Cardio-pulmonary resuscitation system based on artificial intelligence and application method

Abstract

The invention provides a cardio-pulmonary resuscitation system based on artificial intelligence and an application method, and the system comprises a sudden cardiac arrest early warning EMS system which collects physiological information, carries out the feature extraction and classification reasoning, diagnoses whether the heart is suddenly stopped or not according to the physiological information, and if yes, automatically dials 120 and sends the position to a medical system; the automatic defibrillation system is used for collecting electrocardiogram information of a patient, processing electrocardiogram data of ventricular fibrillation waveforms, constructing an electrocardiogram ventricular fibrillation recognition model, and sending out a voice prompt and defibrillating after the ventricular fibrillation is diagnosed; the automatic external chest compression system is used for starting automatic external chest compression after cardiac arrest of the patient is determined, identifying carotid artery pulse and continuously monitoring blood pressure, and adjusting amplitude and frequency of external chest compression in a self-feedback manner according to identification of carotid artery fluctuation and pressure. The invention provides an external chest compression system capable of accurately evaluating cardiac arrest, quickly starting an EMS system, accurately identifying ventricular fibrillation rhythm, starting an automatic defibrillation system and having a self-feedback function.

Description

Technical Field

[0001] The present invention relates to the field of computer-aided technology, and more particularly to an artificial intelligence-based cardiopulmonary resuscitation system and an application method. Background Art

[0002] Sudden cardiac death (SCD) is a serious public health issue that threatens public health and safety. The occurrence of SCD not only brings devastating damage to families but also increases the social burden. To this end, the Chinese government has implemented a series of policies and measures to improve the public's ability to quickly perform cardiopulmonary resuscitation (CPR) on patients with cardiac arrest. However, since the vast majority of the public lacks basic medical first aid knowledge, the effectiveness of rescue efforts in the event of sudden cardiac arrest is limited.

[0003] With the development of information technology, the clinical application of artificial intelligence (AI) continues to deepen. Through deep learning and human-computer interaction, AI can integrate and correlate different types of information to perform increasingly complex tasks. Currently, AI is widely used in some clinical departments, and has also achieved significant advancements in 3D human anatomy teaching models. However, the role of AI in rescuing critically ill patients has yet to be fully realized, and treatment still primarily relies on the personal clinical experience of medical staff.

[0004] In order to solve the problem of unsatisfactory rescue effects after people encounter sudden cardiac arrest events and alleviate the problem of shortage of medical personnel due to cardiopulmonary resuscitation, it is planned to use AI technology to monitor the basic vital signs of high-risk patients in real time through portable wearable devices, build an emergency medical service (EMS) system with early warning, early detection and early activation of cardiac arrest, and develop cardiopulmonary resuscitation devices that can automatically press according to changes in the electrocardiogram to improve the success rate of treatment of cardiac arrest patients. Summary of the Invention

[0005] In response to the technical problems existing in the prior art, the present invention provides an artificial intelligence-based cardiopulmonary resuscitation system and application method, aiming to provide a system that can accurately assess cardiac arrest and quickly activate the EMS system, accurately identify ventricular fibrillation rhythm and activate the automatic defibrillation system, and a chest compression system with self-feedback.

[0006] According to a first aspect of the present invention, there is provided an artificial intelligence-based cardiopulmonary resuscitation system, comprising: The cardiac arrest early warning EMS system is used to collect physiological information through a portable wearable device, process the collected data, extract features and perform classification reasoning on the processed data, and diagnose whether the heart has suddenly stopped according to the physiological information. If so, it automatically calls 120 and sends the location to the medical system; The automatic defibrillation system is used to collect the patient's electrocardiogram information, process the electrocardiogram data of ventricular fibrillation waveforms by selecting the optimal wavelet basis and threshold suitable for the characteristics of the electrocardiogram signal, and build an electrocardiogram ventricular fibrillation recognition model. When ventricular fibrillation is diagnosed, a voice reminder is issued and defibrillation is performed; The automatic chest compression system is used to start automatic chest compression when it is determined that the patient has suffered cardiac arrest, identify the carotid pulse and continuously monitor blood pressure. Based on the identification of carotid artery fluctuations and pressure, the amplitude and frequency of chest compressions are adjusted through self-feedback.

[0007] On the basis of the above technical solution, the present invention can also make the following improvements.

[0008] Preferably, collecting physiological information through a portable wearable device and processing the collected data includes: Pulse and blood oxygen data are collected through the bracelet, marked and stored by timestamp. Through pattern recognition and trend analysis of historical data, the dynamic changes of patients' vital signs are captured, and the collected pulse and blood oxygen data are checked for integrity and consistency, missing values are supplemented, and abnormal values in the collection are eliminated. Preferably, the feature extraction and classification reasoning of the processed data includes: Perform time series modeling on pulse and blood oxygen data and use the exponentially weighted moving average algorithm to smooth the data to ensure that the model can capture trend characteristics; During the feature extraction process, the first-order and second-order features of peak value, fluctuation amplitude and change rate are extracted based on the periodicity and amplitude characteristics of pulse and blood oxygen signals. At the same time, time delay features are introduced to enhance the model's ability to model time dependencies.

[0009] Preferably, when the cardiac arrest warning EMS system detects a pulse of ≤40 beats / minute and lasts for 10 seconds, a yellow warning is triggered, and an extreme bradycardia warning is sent to close contacts; when the pulse is ≤30 beats / minute and lasts for 10 seconds, a red warning is triggered, and the mobile phone automatically sends an emergency call to the EMS center and uses GPS to locate the user's position to ensure a quick response in an emergency.

[0010] Preferably, the automatic defibrillation system includes an ECG acquisition and defibrillation device, which adopts intelligent charging management, automatically maintains the charging state when connected to the central device, and immediately starts single-electrode ECG signal acquisition after disconnection.

[0011] Preferably, the ECG acquisition and defibrillation device transmits the collected ECG signals to the central device. After identifying the ventricular fibrillation ECG waveform, the central device will give a voice reminder that defibrillation is about to begin, and say "Defibrillation is about to begin, please do not have physical contact with the patient!"; during the entire resuscitation process, the ECG acquisition and defibrillation device records the ECG waveform in real time and transmits it to the central device for long-term monitoring; if the ECG signal shows that the patient has recovered normal sinus rhythm, the central device will give a voice reminder that "sinus rhythm has been restored" and call the patient again.

[0012] Preferably, the defibrillation process of the automatic defibrillation system includes: First, the ECG data containing ventricular fibrillation waveforms were processed and standardized to remove abnormal samples. A high-quality dataset was constructed through quality control. Comprehensive data cleaning was also performed. Wavelet transform was used to analyze and extract features from the ECG signals. Key features were extracted from the signals by selecting the optimal wavelet basis and threshold that were suitable for the ECG signal characteristics. Convolutional neural networks and local attention image recognition are used to construct an ECG ventricular fibrillation recognition model; the input ECG data is processed through multiple convolutional layers of the convolutional neural network. Each convolutional layer is equipped with multiple filters to detect waveform features in different directions and scales. The pooling layer further samples the feature map, reduces the feature dimension, retains key information while improving computational efficiency, and constitutes feature data extracted continuously at multiple scales under the neural convolutional network module.

[0013] Preferably, the automatic chest compression system has the function of automatically identifying venous pulse, monitoring changes in transcutaneous blood oxygen saturation and carotid artery pressure, and timely adjusting the compression depth and compression frequency according to the carotid artery pulse pressure difference and blood oxygen saturation.

[0014] Preferably, the automatic chest compression system integrates an ultrasonic transducer, a pressure sensor, and a photoplethysmography sensor placed at the carotid artery to accurately identify the carotid pulse and continuously monitor blood pressure; including: A pressure sensor is used to obtain weak mechanical fluctuations in the carotid artery area, and combined with the periodic characteristics of blood volume changes reflected in the photoplethysmography sensor signal, an initial pulse recognition model is constructed. An ultrasonic transducer is used to measure changes in arterial wall size and excite multi-frequency sound waves. When the arterial wall reaches a resonant state, the synchronized vibration frequency is used to infer the arterial wall tension, and then the real-time blood pressure in the blood vessels is calculated.

[0015] According to a second aspect of the present invention, there is provided a method for applying an artificial intelligence-based cardiopulmonary resuscitation system, comprising the following steps: Build a wearable, portable cardiac arrest early warning EMS system. This system collects physiological information through a convenient wearable device, processes the collected data, extracts features and performs classification inference on the processed data, and diagnoses cardiac arrest based on the physiological information. If so, it automatically calls 120 and sends the patient's location to the medical system. Build a cardiac arrest AI model and automatic defibrillation system based on surface ECG data. Collect the patient's ECG information, select the optimal wavelet basis and threshold suitable for the ECG signal characteristics to process the ECG data of ventricular fibrillation waveforms, and build an ECG ventricular fibrillation recognition model. Once ventricular fibrillation is diagnosed, a voice reminder is issued and defibrillation is performed. Build an intelligent automatic chest compression system that can accept feedback information regulation. When it is determined that the patient has suffered cardiac arrest, automatic chest compression is started, and the carotid pulse and continuous blood pressure monitoring are performed. Based on the identification of carotid artery fluctuations and pressure, the amplitude and frequency of chest compressions are adjusted through self-feedback.

[0016] Technical effects and advantages of the present invention: The present invention discloses an artificial intelligence-based cardiopulmonary resuscitation system and application method. The system obtains the vital signs of a patient monitored by a wearable wristband, inputs the vital sign information into a trained cardiac arrest diagnostic model for analysis, and activates the EMS system. The system recognizes the patient's electrocardiogram (ECG) signal based on the wearable device, inputs a trained ventricular fibrillation model, and activates automatic defibrillation and, at the same time, chest compression with self-feedback function. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of an artificial intelligence-based cardiopulmonary resuscitation system provided by an embodiment of the present invention; Figure 2 A schematic diagram of a wearable and portable cardiac arrest warning EMS system according to an embodiment of the present invention; Figure 3 A schematic diagram of an automatic defibrillator system provided in an embodiment of the present invention; Figure 4 A schematic diagram of an intelligent automatic chest compression system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] Sudden cardiac death (SCD) refers to a critical phenomenon in which the heart's electrical and mechanical activity suddenly ceases, leading to systemic circulatory disruption. Without timely treatment, the patient will die within minutes. It has a high incidence and mortality rate, and there are significant regional disparities in the success rate of cardiac arrest treatment. This is primarily due to differences in the proportion and time it takes for cardiac arrest patients to receive emergency medical assistance. Big data shows that the time from the discovery of a patient's cardiac arrest to the activation of EMS is approximately 14.4 minutes, while the arrival time of EMS personnel is approximately 26.7 minutes, both of which exceed the optimal clinical treatment time. The earlier defibrillation is performed within 10 minutes of cardiac arrest, the higher the patient's survival rate. However, the survival rate of patients discharged from the hospital after 10 minutes is only 13.2%. Chest compression requires professional training. Even professional medical staff will become physically exhausted after a long period of manual chest compression and will not be able to perform continuous and effective compressions consistently. Therefore, how to accurately and quickly identify cardiac arrest through intelligent monitoring equipment and promptly initiate EMS to perform high-quality chest compressions is of great value in improving the success rate of resuscitation of cardiac arrest patients.

[0020] It is understandable that based on the defects in the background technology, the embodiment of the present invention proposes a cardiopulmonary resuscitation system based on artificial intelligence, specifically Figure 1 As shown, the system includes: The cardiac arrest warning EMS system is used to collect physiological information through a convenient wearable wristband, process the collected data, extract features and perform classification reasoning on the processed data, and diagnose whether the heart has suddenly stopped according to the physiological information. If so, it automatically calls 120 and sends the location to the medical system; like Figure 2 As shown in the figure, the cardiac arrest warning EMS system collects pulse and blood oxygen information based on the wristband worn by the patient. If the pulse is ≤40 beats / minute and lasts for more than 10 seconds, the wristband will vibrate, and the mobile phone will automatically contact the close contacts set by the patient and give an extremely bradycardia warning; if the pulse is ≤30 beats / minute and lasts for 10 seconds, it is considered that cardiac arrest has occurred. The mobile phone will automatically call 120 and send the location to the medical system.

[0021] The early warning process of the cardiac arrest early warning EMS system includes: 1.1. Collection of physiological information: First, you need to register a developer account and obtain the corresponding application key or token in order to legally access the data interface of the smart bracelet-related APP. By calling the API interface, the pulse and blood oxygen vital signs data uploaded by the bracelet are collected in real time. To ensure the real-time and stability of the data, the WebSocket protocol is used to establish a persistent connection, so that the server can actively push the latest data to the client, reducing data delays caused by polling. In addition, to prevent data loss or transmission interruption, a data buffer is set up, and a data streaming storage and backup mechanism is adopted to ensure that the data can be automatically retransmitted and restored in the event of network instability or interruption. The collected pulse and blood oxygen data are marked and stored in categories according to timestamps. Through pattern recognition and trend analysis of historical data, the dynamic changes of the patient's vital signs are captured, thereby providing data support for subsequent anomaly detection, early warning model training and personalized intervention strategies.

[0022] 1.2. Data preprocessing: Check the integrity and consistency of the collected pulse and blood oxygen vital signs data, supplement missing values, eliminate outliers in the collection, and ensure the accuracy and reliability of the data. Use the exponentially weighted moving average (EWMA) to smooth the time series data to eliminate short-term fluctuations and noise interference. At the same time, for possible short-term signal interruptions or data loss, use interpolation methods based on historical patterns or time series models (such as ARIMA) to fill missing values to maintain the continuity and integrity of the data. In order to enhance the adaptability of the model to different individuals, the data needs to be standardized and normalized to eliminate the differences in vital signs baselines between different individuals and ensure the consistency of the model input data. The core formula of the exponentially weighted moving average (EWMA) is as follows:

[0023] in, It's time The exponentially weighted moving average of It's time The original data value of is the smoothing factor , usually take , is the weighted average of the previous moment.

[0024] 1.3. Feature extraction and classification inference on the processed data: Time series modeling is performed on pulse and blood oxygen data, and the data is smoothed using EWMA technology to ensure that the model can capture trend characteristics. During the feature extraction process, first-order and second-order features such as peak value, fluctuation amplitude, and rate of change are extracted based on the periodicity and amplitude characteristics of the pulse and blood oxygen signals. At the same time, time delay features (such as the distance between adjacent peaks and the rate of change of amplitude) are introduced to enhance the model's ability to model time-dependent relationships. To improve feature effectiveness, principal component analysis (PCA) or linear discriminant analysis (LDA) is used to reduce the dimensionality of the feature space, remove redundancy, and reduce noise interference.

[0025] During the modeling phase, the CART (Classification and Regression Trees) algorithm is used. Recursive partitioning is performed using a greedy strategy, and the optimal partitioning features are selected based on information gain (Gain) or the Gini index (Gini). The tree depth and number of split nodes are dynamically adjusted to prevent overfitting or underfitting. Cross-validation and pruning techniques are introduced to optimize the tree structure and improve the model's generalization ability. To enhance the model's robustness on unbalanced datasets, the SMOTE method is introduced to synthetically augment minority class data. At the same time, the Focal Loss function is used to adjust class weights and balance classification accuracy. The formulas for the Gini function or information gain (Gain) function used in this process are as follows: and

[0026] in, is the number of samples that are moved to the left node, is the number of samples divided to the right node, N is the total number of samples divided, The number of CK classes on the left node, The number of CR classes on the left node.

[0027] During the model deployment phase, the model is integrated into the mobile device based on a real-time inference framework. Lightweight optimization reduces the computational burden and ensures the model's responsiveness and stability in real-time environments. When the system detects a pulse of ≤40 beats / minute for 10 seconds, a yellow alert is triggered, sending an extreme bradycardia warning to close contacts. A red alert is triggered when the pulse is ≤30 beats / minute for 10 seconds. The phone automatically sends an emergency call to the EMS center and uses GPS to locate the user, ensuring a rapid response in emergencies.

[0028] The automatic defibrillation system is used to collect the patient's electrocardiogram information, process the electrocardiogram data of ventricular fibrillation waveforms by selecting the optimal wavelet basis and threshold suitable for the characteristics of the electrocardiogram signal, and build an electrocardiogram ventricular fibrillation recognition model. When ventricular fibrillation is diagnosed, a voice reminder is issued and defibrillation is performed; like Figure 3 As shown, the automated defibrillation system includes a portable ECG acquisition and defibrillation device. This portable defibrillator utilizes intelligent charging management, automatically maintaining its charge status when connected to a central unit and immediately starting single-electrode ECG signal acquisition upon disconnection. To ensure real-time transmission, an optimized Bluetooth Low Energy (BLE) communication architecture is employed, incorporating data compression technology to reduce transmission overhead. Adaptive frequency hopping strategies and dynamic protocol stack parameter adjustment improve wireless transmission efficiency, reduce latency, and increase throughput. The system is also tested in real time, including metrics such as data transmission latency and packet loss rate. Acceptable real-time thresholds are set based on requirements. Furthermore, the device incorporates a defibrillation function, transmitting the acquired ECG signals to the central unit. Upon identifying a ventricular fibrillation ECG waveform, the central unit issues a voice notification, stating, "Defibrillation is imminent. Please avoid physical contact with the patient!" Throughout the resuscitation process, the device records the ECG waveform in real time and transmits it to the central unit for long-term monitoring. If the ECG signal indicates the patient has returned to normal sinus rhythm, the central unit issues a voice notification, stating, "Sinus rhythm has been restored," and calls out to the patient.

[0029] The defibrillation process of the automatic defibrillator system includes: 2.1. Processing large amounts of ECG data containing ventricular fibrillation waveforms: Standardize the data, remove abnormal samples, and build a high-quality data set through strict quality control. At the same time, carry out comprehensive data cleaning and study the use of wavelet transform to analyze and extract features from ECG signals. By selecting the optimal wavelet basis and threshold that are suitable for the characteristics of the ECG signal, the key features in the signal can be effectively extracted. The wavelet transform performs multi-scale decomposition and refinement analysis of the signal through operations such as scaling and translation, which helps to separate noise components from complex ECG signals while retaining important waveform features. This method can significantly improve the signal-to-noise ratio and extraction accuracy of the signal, providing a solid data foundation for subsequent model training and ECG recognition. The formula is as follows:

[0030] Where: a is the scale factor, b is the translation, f(t) is the original signal, and ψ(t) is the wavelet function.

[0031] Annotation tools were used to calibrate the ECG images. Labels included normal ECG, supraventricular premature beats, atrial fibrillation, atrial flutter, ventricular premature beats, and ventricular fibrillation. The dataset was then divided into training, validation, and test sets for subsequent training of the ECG ventricular fibrillation recognition model.

[0032] 2.3. Integrating a Deep Learning Network for ECG Ventricular Fibrillation Identification: To accurately identify ventricular fibrillation waveforms, a high-precision ventricular fibrillation identification model integrating a neural network was constructed to provide technical support for clinical decision-making. A CNN and AutoFocus Transformer were used to construct an ECG ventricular fibrillation identification model. First, the input ECG data was processed through multiple convolutional layers of the CNN network. Each convolutional layer was equipped with multiple filters to detect waveform features at different directions and scales, such as local patterns such as edges, peaks, and troughs in the ECG signal. The pooling layer further samples the feature map, reducing the feature dimension, retaining key information while improving computational efficiency, and forming feature data extracted continuously at multiple scales by the neural convolutional network module.

[0033] The U-Net++ module focuses on fine-grained ECG segmentation, aiming to accurately segment key regions such as ventricular fibrillation waveforms. Its architecture consists of an encoder, a decoder, and dense skip connections, forming an improved nested U-shaped network. The encoder extracts input features layer by layer, using a series of convolutional layers to capture fine-grained features. Pooling gradually reduces the resolution, extracting layer-by-layer deep features from the input data to generate multi-scale feature maps. U-Net++ uses a dense skip connection structure to establish an interactive channel between the encoder and decoder for features at different scales, fusing features from different levels to form a nested U-shaped network. The decoder gradually restores spatial resolution through deconvolution operations, reshaping the multi-scale feature maps into a segmentation result of the same size as the input data. In this process, the decoder not only utilizes skip connections to fuse low-level detailed features with high-level semantic features, but also enhances the segmentation of key regions through context integration, providing high-quality input features for subsequent classification tasks.

[0034] To increase global dependencies, the AutoFocus Transformer is integrated to capture long-range dependencies and spatial contextual information between ECG features on a global scale. The AutoFocus Transformer model uses its adaptive focus attention mechanism and transformer architecture to perform efficient multi-scale feature extraction. The encoder part uses multi-layer self-attention modules to replace traditional convolutional layers, generating rich multi-scale feature representations by capturing global contextual information and local details. In addition, the addition of the AutoFocus mechanism further enhances the dynamic adaptability of the model, and can automatically adjust the focus to specific abnormal areas of the ECG to filter out redundant information. During the decoding process, the AutoFocus Transformer uses skip connections to combine low-level features with high-level features in the encoder, which can effectively extract significant distinguishable features that combine global and local information from complex high-dimensional signals, providing deep support for ECG classification and prediction, and further improving the performance and robustness of the model.

[0035] Boosting ensemble learning techniques are then applied to improve classification accuracy by combining multiple weak learners to construct a strong learner. By constructing a Boost algorithm model for classification based on ECG features, the overall model performance is gradually enhanced by iteratively training weak learners and adjusting sample weights at each step. Each new weak learner corrects the errors of the previous learner, gradually improving the model's ability to recognize abnormal waveforms. A greedy strategy-based combination method uses all the original features and cross-features used in the spanning tree with all the categorical features in the dataset to cross-capture the interactions between features, improving the model's expressiveness and generalization capabilities.

[0036] The automatic chest compression system is used to start automatic chest compression when it is determined that the patient has suffered cardiac arrest, identify the carotid pulse and continuously monitor blood pressure. Based on the identification of carotid artery fluctuations and pressure, the amplitude and frequency of chest compressions are adjusted through self-feedback.

[0037] like Figure 4 As shown in the figure, the intelligent automatic chest compression system consists of two parts. One part is the automatic identification of venous pulse to monitor changes in transcutaneous blood oxygen saturation and carotid artery pressure; the other part is the design of an automatic compression device, which can adjust the compression depth and frequency in time according to the carotid artery pulse pressure difference and blood oxygen saturation.

[0038] The intelligent automatic chest compression system comprises: Automatic identification of carotid pulse and continuous non-invasive blood pressure monitoring: The device is attached to the patient's neck and integrates an ultrasonic transducer, a pressure sensor, and a photoplethysmography (PPG) sensor to accurately identify the carotid pulse and continuously monitor blood pressure. The system first uses a pressure sensor to obtain weak mechanical fluctuations in the carotid artery area, and combines this with the periodic characteristics of blood volume changes in the PPG signal to construct an initial pulse recognition model. The ultrasonic transducer is then used to measure changes in arterial wall size and excite multi-frequency sound waves. When the arterial wall reaches a resonant state, its synchronized vibration frequency can be used to infer arterial wall tension, thereby calculating the real-time blood pressure in the blood vessels and achieving non-invasive and continuous blood pressure monitoring. To improve measurement accuracy, the system integrates a pressure sensor and uses Kalman filtering for denoising and a long short-term memory network (LSTM) to predict blood pressure trends, while also combining wavelet transforms for signal optimization. In addition, the device can monitor transcutaneous blood oxygen saturation to accurately assess ventilation effectiveness, and use time-frequency analysis and heart rate variability (HRV) calculations to assess the recovery of autonomous cardiac rhythm. It also uses principal component analysis (PCA) to extract key blood pressure features to improve computational efficiency and feedback speed, thereby achieving intelligent vital signs monitoring and resuscitation quality optimization.

[0039] Automatically adjust the mechanical compression device through feedback. The present invention proposes to design a cardiopulmonary resuscitation system and its components with a feedback device, which can evaluate the compression depth, compression frequency, chest wall pressure and interruption time, thereby improving the patient's spontaneous circulation recovery rate and long-term survival rate and reducing the occurrence of adverse events. It is planned to use real-time feedback from a flexible pressure sensor to measure the exact position and pressure of the compression, and transmit it to the information center based on the results of the compression. In order to simulate the pressure changes caused by manual chest compression, the pressure value and downward pressure amplitude of the manual chest compression are first collected to establish a mechanical model. The mechanical chest compression device is set according to the mechanical model to avoid adverse events caused by excessive pressure. At the same time, the effective fluctuations generated by chest compression can be sensed by a device attached to the carotid artery to automatically monitor blood pressure. At the same time, this information is transmitted to the information center for artificial intelligence analysis, and instructions are transmitted to the automatic chest compression device to adjust the compression amplitude and frequency according to the results.

[0040] The intelligent automatic chest compression system automatically completes chest compressions including: 3.2.1. Data Acquisition: The automated chest compression device is equipped with a flexible pressure sensor that can measure the anteroposterior diameter of the thorax and compression depth, pressure, and frequency in real time. An accelerometer is also used to monitor compression amplitude and speed. The device on the patient's neck integrates a photoplethysmography (PPG) sensor and infrared / near-infrared spectroscopy sensors to enable non-invasive blood pressure, pulse pressure difference, and pulse monitoring. Combined with a pressure sensor, it identifies the carotid artery pulse to provide continuous vital sign data.

[0041] 3.2.2. Constructing a feedback control system: This system uses a mechanical chest compression device based on vital sign data to dynamically adjust compression depth, speed, and force according to sensor data. A reinforcement learning algorithm (Deep Q-Network, DQN) is used to optimize the mechanical chest compression strategy. By collecting key vital signs such as MAP (mean arterial pressure) and pulse waveform amplitude as state inputs and reward basis, a closed-loop control model is constructed to optimize compression depth and frequency. Compression parameters are adjusted in real time using Lyapunov adaptive control to stabilize blood pressure within the target range. Fuzzy logic control is also used to adjust compression force based on heart rate, blood pressure, and chest wall recoil rate to ensure maximum blood perfusion and minimize injury risk. The user interface provides real-time data visualization, allowing medical staff to monitor patient status and manually adjust compression strategies through the web or mobile terminal.

[0042] 3.2.3. Parameter adjustment strategy: Set a safe blood pressure target range (MBP 50-60 mmHg), a pulse rate of 80-120 bpm, and an initial chest compression depth of 5 cm and a compression frequency of 100 times / min. If the blood pressure is below 50 mmHg, increase the compression depth by 0.5 cm. If it is still below 50 mmHg, increase the compression frequency by 10 times / min. If the compression amplitude reaches 6 cm and the compression frequency reaches 120 times / min, and the blood pressure is still not ideal, adjust the compression position. If it is still not ideal, a warning will be issued, and the operator needs to consider other factors that may cause low resuscitation blood pressure or other treatment options such as extracorporeal circulation. If blood pressure rises above 80 mmHg, the compression rate is reduced by 5 times / minute. If it remains above 80 mmHg, the compression amplitude is reduced by 0.5 cm. If it remains above 80 mmHg, the compression rate is further reduced by 5 times / minute or the compression amplitude is reduced by 0.5 cm. If the blood pressure remains above 80 mmHg after the compression amplitude reaches 3 cm, a 3-5 second pause is performed to determine whether spontaneous circulation has been restored based on blood pressure, pulse pressure difference, and pulse rate. If spontaneous circulation has been restored, chest compressions are discontinued. A self-feedback system continuously monitors system performance and adjusts controller parameters to optimize control effectiveness. Simultaneously, the system continuously collects key parameters during the compression process in the background to build a high-quality dataset. Based on this historical data, a recurrent neural network (RNN) is used to train a predictive model to predict the optimal compression depth and frequency under specific vital sign conditions. This model can assist in DQN policy decision-making and serve as a pre-training or policy correction module to improve the system's responsiveness and adaptability under complex physiological conditions.

[0043] 3.2.4. Control system: This control system uses reinforcement learning and a direct adaptive control algorithm based on Lyapunov stability, aiming to achieve dynamic adjustment and optimal control of compression parameters, so that key vital signs are stabilized within the set target range. The system first constructs an initial controller and sets reasonable control parameters; during operation, it collects state data including process variables (such as MAP), control variables (such as compression depth, frequency, and force), and control errors in real time. Based on the collected information, the system evaluates the current parameter control effect and updates the parameters through an adaptive adjustment algorithm derived from reinforcement learning and the Lyapunov criterion. At the same time, considering the control delay of the system, the controller uses the state and parameter information of the previous sampling period to calculate the control law, thereby effectively reducing the online computing burden. This process is continuously iterated in a closed-loop manner to ensure that the control parameters continuously adapt to changes in the patient's physiological state, achieving stable and efficient individualized compression control.

[0044] In summary, the embodiments of the present invention provide an artificial intelligence-based cardiopulmonary-cerebral resuscitation system, which can construct a deep learning model of cardiac arrest based on surface electrocardiogram data to accurately identify high-risk patients, and for patients whose hearts still cannot be revived after defibrillation, an intelligent chest compression resuscitation device can be used to continuously and automatically adjust the compression depth and frequency according to their pulse pressure difference, oxygen and other indicators, aiming to improve the success rate of cardiac arrest patients.

[0045] According to a second aspect of the present invention, there is provided an application method of an artificial intelligence-based cardiopulmonary resuscitation system, which is used in the artificial intelligence-based cardiopulmonary resuscitation system described above, the method comprising: Build a wearable, portable cardiac arrest early warning EMS system. This system collects physiological information through a convenient wearable device, processes the collected data, extracts features and performs classification inference on the processed data, and diagnoses cardiac arrest based on the physiological information. If so, it automatically calls 120 and sends the patient's location to the medical system. Build a cardiac arrest AI model and automatic defibrillation system based on surface ECG data. Collect the patient's ECG information, select the optimal wavelet basis and threshold suitable for the ECG signal characteristics to process the ECG data of ventricular fibrillation waveforms, and build an ECG ventricular fibrillation recognition model. Once ventricular fibrillation is diagnosed, a voice reminder is issued and defibrillation is performed. Build an intelligent automatic chest compression system that can accept feedback information regulation. When it is determined that the patient has suffered cardiac arrest, automatic chest compression is started, and the carotid pulse is accurately identified and blood pressure is continuously monitored. Based on the identification of carotid artery fluctuations and pressure, the amplitude and frequency of chest compressions are adjusted through self-feedback.

[0046] It can be understood that the artificial intelligence-based cardiopulmonary resuscitation method provided by the present invention corresponds to the artificial intelligence-based cardiopulmonary resuscitation system provided in the aforementioned embodiments. The relevant technical features of an artificial intelligence-based cardiopulmonary resuscitation method can refer to the relevant technical features of an artificial intelligence-based cardiopulmonary resuscitation system, which will not be repeated here.

[0047] Through the description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0048] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An artificial intelligence-based cardiopulmonary resuscitation system, characterized in that: include: The cardiac arrest early warning EMS system is used to collect physiological information through a portable wearable device, process the collected data, extract features and perform classification reasoning on the processed data, and diagnose whether the heart has suddenly stopped according to the physiological information. If so, it automatically calls 120 and sends the location to the medical system; The automatic defibrillation system is used to collect the patient's electrocardiogram information, process the electrocardiogram data of ventricular fibrillation waveforms by selecting the optimal wavelet basis and threshold suitable for the characteristics of the electrocardiogram signal, and build an electrocardiogram ventricular fibrillation recognition model. When ventricular fibrillation is diagnosed, a voice reminder is issued and defibrillation is performed; The automatic chest compression system is used to start automatic chest compression when it is determined that the patient has suffered cardiac arrest, identify the carotid pulse and continuously monitor blood pressure. Based on the identification of carotid artery fluctuations and pressure, the amplitude and frequency of chest compressions are adjusted through self-feedback.

2. The artificial intelligence-based cardiopulmonary resuscitation system according to claim 1, characterized in that: The collecting of physiological information by a portable wearable device and processing of the collected data include: Pulse and blood oxygen data are collected through the bracelet, marked and stored by timestamp. Through pattern recognition and trend analysis of historical data, the dynamic changes of patients' vital signs are captured, and the collected pulse and blood oxygen data are checked for integrity and consistency, missing values are supplemented, and abnormal values in the collection are eliminated.

3. The artificial intelligence-based cardiopulmonary resuscitation system according to claim 1, characterized in that: The feature extraction and classification reasoning of the processed data includes: Perform time series modeling on pulse and blood oxygen data and use the exponentially weighted moving average algorithm to smooth the data to ensure that the model can capture trend characteristics; During the feature extraction process, the first-order and second-order features of peak value, fluctuation amplitude and change rate are extracted based on the periodicity and amplitude characteristics of pulse and blood oxygen signals. At the same time, a time delay feature enhancement model is introduced to model the time dependency.

4. The artificial intelligence-based cardiopulmonary resuscitation system according to claim 1, characterized in that: When the cardiac arrest warning EMS system detects a pulse of ≤40 beats / minute and lasts for 10 seconds, a yellow warning is triggered and an extreme bradycardia warning is sent to close contacts; when the pulse is ≤30 beats / minute and lasts for 10 seconds, a red warning is triggered and the mobile phone automatically sends an emergency call to the EMS center and uses GPS to locate the user's location to ensure a quick response in an emergency.

5. The artificial intelligence-based cardiopulmonary resuscitation system according to claim 1, characterized in that: The automatic defibrillation system includes a portable ECG acquisition and defibrillation device, which adopts intelligent charging management, automatically maintains the charging state when connected to the central device, and immediately starts single-electrode ECG signal acquisition after disconnection.

6. The artificial intelligence-based cardiopulmonary resuscitation system according to claim 5, characterized in that: The portable ECG acquisition and defibrillator device transmits the collected ECG signals to the central device. After identifying the ventricular fibrillation ECG waveform, the central device will issue a voice reminder that defibrillation is about to begin, saying "Defibrillation is about to begin, please do not have physical contact with the patient!"; during the entire resuscitation process, the ECG acquisition and defibrillator device records the ECG waveform in real time and transmits it to the central device for long-term monitoring; if the ECG signal shows that the patient has recovered normal sinus rhythm, the central device will issue a voice reminder that "sinus rhythm has been restored" and call the patient again.

7. The artificial intelligence-based cardiopulmonary resuscitation system according to claim 1, characterized in that: The defibrillation process of the automatic defibrillator system includes: Processing ECG data containing ventricular fibrillation waveforms, standardizing data collected by portable wearable devices, removing abnormal samples, and constructing a data set through quality control. Comprehensive data cleaning is also performed. Wavelet transform is used to analyze and extract features from ECG signals, extracting key features from the signal by selecting the optimal wavelet basis and threshold that are suitable for the characteristics of the ECG signal. Convolutional neural networks and local attention image recognition are used to construct an ECG ventricular fibrillation recognition model, which involves processing the input ECG data through multiple convolutional layers of the convolutional neural network. Each convolutional layer is equipped with multiple filters to detect waveform features in different directions and scales. The pooling layer samples the feature map, reduces the feature dimension, retains key information, and constitutes the feature data extracted at continuous multi-scale under the neural convolutional network module.

8. The artificial intelligence-based cardiopulmonary resuscitation system according to claim 1, characterized in that: The automatic chest compression system has the function of automatically identifying venous pulses, monitoring changes in transcutaneous blood oxygen saturation and carotid artery pressure, and timely adjusting compression depth and compression frequency according to the carotid artery pulse pressure difference and blood oxygen saturation.

9. The artificial intelligence-based cardiopulmonary resuscitation system according to claim 8, characterized in that: The automatic chest compression system integrates an ultrasonic transducer, a pressure sensor, and a photoplethysmography sensor placed at the carotid artery to accurately identify the carotid pulse and continuously monitor blood pressure; it includes: A pressure sensor is used to capture the weak mechanical fluctuations in the carotid artery region. This is combined with the periodic characteristics of blood volume changes in the photoplethysmography sensor signal to construct an initial pulse recognition model. The ultrasonic transducer is used to measure the changes in arterial wall size and stimulate multi-frequency sound waves. When the arterial wall reaches a resonant state, the synchronized vibration frequency is used to infer the arterial wall tension, and then calculate the real-time blood pressure in the blood vessels.

10. An application method of an artificial intelligence-based cardiopulmonary resuscitation system, used in an artificial intelligence-based cardiopulmonary resuscitation system according to any one of claims 1 to 9, characterized in that: The method comprises: Build a wearable, portable cardiac arrest early warning EMS system. This system collects physiological information through a convenient wearable device, processes the collected data, extracts features and performs classification inference on the processed data, and diagnoses cardiac arrest based on the physiological information. If so, it automatically calls 120 and sends the patient's location to the medical system. Build a cardiac arrest AI model and automatic defibrillation system based on surface ECG data. Collect the patient's ECG information, select the optimal wavelet basis and threshold suitable for the ECG signal characteristics to process the ECG data of ventricular fibrillation waveforms, and build an ECG ventricular fibrillation recognition model. Once ventricular fibrillation is diagnosed, a voice reminder is issued and defibrillation is performed. Build an automatic chest compression system that can accept feedback information and control. When it is determined that the patient has suffered cardiac arrest, automatic chest compression is started, and the carotid pulse and continuous blood pressure monitoring are performed. Based on the identification of carotid artery fluctuations and pressure, the amplitude and frequency of chest compressions are adjusted through self-feedback.

Images