 15 seconds in advance, with a warning accuracy of 94.2%. Compared with traditional fixed-parameter superimposed ventilation equipment, the model of this invention can reduce the ventilatory asymmetry index by 47.3%, narrow the fluctuation range of SpO2 to ±1.2%, shorten the time to achieve target oxygenation by 62%, reduce the incidence of ventilator-associated lung injury from 8.6% to 0.8%, and shorten the average weaning time by 36.5 hours. These data are preliminary results from a prospective small-sample trial of this project. Subsequent clinical trials will be conducted in accordance with medical device registration regulations to supplement and validate these results.

[0078] 4. Safety Early Warning and Emergency Response Phase

[0079] Throughout the ventilation process, the safety warning unit monitors the alarm triggering conditions in real time and executes corresponding warning and protection strategies based on the risk level:

[0080] When a Level 1 alarm is triggered, a notification message will pop up on the screen, and a low-frequency audio-visual alert will be activated to remind medical staff to pay attention to the patient's condition.

[0081] When a Level 2 warning alarm is triggered, a continuous audible and visual alarm will be activated, the abnormal parameters will be highlighted on the screen, and the system will automatically fine-tune the ventilation parameters to a safe range.

[0082] When a Level 3 emergency alarm is triggered, a high-decibel audible and visual alarm is activated, and the physical button indicator lights flash simultaneously. The system immediately and automatically switches to the preset safe ventilation mode and locks the risk parameter adjustment to avoid misoperation and maximize the protection of the patient's life.

[0083] The beneficial effects of this invention are:

[0084] (1) This invention realizes for the first time the integrated design of "multimodal monitoring-intelligent closed-loop adjustment" in open superimposed ventilation scenario. It integrates EIT local ventilation monitoring and blood oxygen macro-oxygenation monitoring to construct a dual-dimensional monitoring system of "micro ventilation-macro oxygenation". It fundamentally solves the core problem of "blind superposition" of traditional superimposed ventilation equipment and significantly improves the accuracy of ventilation parameter adjustment.

[0085] (2) This invention captures the dynamic changes of lung micro-ventilation in real time through EIT technology. Combined with the self-developed and trained PID+CNN-LSTM hybrid algorithm model, it can dynamically optimize the core parameters of high-frequency and conventional ventilation according to the individual differences of patients and their real-time respiratory status, so as to achieve true individualized ventilation support. The ventilation adaptability is greatly improved, which can cover patients of all weight ranges from 3kg to 150kg and is suitable for 8 clinical indications such as ARDS, postoperative atelectasis, and neonatal respiratory distress.

[0086] (3) The present invention sets up a multi-dimensional lung injury risk warning and a three-level safety alarm mechanism, which can predict ventilation risk 15 seconds in advance and realize automatic emergency adjustment simultaneously, significantly reducing the risk of ventilator-associated lung injury and improving ventilation safety.

[0087] (4) This invention optimizes the airway coupling design of the coaxial annular nozzle and utilizes the Venturi effect to achieve interference-free superposition of high-frequency and normal-frequency airflow, with cross-interference attenuation of more than 92%. At the same time, it adopts an open airway design with low dead space and low resistance, and is compatible with various clinical interfaces such as jet laryngoscope, 3 / 4-cavity jet catheter, and standard endotracheal intubation adapter. It can adapt to the respiratory support needs of multiple scenarios such as ICU intensive care, emergency resuscitation, and laryngotracheal surgery, and has extremely strong clinical applicability.

[0088] The ventilator provided by this invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. For those skilled in the art, the technical solutions of this invention are not limited to the solutions defined in the specific embodiments. Technical solutions formed by other modifications that can be obviously implemented based on ordinary technical knowledge in the art are all within the protection scope of this invention.

Claims

1. A ventilator, comprising a high-frequency ventilation module, a conventional-frequency ventilation module, an airway coupling unit, a multimodal monitoring unit, an intelligent main control unit, and a human-machine interaction and safety warning unit; characterized in that, The multimodal monitoring unit includes an EIT (exhalation-internal pressure) lung impedance monitoring subunit, a blood oxygen saturation monitoring subunit, an airway parameter monitoring subunit, and an exhaled gas analysis subunit. The airway parameter monitoring subunit incorporates a high-precision pressure sensor subunit, a rapid oxygen sensor subunit, and a carbon dioxide sensor subunit to collect peak airway pressure, plateau pressure, positive end-expiratory pressure, oxygen concentration, and end-expiratory carbon dioxide concentration in real time. The exhaled gas analysis subunit acquires samples through a gas acquisition interface and analyzes changes in the concentrations of carbon dioxide and oxygen in exhaled gas.

2. The ventilator as described in claim 1, characterized in that: The gas path coupling unit includes a high-frequency ventilation input interface, a normal-frequency ventilation input interface, a gas flow coupling cavity, a gas acquisition interface, and an open gas path interface. The high-frequency ventilation input interface is connected to the high-frequency ventilation module, and the normal-frequency ventilation input interface is connected to the normal-frequency ventilation module.

3. The ventilator as described in claim 2, characterized in that: The high-frequency ventilation input interface and the normal-frequency ventilation input interface are simultaneously connected to the airflow coupling cavity.

4. The ventilator as described in claim 3, characterized in that: The airflow coupling unit has a flow sensor to monitor the flow parameters of the jet airflow in real time; the gas acquisition interface is connected to the airway parameter monitoring subunit and the exhaled gas analysis subunit to collect gas samples from the patient.

5. The ventilator as described in claim 4, characterized in that: The open airway interface is used to adapt to at least one of the following: jet laryngoscope, 3 / 4-cavity jet tubing, and standard endotracheal tube adapter.

6. The ventilator according to any one of claims 1-5, characterized in that: The intelligent main control unit has built-in data fusion module, ventilation status assessment module, and intelligent adjustment module, which realizes synchronous processing of multi-source data, hierarchical assessment of respiratory status, and closed-loop adjustment of ventilation parameters.

7. A method for controlling a ventilator as described in any one of claims 1-6, comprising the following steps: S1. Data Acquisition: After the system starts, each sub-unit of the multimodal monitoring unit starts data acquisition synchronously. The data fusion module of the multimodal monitoring unit synchronizes the data acquired by each sub-unit in time, removes signal noise, extracts effective feature data, preprocesses the multimodal data, and then transmits it to the ventilation status assessment module of the intelligent main control unit. S2. The ventilation status assessment module built into the intelligent main control unit receives the preprocessed multimodal data mentioned above and, based on the preset graded assessment model, completes the real-time assessment of the patient's oxygenation status, ventilation status, and lung injury risk status; S3. The intelligent adjustment module built into the intelligent main control unit receives the ventilation status assessment result and generates the optimal ventilation parameter adjustment command based on the PID+CNN-LSTM hybrid algorithm model; in, The PID algorithm is used for real-time closed-loop fine-tuning of parameters to eliminate steady-state error of parameters. The parameters output by the model are used as setpoints and the airway parameters monitored in real time are used as feedback values ​​to perform millisecond-level parameter fine-tuning. The results are transmitted to the high-frequency ventilation and normal-frequency ventilation modules through high-speed solenoid valves. The CNN-LSTM hybrid model is used for long-term prediction of ventilation status and global optimization of parameters. Based on the ventilation status assessment results, the operating parameters of the high-frequency ventilation module and the normal-frequency ventilation module are dynamically adjusted in a closed loop. The model takes 10 consecutive frames of EIT images and 5 consecutive seconds of multimodal physiological parameters as input and outputs the globally optimal combination of ventilation parameters. S4. While performing parameter adjustment based on the PID+CNN-LSTM hybrid algorithm model, continuously collect the adjusted monitoring data to form a closed-loop control of "monitoring-evaluation-adjustment-feedback". At the same time, input the adjustment data into the PID+CNN-LSTM hybrid algorithm model to continuously optimize the algorithm's adaptability and achieve individualized ventilation support for different patients.

8. The control method for a ventilator as described in claim 7, characterized in that, The real-time evaluation in step S2 includes: Oxygenation status is assessed based on real-time SpO2 values, which are then classified as normal, mildly deficient, or severely deficient. Specifically, an SpO2 value of 95% or higher is considered normal, between 90% and 94% indicates mild oxygenation deficiency, and less than 90% indicates severe oxygenation deficiency. Ventilation status assessment is based on the ventilation uniformity index calculated from the EIT images, which determines whether the ventilation is uniform, slightly uneven, or severely uneven. Specifically, an EIT ventilation uniformity index greater than or equal to 0.6 indicates uniform ventilation, between 0.4 and 0.59 indicates slightly uneven ventilation, and less than 0.4 indicates severely uneven ventilation. Risk assessment identifies the risk level of alveolar overdistension and collapse based on plateau pressure values ​​and local impedance changes in EIT images; among which, plateau pressure P... plat A lung overexpansion is considered high-risk when the lung temperature is ≥28 cmH2O and the EIT shows local hyperventilation; a lung collapse is considered high-risk when the EIT shows a relative decrease of more than 20% in the average local ventilation impedance across multiple frames within the same respiratory cycle.

9. The control method for a ventilator as described in claim 8, characterized in that, In step S3, when the SpO2 value is less than 90% and the ventilation uniformity is less than 0.4, it is assessed as severe hypoxia and severe ventilation heterogeneity. At this time, the high-frequency jet pressure is increased by 3%-8% and the frequency is increased by 5%-10%, while the normal frequency ventilation pressure is increased by 3%-6% and the inspiratory-to-expiratory ratio is adjusted to 1:1.5 to 1:2, and the oxygen concentration is increased by 3%-5% simultaneously. When SpO2 is monitored to be greater than or equal to 95% and ventilation uniformity is greater than or equal to 0.6, oxygenation is assessed as normal and ventilation is uniform. At this time, the high-frequency jet ventilation frequency is reduced by 5%-10%, while the normofrequency ventilation pressure is reduced by 3%-6%, and the inspiratory-to-expiratory ratio is adjusted to 1:2 to 1:2.

5. This reduces ventilation intensity while ensuring oxygenation, thereby minimizing the risk of lung injury. When P is detected plat When alveolar hyperdistension is detected at 28 cmH2O or higher and an early warning of alveolar hyperdistension is given, it is assessed as a high risk of alveolar hyperdistension. At this time, the high-frequency jet pressure is reduced by 5%-8%, the conventional ventilation pressure is reduced by 5%-8%, and the positive end-expiratory pressure (PEEP) is reduced by 1 cmH2O to avoid further alveolar damage. When a warning of local alveolar collapse is detected and the SpO2 value is normal, it is assessed as a risk of local alveolar collapse. At this time, the frequency of high-frequency pulmonary embolism is increased by 8%-12%, the duration of pulmonary embolism is extended by 5%-10%, the frequency of conventional ventilation is increased by 5%-10%, the inspiratory-to-expiratory ratio is adjusted to 1:2, and the positive end-expiratory pressure (PEEP) is increased by 1 cmH2O to promote the re-expansion of collapsed alveoli.

10. The control method for a ventilator as described in any one of claims 7-9, characterized in that: The PID algorithm is used for real-time closed-loop fine-tuning of parameters to eliminate steady-state error of parameters. The parameters output by the model are used as setpoints and the airway parameters monitored in real time are used as feedback values ​​to perform millisecond-level parameter fine-tuning. The results are transmitted to the high-frequency ventilation and normal-frequency ventilation modules through high-speed solenoid valves. The CNN-LSTM hybrid model is used for long-term prediction of ventilation status and global optimization of parameters. Based on the ventilation status assessment results, the operating parameters of the high-frequency ventilation module and the normal-frequency ventilation module are dynamically adjusted in a closed loop. Taking 10 consecutive frames of EIT images and 5 consecutive seconds of multimodal physiological parameters as input, the model outputs the globally optimal combination of ventilation parameters.

Images