Multimodal respiratory acquisition device, respiratory state prediction method, apparatus, and medium

Abstract

The present application relates to the technical field of health monitoring, and discloses a multi-modal respiration acquisition device, a respiration state prediction method, equipment and a medium, the device comprising a hardware acquisition layer and a data transmission layer; the hardware acquisition layer comprises a plurality of sensors, and multi-modal respiration signals are acquired based on a muscle-skeleton-airflow conduction mechanism; the data transmission layer controls the synchronous acquisition of the plurality of sensors. The present application integrates a plurality of sensors based on a muscle-skeleton-airflow conduction mechanism, synchronously captures multi-dimensional data of muscle activity, skeleton linkage and airflow changes in the respiration process, and avoids inaccurate respiration state prediction caused by a single data source. At the same time, the data transmission layer controls the synchronous acquisition of the plurality of sensors, ensures that different-dimensional respiration signals are aligned in the time dimension, avoids signal deviation caused by chaotic acquisition timing, provides an accurate timing basis for subsequent data processing, improves the accuracy and reliability of respiration state prediction, and is suitable for clinical scenarios with high requirements for respiration monitoring accuracy.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, specifically to a multimodal respiratory acquisition device, a respiratory state prediction method, equipment, and medium. Background Technology

[0002] In modern medicine, especially in intensive care units, accurate perception and assessment of a patient's respiratory status are crucial, directly impacting disease diagnosis, treatment adjustments, and prognosis. Existing respiratory monitoring devices often collect only single types of physiological signals, such as airflow signals from the ventilator or simple chest and abdominal movement signals, leading to inaccurate respiratory status predictions. Summary of the Invention

[0003] This invention provides a multimodal respiratory acquisition device, a respiratory state prediction method, equipment, and medium to solve the problem that existing respiratory acquisition devices often acquire only a single type of physiological signal, leading to inaccurate respiratory state prediction.

[0004] In a first aspect, the present invention provides a multimodal respiratory acquisition device, which includes a hardware acquisition layer and a data transmission layer;

[0005] The hardware acquisition layer includes multiple sensors for acquiring multimodal respiratory signals from patients based on the muscle-skeletal-airflow conduction mechanism;

[0006] The data transmission layer is used to establish a communication connection with the hardware acquisition layer and control multiple sensors to collect data synchronously.

[0007] This invention integrates multiple sensors based on the muscle-skeletal-airflow conduction mechanism through a hardware acquisition layer. This overcomes the limitations of traditional respiratory monitoring devices that only acquire single signals. It can simultaneously capture multi-dimensional data on muscle activity, skeletal coordination, and airflow changes during respiration, comprehensively reflecting the patient's core respiratory state, including intensity, rhythm, and muscle effort. This effectively avoids the bias caused by a single data source and provides more comprehensive data support for respiratory status prediction. Simultaneously, the data transmission layer establishes stable communication with the hardware acquisition layer and controls the synchronous acquisition of multiple sensors. This ensures high temporal alignment of respiratory signals from different dimensions, avoiding signal deviations caused by inconsistent acquisition timing. This provides a precise temporal basis for subsequent data processing, further improving the accuracy and reliability of respiratory status prediction. It is particularly suitable for clinical scenarios with high requirements for respiratory monitoring accuracy, such as intensive care units, demonstrating feasibility and clinical applicability, and providing an efficient and accurate solution for respiratory status monitoring of critically ill patients.

[0008] In one alternative implementation, the multiple sensors in the hardware acquisition layer include an electromyography sensor, a flow sensor, a tension band sensor, and a chest and abdomen sensor.

[0009] Electromyography (EMG) sensors are used to collect surface electromyographic signals from the patient's respiratory muscles.

[0010] A flow sensor is used to collect the flow rate signal of a patient's respiratory airflow;

[0011] Tension band sensors are used to collect chest and abdominal tension signals during a patient's respiratory movements;

[0012] The chest and abdomen sensor is used to collect chest and abdominal movement signals during a patient's respiratory movements.

[0013] This embodiment captures physiological signals from three key dimensions during respiration: respiratory muscle activity, thoracic and abdominal skeletal coordination, and airway airflow changes. This comprehensively reflects the patient's core respiratory state, such as respiratory intensity, respiratory rhythm, and respiratory muscle effort, avoiding inaccurate respiratory state predictions caused by a single data source.

[0014] In one optional implementation, the data transmission layer includes an analog-to-digital conversion module, and the flow sensor and tension band sensor are respectively connected to the analog-to-digital conversion module via signal lines;

[0015] The analog-to-digital conversion module is used to perform analog-to-digital conversion on flow signals and chest and abdominal tension signals.

[0016] This embodiment uses an analog-to-digital conversion module to convert the flow signal and the chest and abdominal tension signal into digital signals, making them suitable for the digital signal requirements of subsequent data transmission and processing, thus laying the foundation for the synchronous integration of multimodal signals.

[0017] In one alternative implementation, the chest and abdominal sensors and the electromyography (EMG) sensors communicate serially with the data transmission layer, respectively.

[0018] In this embodiment, the chest and abdominal sensors and electromyography sensors communicate serially with the data transmission layer, providing a stable transmission basis for subsequent time synchronization calibration of multimodal data and further reducing the error in respiratory status judgment caused by communication delay.

[0019] Secondly, the present invention provides a method for predicting respiratory status, the method comprising:

[0020] Acquire sample intubation invasive diaphragm data for each test patient, and simultaneously acquire sample multimodal respiratory signals of the test patients based on the multimodal respiratory acquisition device of the first aspect or any of its corresponding embodiments.

[0021] The multimodal respiratory signals and invasive diaphragmatic data of the test patients were combined into sample data. The respiratory state prediction model was trained based on the sample data of all test patients to obtain the target prediction model.

[0022] The multimodal respiratory acquisition device based on the first aspect or any corresponding embodiment acquires the target multimodal respiratory signal of the target patient;

[0023] A target prediction model is used to predict the respiratory nodes of the target patient based on the target multimodal respiratory signal.

[0024] This invention acquires invasive diaphragmatic data from test patients during intubation and uses it as the gold standard for determining respiratory nodes. However, considering the difficulties, high costs, and potential invasiveness associated with invasive diaphragmatic data acquisition, a multimodal respiratory acquisition device is used to collect multimodal respiratory signals. This data is then combined with the invasive diaphragmatic data to train a model, enabling the model to learn the mapping relationship between signals at different conduction stages and diaphragmatic activity. In practical use, the trained target prediction model is invoked, and the target patient's target multimodal respiratory signal is input. The model, by exploring the temporal conduction patterns of muscle-skeletal-airflow and fully utilizing the complementary advantages of different modal signals, accurately reconstructs the key respiratory nodes defined by diaphragmatic activity, ultimately outputting the predicted respiratory nodes for the target patient. This achieves the predictive effect of the gold standard signal for invasive intubation using non-invasive in vitro detection signals, demonstrating significant clinical application value.

[0025] In one alternative implementation, the sample cannulation invasive diaphragm data carries multiple respiratory node tags;

[0026] A respiratory status prediction model was trained based on sample data from all test patients to obtain the target prediction model, which includes:

[0027] All sample data from test patients were divided into training and validation sets;

[0028] For each sample data in the training set, the multimodal respiratory signals of the sample data are extracted into multiple time sequences according to a preset time window;

[0029] The Transformer architecture in the respiratory state prediction model is used to predict multiple respiratory nodes based on multiple time series of each sample data in the training set.

[0030] The training loss is calculated based on multiple respiratory nodes from multiple samples and multiple respiratory node labels from intubation-invasive diaphragm data in the sample data.

[0031] The breathing state prediction model is optimized based on the training loss. It then returns to the Transformer architecture used in the breathing state prediction model, and makes predictions based on multiple time series of each sample data in the training set to obtain the breathing nodes of the sample. This process continues until the performance of the breathing state prediction model on the validation set reaches the preset conditions. Finally, the last optimized breathing state prediction model is used as the target prediction model.

[0032] This embodiment provides training data through respiratory node labels, and combines the multi-head self-attention mechanism, dual decoding layer and independent prediction head of the Transformer architecture to deeply mine the single-modal temporal dependence and cross-modal correlation of multimodal signals, so as to achieve accurate prediction of key respiratory nodes and achieve prediction accuracy that reaches the gold standard of invasive non-invasive acquisition.

[0033] In one alternative implementation, the Transformer architecture includes a feedforward layer, a multi-head self-attention layer, an encoding layer, a decoding layer, a masked multi-head attention mechanism, and a prediction head;

[0034] Using the Transformer architecture in the respiratory state prediction model, predictions are made based on multiple time-series sequences of each sample data in the training set, resulting in multiple sample respiratory nodes, including:

[0035] For each sample data, a feedforward layer is used to transform multiple time series sequences of the sample data into an intermediate-order feature matrix;

[0036] A multi-head self-attention layer is used to process the intermediate-order feature matrix to obtain the fused feature matrix;

[0037] An encoding layer is used to encode the fused feature matrix to obtain a higher-order feature matrix;

[0038] A multi-head self-attention layer is used to enhance the features of the high-order feature matrix to obtain the target feature matrix.

[0039] A decoding layer is used to decode based on the target feature matrix to obtain the feature vector of the candidate respiratory node;

[0040] A masked multi-head attention mechanism is used to mask the feature vectors of candidate breathing nodes to obtain sample feature vectors.

[0041] A prediction head is used, and optimization is performed based on the sample feature vector to obtain multiple sample breathing nodes.

[0042] This embodiment effectively solves the core pain points of difficult respiratory signal feature extraction, multi-modal fusion contradictions, and large triggering delay through the full-link collaborative design of the Transformer architecture, and ultimately achieves high-precision prediction of key respiratory nodes.

[0043] Thirdly, the present invention provides a respiratory state prediction device, the device comprising:

[0044] The acquisition module is used to acquire sample intubation invasive diaphragm data of each test patient, and at the same time, based on the multimodal respiratory acquisition device of the first aspect or any corresponding embodiment, it acquires sample multimodal respiratory signals of the test patient.

[0045] The training module is used to combine the multimodal respiratory signals and invasive diaphragmatic data of the test patients into sample data, and train the respiratory state prediction model based on the sample data of all test patients to obtain the target prediction model.

[0046] The acquisition module is used to acquire the target multimodal respiratory signal of the target patient based on the multimodal respiratory acquisition device of the first aspect or any corresponding embodiment;

[0047] The prediction module is used to predict the respiratory nodes of the target patient by using a target prediction model based on the target multimodal respiratory signal.

[0048] Fourthly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the respiratory state prediction method of the first aspect or any corresponding embodiment described above.

[0049] Fifthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the respiratory state prediction method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0050] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0051] Figure 1 This is a schematic diagram of a multimodal respiratory data acquisition device according to an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of another multimodal respiratory data acquisition device according to an embodiment of the present invention;

[0053] Figure 3 This is a flowchart of a respiratory state prediction method according to an embodiment of the present invention;

[0054] Figure 4 This is a structural block diagram of a respiratory state prediction device according to an embodiment of the present invention;

[0055] Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0058] In modern medicine, especially in intensive care units, accurate perception and assessment of a patient's respiratory status are crucial, directly impacting disease diagnosis, treatment adjustments, and prognosis. Existing respiratory data acquisition devices often collect single-type physiological signals, such as only airflow signals from the ventilator or simple chest and abdominal movement signals, leading to inaccurate respiratory status prediction. This invention integrates multiple sensors at the hardware acquisition layer based on the muscle-skeleton-airflow conduction mechanism, overcoming the limitations of traditional respiratory data acquisition devices that only acquire single signals. It can simultaneously capture multi-dimensional data on muscle activity, skeletal coordination, and airflow changes during respiration, comprehensively reflecting the patient's core states such as respiratory intensity, rhythm, and muscle effort. This effectively avoids the biased judgment caused by a single data source and provides more comprehensive data support for respiratory status prediction.

[0059] Figure 1 This is a schematic diagram of a multimodal respiratory data acquisition device according to an embodiment of the present invention, such as... Figure 1 As shown, the device includes a hardware acquisition layer and a data transmission layer; the hardware acquisition layer includes multiple sensors for acquiring multimodal respiratory signals from patients based on the muscle-skeletal-airflow conduction mechanism; the data transmission layer is used to establish a communication connection with the hardware acquisition layer and control multiple sensors to acquire data synchronously.

[0060] Specifically, the patients in this embodiment of the invention are those with stable vital signs and no skin allergy symptoms in a clinical setting. The hardware acquisition layer innovatively integrates multiple sensors screened based on the muscle-skeletal-airflow conduction mechanism. This mechanism refers to the increasing signal delay between the muscle electrical signals, skeletal motion signals, and respiratory airflow signals and the breathing intention; however, these signals become increasingly easier to acquire using sensors, making acquisition less difficult. Compared to traditional respiratory acquisition devices that only collect a single type of data, this device can simultaneously acquire information from multiple physiological dimensions, comprehensively reflecting the patient's core respiratory state, such as respiratory intensity, respiratory rhythm, and respiratory muscle effort, avoiding inaccurate respiratory state predictions caused by a single data source.

[0061] The data transmission layer uses the STM32F407 development board as the core control unit. This board is equipped with a high-performance ARM Cortex-M4 core and has multi-peripheral synchronous control capabilities, which can strictly control the time synchronization error of each acquisition channel within milliseconds. This design ensures that multiple sensors can achieve synchronous signal acquisition without time difference, effectively avoiding analysis deviations caused by signal timing disorder, significantly improving the accuracy of respiratory status judgment, and providing a precise timing foundation for subsequent data processing.

[0062] pass Figure 1 The device architecture shown drives the transformation of respiratory monitoring devices from single signal monitoring to high-precision monitoring; at the same time, this portable design can be adapted to scenarios such as home respiratory rehabilitation monitoring and sleep apnea syndrome screening, helping home medical care and chronic disease management, and has a wide range of applications.

[0063] In some alternative implementations, such as Figure 1 As shown, the hardware acquisition layer includes multiple sensors such as an electromyography (EMG) sensor, a flow sensor, a tension band sensor, and a chest and abdomen sensor. The EMG sensor is used to acquire surface EMG signals of the patient's respiratory muscles; the flow sensor is used to acquire flow signals of the patient's respiratory airflow; the tension band sensor is used to acquire chest and abdominal tension signals of the patient during respiratory movements; and the chest and abdomen sensor is used to acquire chest and abdominal movement signals of the patient during respiratory movements.

[0064] Specifically, the electromyography (EMG) sensor is attached to the patient's diaphragm at the corresponding location on the body surface (determined by ultrasound), ensuring good contact between the electrode and the skin. It is used to collect surface EMG signals from the patient's respiratory muscles (such as the diaphragm and intercostal muscles), with a particular focus on capturing the diaphragm's EMG signals. The amplitude, frequency, and other characteristics of the surface EMG signals directly reflect the intensity and temporal patterns of respiratory muscle activity, providing direct and accurate data for clinical assessment of the patient's respiratory effort (e.g., the amount of muscle force exerted during spontaneous breathing) and diaphragmatic function (e.g., whether the diaphragm's contraction ability is normal).

[0065] The flow sensor is a portable gas flow measurement sensor that connects to the ventilator's airway tubing. It can simultaneously capture airflow changes during ventilator-assisted ventilation or spontaneous breathing. Specifically, the flow sensor collects the velocity and flow rate changes of the patient's respiratory airflow, outputting an analog signal with a fast 5ms response time. It can instantly capture instantaneous fluctuations in airflow and has a measurement accuracy of ±(2.5+0.5FS) (FS is the sensor's full scale). It can accurately quantify the intensity changes of respiratory airflow, providing reliable data support for judging the patient's respiratory intensity (such as tidal volume) and respiratory rhythm (such as airflow fluctuation period).

[0066] The tension band sensor is integrated into the pressure sensor of the chest and abdominal bandage (the tightness should be comfortable for the patient and not affect respiratory movements). Based on the principle of strain resistance, it works by sensing the tension of the bandage as the patient breathes, with the expansion and contraction of the chest and abdomen changing the bandage's tension. The sensor outputs a corresponding analog voltage signal by detecting these tension changes, and its sampling frequency is as high as 1kHz, allowing for real-time tracking of tension dynamics. Its function is not only to assist in determining the amplitude of chest and abdominal respiratory movements, but also to assess motor coordination through differences in tension distribution. Furthermore, it complements the chest and abdominal sensor, enhancing the comprehensiveness of respiratory movement monitoring.

[0067] The chest and abdomen sensor is an IMU (Inertial Measurement Unit) sensor based on the MPU9250 chip, fixed to the inside of the chest and abdomen bandage (the IMU module is tightly fitted to the chest and abdomen skin to prevent loosening). It is a portable respiratory motion measurement module, integrating a three-axis accelerometer and a three-axis gyroscope, outputting SPI (Serial Peripheral Interface) digital signals with a sampling frequency of up to 1kHz. Wearing it on the patient's chest and abdomen, it can accurately capture minute movements of the chest and abdomen during respiration (such as changes in the amplitude, speed, and direction of expansion / contraction), achieving non-invasive respiratory motion monitoring on the body surface. It has high response sensitivity and can reflect the dynamic characteristics of respiratory motion in real time, providing refined data for analyzing the patient's breathing patterns (such as the proportion of thoracic and abdominal breathing).

[0068] By simultaneously capturing physiological signals from three key dimensions during respiration—respiratory muscle activity, thoracic and abdominal skeletal coordination, and airway airflow changes—the four sensors comprehensively reflect the patient's core respiratory state, including respiratory intensity, respiratory rhythm, and respiratory muscle effort, thus avoiding inaccurate respiratory state predictions caused by a single data source.

[0069] In some alternative implementations, such as Figure 1As shown, the data transmission layer includes an analog-to-digital conversion module. The flow sensor and tension band sensor are connected to the analog-to-digital conversion module via signal lines. The analog-to-digital conversion module is used to convert the flow signal and the chest and abdominal tension signal from analog to digital.

[0070] Specifically, the core control unit of the data transmission layer, the STM32F407 development board, integrates an analog-to-digital converter (ADC) module for signal format conversion. Figure 2 This is a schematic diagram of another multimodal respiratory data acquisition device according to an embodiment of the present invention, such as... Figure 2 As shown, since the respiratory airflow signal collected by the flow sensor and the chest and abdominal tension signal collected by the tension band sensor are both analog signals, they cannot be directly processed by the STM32F407 development board. Therefore, a dedicated signal line is needed to precisely connect the signal output terminals of the two types of sensors to the signal input terminal of the analog-to-digital converter module. After the connection is completed, the analog-to-digital converter module will convert the input analog signal into a digital signal. This conversion process can achieve 16-bit precision signal quantization, ensuring that the signal-to-noise ratio of the converted digital signal is not less than 60dB. This preserves the subtle features of the original analog signal (such as instantaneous airflow fluctuations and small tension changes) and adapts to the requirements of subsequent data transmission and processing, laying the foundation for the synchronous integration of multimodal signals.

[0071] In some alternative implementations, such as Figure 2 As shown, the chest and abdominal sensors and electromyography sensors communicate serially with the data transmission layer.

[0072] Specifically, SPI communication is used for serial communication. Both the chest and abdominal sensors and the electromyography (EMG) sensors establish communication connections with the STM32F407 development board via SPI. To ensure that the signals collected by the two types of sensors can be transmitted to the STM32F407 development board in real time without delay, the SPI communication rate needs to be configured to be no less than 1kHz. This rate not only matches the 1kHz sampling frequency of the sensors, achieving synchronization of acquisition and transmission rhythms and avoiding data backlog or loss, but also ensures the timeliness of signal transmission, providing a stable transmission foundation for subsequent time synchronization calibration of multimodal data, and further reducing respiratory status judgment errors caused by communication delays.

[0073] Optionally, to ensure stable operation of the data transmission layer, the device is equipped with a dedicated power supply module. This power supply uses a 5V / 2A DC power supply specification, which matches the power supply requirements of the STM32F407 development board and peripheral modules, while preventing equipment downtime and data transmission interruptions due to voltage fluctuations or insufficient current. The power supply and data transmission layer are connected via a dedicated power supply interface. During connection, it is essential to ensure correct polarity, and the power supply status can be monitored in real time via a power indicator light. Furthermore, the power module has overcurrent protection; it automatically cuts off power to the data transmission layer in case of abnormal current, preventing hardware damage and further ensuring the continuity and safety of the entire data transmission link, providing stable power support for the synchronous acquisition and real-time transmission of multimodal breathing signals.

[0074] According to an embodiment of the present invention, a method for predicting respiratory state is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0075] This embodiment provides a method for predicting respiratory status, applied to a multimodal respiratory acquisition device. Figure 3 This is a flowchart of a respiratory state prediction method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0076] Step S301: Acquire sample intubation invasive diaphragm data for each test patient, and simultaneously acquire sample multimodal respiratory signals from the test patients based on the multimodal respiratory acquisition device provided in the above embodiment.

[0077] Specifically, the gold standard device for invasive diaphragmatic data acquisition via esophageal intubation for ventilators directly contacts the core area of ​​diaphragmatic physiological activity through an invasive acquisition method, accurately capturing dynamic data of diaphragmatic contraction and relaxation. As the core organ of respiratory power, the diaphragm's contraction initiation time corresponds to the start of inspiration, its peak contraction time corresponds to the peak respiratory time, and its relaxation initiation time corresponds to the start of expiration. This data directly defines the true timestamp of respiratory nodes and is the gold standard for determining respiratory nodes.

[0078] The components of the multimodal respiration collection device are arranged according to... Figure 2 The communication connection is established as shown. Multiple sensors are worn on different parts of the test patient. After confirming that all parts are connected normally and the parameters are configured correctly, the respiratory signals of the test patient are collected, and sample multimodal respiratory signals including surface electromyography signals, flow signals, chest and abdominal tension signals, and chest and abdominal movement signals are obtained.

[0079] It should be noted that the diaphragmatic data from intubation and the multimodal respiratory signals of the sample need to be collected simultaneously to ensure that their timing is aligned.

[0080] Step S302: Combine the multimodal respiratory signals and invasive diaphragmatic data of the test patients into sample data, and train the respiratory state prediction model based on the sample data of all test patients to obtain the target prediction model.

[0081] Specifically, considering the drawbacks of invasive diaphragmatic data acquisition, such as high difficulty and cost, and the potential for invasive injury to patients, this invention aims to approximate the predictive performance of invasive gold standard data using multimodal respiratory signals acquired non-invasively outside the body. Specifically, sample multimodal respiratory signals and sample invasive diaphragmatic data collected from a test patient are used as the sample data for that patient. A respiratory state prediction model is then trained based on all sample data, enabling the model to learn the mapping relationship between signals at different conduction stages and diaphragmatic activity. This achieves the modeling goal of approximating invasive gold standard signals with multimodal in vitro signals, ultimately training a target prediction model that balances signal delay and reliability.

[0082] Step S303: Based on the multimodal respiratory acquisition device provided in the above embodiments, acquire the target multimodal respiratory signal of the target patient.

[0083] Specifically, in practical applications, refer to step S301 for the process of collecting sample multimodal respiratory signals from test patients, and collect target multimodal respiratory signals from target patients.

[0084] Step S304: Using a target prediction model, prediction is performed based on the target multimodal respiratory signal to obtain the predicted respiratory nodes of the target patient.

[0085] Specifically, the trained target prediction model is invoked, and the target patient's target multimodal respiratory signal is input. By exploring the temporal transmission patterns of muscle-skeletal-airflow, the model fully utilizes the complementary advantages of different modal signals to accurately reconstruct the key respiratory nodes defined by diaphragmatic activity. Finally, it outputs predicted respiratory nodes such as the onset of inspiration, the onset of expiration, and the peak respiratory rate of the target patient. This achieves the predictive effect of the gold standard signal for invasive intubation using non-invasive external detection signals, and has significant clinical application value.

[0086] This invention acquires invasive diaphragmatic data from test patients during intubation and uses it as the gold standard for determining respiratory nodes. However, considering the difficulties, high costs, and potential invasiveness associated with invasive diaphragmatic data acquisition, a multimodal respiratory acquisition device is used to collect multimodal respiratory signals. This data is then combined with the invasive diaphragmatic data to train a model, enabling the model to learn the mapping relationship between signals at different conduction stages and diaphragmatic activity. In practical use, the trained target prediction model is invoked, and the target patient's target multimodal respiratory signal is input. The model, by exploring the temporal conduction patterns of muscle-skeletal-airflow and fully utilizing the complementary advantages of different modal signals, accurately reconstructs the key respiratory nodes defined by diaphragmatic activity, ultimately outputting the predicted respiratory nodes for the target patient. This achieves the predictive effect of the gold standard signal for invasive intubation using non-invasive in vitro detection signals, demonstrating significant clinical application value.

[0087] This embodiment provides a method for predicting respiratory status, applied to a multimodal respiratory data acquisition device. The method specifically includes the following steps:

[0088] Step S401: Acquire invasive diaphragmatic data for each test patient, and simultaneously acquire multimodal respiratory signals from the test patients using the multimodal respiratory acquisition device provided in the above embodiment. For details, please refer to... Figure 3 Step S301 of the illustrated embodiment will not be described again here.

[0089] Step S402: Combine the multimodal respiratory signals and intubation diaphragm data of the test patients into sample data, train the respiratory state prediction model based on the sample data of all test patients, and obtain the target prediction model. The intubation diaphragm data carries multiple respiratory node labels.

[0090] Specifically, step S402 above trains a respiratory state prediction model based on sample data from all test patients to obtain the target prediction model, including:

[0091] Step S4021: Divide the sample data of all test patients into training set and validation set.

[0092] Specifically, sample data from all test patients, including multimodal respiratory signals and intubation-induced diaphragmatic data, were collected and divided into training and validation sets according to a predetermined ratio. The training set was used for model parameter learning, while the validation set was used to monitor training effectiveness in real time and prevent overfitting.

[0093] Step S4022: For each sample data in the training set, the multimodal respiratory signal of the sample data is extracted into multiple time sequences according to a preset time window.

[0094] Specifically, for each sample data in the training set, based on the temporal continuity of the respiratory signal, the multimodal respiratory signal of the sample is truncated according to a preset time window (e.g., 10s / sample) to obtain multiple time sequences of equal length. During the truncation process, the alignment of each modality signal (surface electromyography, flow rate, chest and abdominal tension, and motion signals) on the time axis is maintained to ensure that each time sequence fully preserves the conduction pattern of muscle-skeletal-airflow.

[0095] Step S4023: Using the Transformer architecture in the respiratory state prediction model, prediction is performed based on multiple time-series sequences of each sample data in the training set to obtain multiple sample respiratory nodes. The Transformer architecture includes a feedforward layer, a multi-head self-attention layer, an encoding layer, a decoding layer, a masked multi-head attention mechanism, and a prediction head.

[0096] Specifically, step S4023 includes:

[0097] Step a1: For each sample data, a feedforward layer is used to transform multiple time-series sequences of the sample data into an intermediate-order feature matrix.

[0098] Specifically, by using linear transformation, ReLU nonlinear activation, and layer normalization, the time series is transformed into a mid-order feature matrix with uniform dimensions and reduced noise, which is then adapted to the input of the subsequent attention mechanism.

[0099] Step a2: A multi-head self-attention layer is used to process the intermediate-order feature matrix to obtain the fused feature matrix.

[0100] Specifically, the intermediate-order feature matrix is ​​split into N parallel attention heads, each assigned a specific task according to the muscle-skeletal-airflow conduction mechanism. For example, head 1 focuses on the temporal lead characteristics of motion signals, capturing the activation timing of the abdominal IMU before inhalation; head 2 focuses on the conduction lag pattern of airflow signals, tracking the attenuation process of airflow intensity during exhalation; and heads 3-5 are dedicated to calculating cross-modal correlations. After all attention heads independently complete the weight calculation, the output feature matrices are concatenated into a high-dimensional matrix, and then the dimension is restored through a linear transformation matrix. Finally, a fusion feature matrix integrating single-modal temporal dependencies and cross-modal correlations is obtained, fully leveraging the complementary advantages of multimodal signals.

[0101] Step a3: Use an encoding layer to encode the fused feature matrix to obtain a high-order feature matrix.

[0102] Specifically, a linear transformation is performed on the fused feature matrix to deeply explore the coordinated transmission patterns of muscle-skeletal-airflow. For example, the linkage pattern of electromyography activation before inhalation → movement initiation → airflow preparation and the decreasing pattern of airflow attenuation → tension relaxation → electromyography weakening during exhalation are extracted. At the same time, residual connections are introduced to directly add the fused features to the processed features, effectively alleviating the gradient vanishing problem in deep training. Combined with layer normalization to stabilize the feature distribution, the final output is a high-order feature matrix with strong abstract representation capabilities.

[0103] Step a4: Use a multi-head self-attention layer to enhance the features of the high-order feature matrix to obtain the target feature matrix.

[0104] Specifically, a multi-head self-attention mechanism is used again to focus on and enhance the specific features of the respiratory nodes: for example, for the inspiratory start node, the synergistic features of the sudden increase in electromyography and the initiation of movement are amplified; for the respiratory peak node, the correlation features between the extreme airflow value and the peak of the movement amplitude are strengthened; and for the expiration start node, the abrupt changes in airflow decay and relaxation of body surface tension are highlighted. At the same time, through dynamic weight adjustment, redundant information unrelated to the respiratory nodes is suppressed, and the cross-modal fusion weights are further optimized, ultimately resulting in a target feature matrix with extremely high recognizability of respiratory node features.

[0105] Step a5: A decoding layer is used to decode based on the target feature matrix to obtain the feature vector of the candidate respiratory node.

[0106] Specifically, the decoding layer is a dual decoding layer. The first layer is used to filter out the approximate temporal range containing breathing nodes from the target feature matrix, such as the inhalation start node appearing within the next 1.2-2.8 seconds. The second layer is used to mine the feature mutation patterns before and after the node within the coarse localization range, such as airflow initiation at the start of inhalation and tension decay at the start of exhalation. By adapting the feature dimension through linear transformation, the core feature vector corresponding to each candidate breathing node is finally output.

[0107] Step a6: Using a masked multi-head attention mechanism, the feature vectors of candidate breathing nodes are masked to obtain sample feature vectors.

[0108] Specifically, a lower triangular mask matrix is ​​introduced to process the feature vectors of candidate breathing nodes. For example, when calculating the features at t=25s, all future feature information at t=25.1s, 25.2s and beyond is masked to ensure that the model makes predictions based only on historical data (t≤25s) and current data, and finally outputs the sample feature vectors.

[0109] Step a7: Using a prediction head, optimization is performed based on the sample feature vector to obtain multiple sample breathing nodes.

[0110] Specifically, a linear transformation is first performed on the basic prediction layer to convert the sample feature vector into the time difference between the current moment and the next breathing node, for example, outputting 0.48s before the start of inhalation. Then, three independent prediction branches are designed for three types of core breathing nodes: the inhalation start head focuses on the characteristics of electromyography surge and motor initiation, calibrating the basic prediction time difference of 0.5s to 0.48s; the exhalation start head corrects the time difference of 0.6s to 0.59s based on the characteristics of airflow decay and tension relaxation; and the peak node head refines the time difference of 0.3s to 0.293s based on the characteristics of airflow extreme value and motor peak value. Finally, combining the current time (e.g., 25.309s), the specific timestamps of the nodes are calculated: inhalation start timestamp: 25.309s + 0.48s = 25.789s; exhalation start timestamp: 25.309s + 0.59s = 25.899s; peak respiratory timetamp: 25.309s + 0.293s = 25.592s. The precise timestamps of the three sample respiratory nodes—inhalation start, exhalation start, and peak respiratory timetamp—are finally output.

[0111] Step S4024: Calculate the training loss based on multiple respiratory nodes of multiple samples and multiple respiratory node labels of intubation invasive diaphragm data in the sample data.

[0112] Specifically, intubated diaphragmatic data is used as the gold standard for respiratory node determination. It carries three core respiratory node labels—inspiratory onset, expiratory onset, and peak respiratory rate—precisely annotated manually, along with the specific timestamps of each respiratory node occurrence. The model's output respiratory nodes are aligned with these gold standard labels according to node type. A loss function is used to calculate the loss between the predicted time and the labeled time for each node type, quantifying the deviation of the model's prediction from the gold standard. For example, if the label for the inspiratory onset node of a sample is 25.792s, and the model's predicted value is 25.789s, the error is 3ms. Combining this with the errors of other nodes, the overall loss for that sample can be calculated.

[0113] Step S4025: Optimize the respiratory state prediction model based on training loss, return to the Transformer architecture in the respiratory state prediction model, and make predictions based on multiple time series of each sample data in the training set to obtain the sample respiratory nodes, until the performance of the respiratory state prediction model on the validation set reaches the preset conditions, and use the last optimized respiratory state prediction model as the target prediction model.

[0114] Specifically, the AdamW optimizer is used to optimize model parameters based on the calculated training loss through backpropagation. Using the optimized model, the process returns to step S4023 to continue processing the next sample data in the training set. After each optimization, the model performance is evaluated using a validation set, such as the Weighted Absolute Percentage Error (WAPE). Training stops when the model's performance on the validation set reaches a preset condition, and the model after the last optimization is used as the target prediction model, which balances prediction accuracy and generalization ability. This preset condition can be defined as the WAPE curve stabilizing over multiple rounds with no further decrease potential.

[0115] Step S403: Based on the multimodal respiratory acquisition device provided in the above embodiment, acquire the target multimodal respiratory signal of the target patient. For details, please refer to... Figure 3 Step S303 of the illustrated embodiment will not be described again here.

[0116] Step S404: Using a target prediction model, predictions are made based on the target multimodal respiratory signals to obtain the predicted respiratory nodes of the target patient. For details, please refer to [link to details]. Figure 3 Step S304 of the illustrated embodiment will not be described again here.

[0117] This invention acquires invasive diaphragmatic data from test patients during intubation and uses it as the gold standard for determining respiratory nodes. However, considering the difficulties, high costs, and potential invasiveness associated with invasive diaphragmatic data acquisition, a multimodal respiratory acquisition device is used to collect multimodal respiratory signals. This data is then combined with the invasive diaphragmatic data to train a model, enabling the model to learn the mapping relationship between signals at different conduction stages and diaphragmatic activity. In practical use, the trained target prediction model is invoked, and the target patient's target multimodal respiratory signal is input. The model, by exploring the temporal conduction patterns of muscle-skeletal-airflow and fully utilizing the complementary advantages of different modal signals, accurately reconstructs the key respiratory nodes defined by diaphragmatic activity, ultimately outputting the predicted respiratory nodes for the target patient. This achieves the predictive effect of the gold standard signal for invasive intubation using non-invasive in vitro detection signals, demonstrating significant clinical application value.

[0118] This embodiment also provides a respiratory state prediction device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0119] This embodiment provides a respiratory status prediction device, such as... Figure 4As shown, it includes:

[0120] The acquisition module 401 is used to acquire sample intubation invasive diaphragm data of each test patient, and at the same time, based on the multimodal respiratory acquisition device provided in the above embodiment, it acquires sample multimodal respiratory signals of the test patients.

[0121] The training module 402 is used to combine the multimodal respiratory signals and invasive diaphragmatic data of the test patients into sample data, and to train the respiratory state prediction model based on the sample data of all test patients to obtain the target prediction model.

[0122] The acquisition module 403 is used to acquire the target multimodal respiratory signal of the target patient based on the multimodal respiratory acquisition device provided in the above embodiments.

[0123] The prediction module 404 is used to use a target prediction model to predict the target patient's respiratory nodes based on the target multimodal respiratory signal.

[0124] In some alternative implementations, the sample cannulation invasive diaphragm data carries multiple respiratory node tags;

[0125] Training module 402 includes:

[0126] The partitioning unit is used to divide the sample data of all test patients into training and validation sets.

[0127] The truncation unit is used to truncate the multimodal respiratory signals of each sample in the training set into multiple time sequences according to a preset time window.

[0128] The prediction unit is used to make predictions based on multiple time-series sequences of each sample data in the training set using the Transformer architecture in the respiratory state prediction model, thereby obtaining multiple sample respiratory nodes.

[0129] The computational unit is used to calculate the training loss based on multiple respiratory nodes of multiple samples and multiple respiratory node labels of sample intubation invasive diaphragm data in the sample data.

[0130] The training unit is used to optimize the respiratory state prediction model based on the training loss. It returns to the Transformer architecture in the respiratory state prediction model and makes predictions based on multiple time series of each sample data in the training set to obtain the sample respiratory nodes. This process continues until the performance of the respiratory state prediction model on the validation set reaches the preset conditions. The last optimized respiratory state prediction model is then used as the target prediction model.

[0131] In some alternative implementations, the Transformer architecture includes a feedforward layer, a multi-head self-attention layer, an encoding layer, a decoding layer, a masked multi-head attention mechanism, and a prediction head;

[0132] Prediction unit, including:

[0133] The transformation subunit is used to transform multiple time-series sequences of sample data into an intermediate-order feature matrix for each sample data using a feedforward layer.

[0134] The first processing subunit is used to process the intermediate-order feature matrix using a multi-head self-attention layer to obtain the fused feature matrix.

[0135] The encoding subunit is used to encode the fused feature matrix using the encoding layer to obtain a higher-order feature matrix.

[0136] The enhancement subunit is used to enhance the features of the high-order feature matrix using a multi-head self-attention layer to obtain the target feature matrix.

[0137] The decoding subunit is used to decode the target feature matrix using the decoding layer to obtain the feature vector of the candidate respiratory node.

[0138] The second processing subunit is used to mask the feature vectors of candidate breathing nodes using a masked multi-head attention mechanism to obtain sample feature vectors.

[0139] The optimization subunit is used to optimize based on the sample feature vector using the prediction head to obtain multiple sample breathing nodes.

[0140] The respiratory state prediction device provided in this embodiment of the invention can execute the respiratory state prediction method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0141] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0142] The following is a detailed reference. Figure 5The diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 501, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 502 or a program loaded from memory 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device. The processor 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0143] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0144] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a memory 508, or installed from a ROM 502. When the computer program is executed by the processor 501, it performs the functions defined in the respiratory state prediction method of the embodiments of the present invention.

[0145] Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0146] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the respiratory state prediction method shown in the above embodiments is implemented.

[0147] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0148] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for predicting respiratory state, characterized in that, The method includes: Data on the diaphragm during intubation of each test patient was acquired. Simultaneously, multimodal respiratory signals from the test patients were collected using a multimodal respiratory acquisition device. This device includes a hardware acquisition layer and a data transmission layer. The hardware acquisition layer includes multiple sensors for acquiring the patient's multimodal respiratory signals based on the muscle-skeleton-airflow conduction mechanism. The data transmission layer establishes a communication connection with the hardware acquisition layer and controls the multiple sensors to acquire signals synchronously. The multiple sensors in the hardware acquisition layer include an electromyography (EMG) sensor, a flow sensor, a tension band sensor, and a chest / abdominal sensor. The EMG sensor is used to acquire the... The system collects surface electromyographic (EMG) signals of the patient's respiratory muscles. A flow sensor is used to acquire the flow rate signal of the patient's respiratory airflow. A tension band sensor is used to acquire the chest and abdominal tension signals of the patient during respiratory movements. A chest and abdominal sensor is used to acquire the chest and abdominal movement signals of the patient during respiratory movements. The data transmission layer includes an analog-to-digital conversion module. The flow sensor and the tension band sensor are respectively connected to the analog-to-digital conversion module via signal lines. The analog-to-digital conversion module is used to perform analog-to-digital conversion on the flow rate signal and the chest and abdominal tension signal. The chest and abdominal sensor and the EMG sensor communicate serially with the data transmission layer. The multimodal respiratory signals and invasive diaphragmatic data of the test patients were combined into sample data. A respiratory state prediction model was trained based on the sample data of all test patients to obtain the target prediction model. Based on the multimodal respiratory acquisition device, target multimodal respiratory signals of the target patient are acquired; Using the target prediction model, predictions are made based on the target multimodal respiratory signals to obtain the predicted respiratory nodes of the target patient.

2. The method according to claim 1, characterized in that, The sample intubation invasive diaphragmatic data carries multiple respiratory node tags. The respiratory state prediction model is trained based on sample data from all test patients to obtain the target prediction model, including: All sample data from test patients were divided into training and validation sets; For each sample data in the training set, the multimodal respiratory signal of the sample data is extracted into multiple time sequences according to a preset time window; Using the Transformer architecture in the respiratory state prediction model, prediction is performed based on multiple time-series sequences of each sample data in the training set to obtain multiple sample respiratory nodes. The training loss is calculated based on the multiple respiratory nodes of the samples and the multiple respiratory node labels of the intubated diaphragm data in the training set. The breathing state prediction model is optimized based on the training loss, and then returned to the step of using the Transformer architecture in the breathing state prediction model to predict the breathing nodes of the sample based on multiple time series of each sample data in the training set, until the performance of the breathing state prediction model on the validation set reaches the preset condition, and the last optimized breathing state prediction model is used as the target prediction model.

3. The method according to claim 2, characterized in that, The Transformer architecture includes a feedforward layer, a multi-head self-attention layer, an encoding layer, a decoding layer, a masked multi-head attention mechanism, and a prediction head; The respiratory state prediction model employs the Transformer architecture to predict multiple respiratory nodes based on multiple time-series sequences of each sample data in the training set, including: For each sample data, the feedforward layer is used to transform multiple time-series sequences of the sample data into an intermediate-order feature matrix; The multi-head self-attention layer is used to process the intermediate-order feature matrix to obtain a fused feature matrix; The fused feature matrix is ​​encoded using the aforementioned encoding layer to obtain a higher-order feature matrix; The multi-head self-attention layer is used to enhance the features of the higher-order feature matrix to obtain the target feature matrix; The decoding layer is used to decode based on the target feature matrix to obtain the candidate respiratory node feature vector; The masked multi-head attention mechanism is used to mask the feature vectors of the candidate breathing nodes to obtain sample feature vectors. The prediction head is used to optimize the sample feature vector to obtain the multiple sample breathing nodes.

4. A respiratory state prediction device, characterized in that, The device includes: The acquisition module is used to acquire invasive diaphragmatic data from each test patient's sample via intubation. Simultaneously, based on a multimodal respiratory acquisition device, it acquires multimodal respiratory signals from the test patients' samples. The multimodal respiratory acquisition device includes a hardware acquisition layer and a data transmission layer. The hardware acquisition layer includes multiple sensors for acquiring the patient's multimodal respiratory signals based on the muscle-skeleton-airflow conduction mechanism. The data transmission layer establishes a communication connection with the hardware acquisition layer and controls the multiple sensors to acquire signals synchronously. The multiple sensors in the hardware acquisition layer include an electromyography (EMG) sensor, a flow sensor, a tension band sensor, and a chest / abdomen sensor. The EMG sensor is used for... The system collects surface electromyographic signals of the patient's respiratory muscles. A flow sensor collects the flow rate signal of the patient's respiratory airflow. A tension band sensor collects the chest and abdominal tension signals of the patient during respiratory movements. A chest and abdominal sensor collects the chest and abdominal movement signals of the patient during respiratory movements. The data transmission layer includes an analog-to-digital conversion module. The flow sensor and the tension band sensor are connected to the analog-to-digital conversion module via signal lines. The analog-to-digital conversion module performs analog-to-digital conversion on the flow rate signal and the chest and abdominal tension signal. The chest and abdominal sensor and the electromyographic sensor communicate serially with the data transmission layer. The training module is used to combine the multimodal respiratory signals and invasive diaphragmatic data of the test patients into sample data, and train the respiratory state prediction model based on the sample data of all test patients to obtain the target prediction model. The acquisition module is used to acquire the target multimodal respiratory signal of the target patient based on the multimodal respiratory acquisition device. The prediction module is used to use the target prediction model to predict the target multimodal respiratory signal and obtain the predicted respiratory nodes of the target patient.

5. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the respiratory state prediction method according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the respiratory state prediction method according to any one of claims 1 to 3.

Images