CN116019443B - Cardiopulmonary resuscitation chest compression compliance detection system and method - Google Patents

Abstract

The present invention discloses a multi-dimensional indicator compliance detection system and method for chest compression. The auxiliary system is composed of an equipment shell, a rubber band, a voice interaction module, an inertial measurement unit module, a microcontroller module, a communication module, a storage module, a touch and display module, and a power module; the auxiliary method includes a compression depth detection method, a compression frequency detection method, an arm compression verticality detection method, and an elbow bending detection method, so as to realize the evaluation of the compression depth, frequency, and arm and elbow posture normativity of the rescuer; the present invention uses a voice interaction method, and the rescuer triggers and guides the startup and working state of the auxiliary system through voice information, and the system guides the rescuer through voice information, thereby effectively improving the quality of chest compression and facilitating standardized, accurate, efficient, and intelligent chest compression medical behavior.

Description

Cardiopulmonary resuscitation chest compression compliance detection system and method

Technical Field

The present invention belongs to the intersecting fields of smart medical care, artificial intelligence, instrument science, computer science, sensor technology, human-computer interaction technology, etc. More specifically, it relates to a cardiopulmonary resuscitation chest compression compliance detection system and method.

Background technique

Cardiopulmonary Resuscitation (CPR) is a medical technology used to treat patients with cardiac arrest and respiratory arrest. It is the most important and basic first aid measure. Cardiopulmonary resuscitation includes three main aspects: chest compression, external ventilation, and electric defibrillation. Among them, chest compression runs through the entire process of cardiopulmonary resuscitation. When the heart stops beating, artificial chest compression helps the patient establish extracorporeal circulation, maintain cerebral blood oxygen, and promote blood return to the heart, helping the patient to restore autonomous circulation. Therefore, chest compression plays a very important role in cardiopulmonary resuscitation.

According to the AHA guidelines, the following requirements are made for standard chest compressions for adults: 1. The compression depth is 50-60 mm and the frequency is 100-120 times/minute; 2. After compression, the chest pressure should be fully released. In addition, the rescuer needs to keep the compression and rebound time balanced to avoid impact compression. When pressing, the arm should apply pressure vertically downward to the center of the line connecting the patient's sternum and breasts, avoid bending the elbow and tilting the arm to avoid kneading and pushing compressions;

Many researchers have realized that non-standard problems in the process of chest compression will affect the treatment effect of cardiopulmonary resuscitation, and have studied and developed chest compression auxiliary equipment. Invention patent CN114983794A designs a chest compression auxiliary equipment, which is in the shape of a flat structure with a built-in accelerometer and pressure sensor, which is used to obtain the compression depth and chest elastic coefficient during chest compression; when the structure is working, it needs to be placed between the patient's chest and the rescuer's palm; to a certain extent, it affects the rescuer's sense of reality of contact with the patient, and what is obtained is always the touch of the surface material of the compression auxiliary equipment. Invention patent CN114404262A designs a wearable cardiopulmonary resuscitation auxiliary equipment, which is similar to the structure of a watch. It uses mechanical transmission, gears, motors and proximity sensors as basic components to evaluate the depth and frequency of compression during chest compression. The mechanical structure transmission has a certain complexity. Therefore, in order to solve the problems that existing chest compression auxiliary equipment has few dimensions for monitoring the quality of chest compression medical behavior, the detection equipment interferes with or causes discomfort to rescuers, and is difficult to use clinically, it is necessary for technical personnel in this field to develop a wearable chest compression multi-dimensional indicator compliance detection system and method for rescuers.

Summary of the invention

In order to solve the above technical problems, the present invention proposes a cardiopulmonary resuscitation chest compression compliance detection system and method. The existing chest compression auxiliary equipment is placed on the patient's sternum, or clamped between the rescuer's left and right hands, or has a more complex mechanical structure, which cannot realize convenient participation in chest compression assistance, and there is a problem of interfering with the rescuer's tactile sense, causing discomfort, affecting the rescuer's judgment on the integrity of the ribs and sternum and the accurate force application position, and has the disadvantages of less detection information and poor convenient interactivity.

To achieve the above object, the technical solution adopted by the present invention is:

The present invention provides a cardiopulmonary resuscitation chest compression compliance detection system, which is composed of a central control processing module, an inertial measurement module, a voice interaction module, a touch display module, a power module, a communication module and a data storage module:

The central control processing module adopts a low-power embedded microprocessor, which is responsible for the logic control, data calculation, and algorithm deployment of the entire system, and performs data communication with the inertial measurement module, voice interaction module, touch display module, power module, communication module, and data storage module;

The inertial measurement module is an integrated unit composed of an accelerometer, a gyroscope, and a magnetometer, and can measure the three-axis inertial data of the system attachment, including acceleration, angular velocity, and attitude quaternion. The inertial measurement module itself has a data preprocessing function, and transmits the real-time collected inertial data to the central control processing module through a data transmission bus. The voice interaction module is composed of a voice interaction controller, a microphone, and a micro speaker;

The voice interaction controller deploys the required voice interaction instructions and voice models in the cardiopulmonary resuscitation scenario. The voice interaction module senses the voice instructions issued by the rescuer in the environment through the microphone, and gives the rescuer guidance and real-time feedback correction through the micro speaker.

The touch display module is composed of a micro touch screen, a touch drive and a display drive circuit. Before the start of chest compression, the auxiliary system is configured through the touch display module. After the completion of chest compression, the compression index detection statistics are viewed through the touch display module.

The power module is composed of a power conversion module, a battery and a charging module, and is responsible for the power supply and discharge of the entire system;

The communication module includes wired communication and wireless communication, which is used to transmit the node data to the network device or PC device; the data storage module is used to store the collected IMU data and chest compression process status data related information.

As a further improvement of the auxiliary system of the present invention: the auxiliary system is equipped with a shell and a rubber strap, the shell is made by 3D printing, injection molding, or mold processing, the shell leaves necessary human-computer interaction interfaces and interfaces to the outside, and the aforementioned modules are sealed and installed in the shell, and the rubber strap is used to connect the shell and the rescuer's wrist.

As a further improvement to the auxiliary system of the present invention: the low-power embedded microprocessor is a single-core or multi-core processor.

As a further improvement of the auxiliary system of the present invention: the touch display module cooperates with the voice interaction module to provide audio-visual feedback when applying chest compressions.

As a further improvement of the auxiliary system of the present invention: the data storage module selects a TF memory card as an external storage device.

The present invention provides a method for a cardiopulmonary resuscitation chest compression compliance detection system, and the specific steps are as follows:

S1: Data collection and segmentation of wristband IMU node. The microcontroller collects the data output by IMU through the UART communication interface, segments the data and adds it to the data window;

S2: Use the attitude quaternion to transform the acceleration of the load system;

S3: Select the acceleration in the vertical direction after coordinate transformation, perform trapezoidal numerical integration calculation, and then input the calculation result into a 0.3 Hz Butterworth high-pass filter to remove the trend term to obtain the velocity in the vertical direction;

S4: dividing the velocity in the vertical direction into velocity curve segments of the pressing stage and the rebound stage;

S5: Performing compression depth detection, performing trapezoidal numerical integration on the segmented compression and rebound stage curves to obtain the compression and rebound stage depths;

S6: According to the fixed sampling rate and the number of data points of the speed curve obtained by segmentation in S4, the time consumed by pressing and rebounding can be obtained;

S7: Install a linear displacement sensor in the chest cavity of the dummy model to collect the actual sternum depression distance caused by the compression of the sternum of the dummy;

S8: Obtain a continuous compression displacement curve, and identify the curve features based on peak detection and frequency domain analysis;

S9: According to the curve characteristics, the reference depth, frequency and time in the pressing and rebounding process are solved;

S10: Compare the solution values of S9, S6, and S5 with the reference values, and use machine learning methods to correct errors.

To obtain more accurate compression depth and frequency information;

S11: The pitch angle is used as a feature vector and input into the long short-term memory artificial neural network LSTM for sequence recognition. The arm compliance is detected by the arm pressing verticality detection method and the elbow bending detection method.

As a further improvement of the method of the present invention, the method of using the attitude quaternion in step S2 to perform coordinate transformation on the acceleration obtained under the load system is as follows:

For the acceleration collected, based on the attitude quaternion, the coordinate transformation to the three-axis acceleration in the geographic coordinate system is solved based on the following algorithm:

p _body =[0,a _bx ,a _by ,a _bz ] ^T

p _world =q·p _body ·q ^-1 =q·p _body ·q*/||q|| ²

Among them, a _bx, a _by , a _bz are the three-axis accelerations in the IMU carrier coordinate system respectively, p _body is a virtual quaternion that describes the acceleration in the carrier system, q is the attitude quaternion in the world coordinate system measured by the IMU, and p _world is the acceleration quaternion in the world coordinate system. Since it is also a virtual quaternion, taking its imaginary part can obtain the three-axis acceleration in the world coordinate system. The three axes are northeast, south, east, and north.

As a further improvement of the method of the present invention, the pressing depth detection method in the pressing stage of step S5 is as follows: based on the following formula:

Where a' is the offset after subtracting the gravity acceleration, a _noise is the wide-band measurement noise, A(t), V(t) and D(t) are the ideal acceleration, velocity and displacement caused by motion;

The velocity curve segments of the pressing stage and the rebound stage are again integrated in a small interval to obtain the motion displacement Distance of the pressing and rebounding process in each cycle. Since the sampling rate Sample and the length of the velocity sequence Length obtained by the above division are known, the time t and velocity v of the pressing and rebounding process and the average frequency f per minute are obtained:

t＝Length/Sample

v＝Distance/t

Where N is the cumulative number of compressions, _t1 is the time taken for the compression process, and _t2 is the time taken for the rebound process. A machine learning regression algorithm is introduced to correct the compression depth error. The linear regression model is used to correct the error between the compression depth solution value and the reference value. Considering a set of data sets {( _xi , _yi ), i = 1, 2, ..., n}, where _xi is the reference depth and _yi is the solution depth, the linear regression model is based on the following form:

y＝x ^T β+ε

Among them, ε～N(0,σ ² ), error variable σ ² and parameter vector β need to be determined based on the existing data set through training and iteration.

As a further improvement of the method of the present invention, the arm pressing verticality detection method and elbow bending detection method in step S11 are as follows:

In order to identify the compliance of the rescuer's arm and elbow when applying pressure, the IMU node is worn on the rescuer's wrist, the pitch angle is continuously observed, and the standard pressing data for a certain period of time is recorded, that is, the angle between the arm and the horizontal plane is between γ ₁ and γ ₂ , γ ₁ and γ ₂ are the thresholds for judging compliance; arm tilt pressing data, that is, the angle between the arm and the horizontal plane is between λ ₁ and λ ₂ , λ ₁ and λ ₂ are the thresholds for judging compliance; elbow bending pressing data, that is, the elbow is not locked during the pressing process, the elbow angle changes and the elbow exerts force. The above-mentioned data sets will be used as feature observations and model training data;

Use LSTM to classify the categories of the model training data sequence. LSTM can take the point set within a certain historical moment of the current sampling point as input and mine the category of the features of the sequence composed of these continuous point sets;

The LSTM neural network structure is composed of a sequence input layer, an lstm layer, a fully connected layer, a softmax layer and a classification layer; the input layer receives the input sequence as a feature vector. When the pitch angle is used as a feature vector, the number of input layers is set to 1, the number of neurons in the lstm layer is set, and the output mode is set to sequence output. Through the fully connected layer, the softmax layer and the classification layer, the category to which a sequence belongs is output;

Use the adam algorithm for training to calculate the exponential moving average of the gradient and the exponential moving average of the square of the gradient:

Among them, β ₁ and β ₂ are the gradient descent factors and mean square gradient descent factors, and Adam uses moving average to update the network parameters:

Before the model training process, adjust the number of neurons and the corresponding solver parameters in the above algorithm formula, observe the value of the MSE of the validation set and the MSE curve during the iteration process. When the MSE no longer converges, the recognition effect is optimal. MSE measures the same descriptive statistics between the predicted value and the true value label:

Where n is the number of eigenvectors, _yi and are the predicted value and the true value respectively.

Compared with the prior art, the present invention has the following beneficial effects:

1. Compared with the existing method of placing the auxiliary equipment directly on the patient's sternum or clamping it between the rescuer's left and right hands, the auxiliary system of the present invention is worn on the rescuer's wrist, which will not interfere with the rescuer's sense of touch or cause discomfort.

2. The proposed method improves the existing double integral method for calculating the compression depth, and introduces algorithm optimization to reduce the error in calculating the compression depth, so that the accuracy of the system can better meet the needs of measuring the compression depth of 50-60 mm;

3. In addition to realizing the monitoring of compression depth and frequency by mainstream chest compression auxiliary equipment, this system expands the dimension of perception of chest compression quality information by introducing machine learning in software methods without introducing hardware complexity, such as adding monitoring of the compliance of the rescuer's arm, elbow and upper body posture.

4. Voice interaction is introduced in the auxiliary system for chest compression scenarios. The rescuer controls the workflow of the auxiliary equipment through voice commands, and the auxiliary equipment feeds back the multi-dimensional indicators detected in real time to the rescuer through voice information. The algorithm of voice interaction is implemented on the wearable device side, without the need for network connection.

5. When using this system to assist chest compression, practice on a cardiopulmonary resuscitation dummy model based on the learning mode, so that the machine learning algorithm of this system can learn the compression habits of different rescuers, and learn based on the rescuer's personalized characteristics such as palm thickness and palm heel bending ability, so that the method of this system can be more generalized.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a schematic diagram of the hardware composition of the wristband-type auxiliary system;

Figure 2 is a schematic diagram of the overall structure of the wristband auxiliary system;

Figure 3 is a schematic diagram of a usage scenario;

Figure 4 shows the task and algorithm framework structure diagram;

Figure 5 is a characteristic curve diagram of the continuous output of the corresponding pitch angle when the arm elbow is pressed in a standard and irregular manner under one condition;

Figure 6 shows the consistency evaluation between the pressing depth reference standard and the depth solution value proposed in this patent;

Figure 7 is a flow chart of the auxiliary system working status.

Attachment Description

1-1. Central control processing module; 1-2. Inertial measurement module; 1-3. Voice interaction module; 1-4. Touch display module; 1-5. Power module; 1-6. Communication module; 1-7. Data storage module; 1-8. Shell and rubber strap; 2-1. Interactive touch screen; 2-2. Shell; 2-3. Integrated circuit board; 2-4. Rubber strap; 3-1. Patient or dummy model; 3-2. Rescuer; 3-3. Auxiliary equipment.

Detailed ways

The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

In the present application, the rescuer wears the device on the wrist and performs chest compressions on patients who need cardiopulmonary resuscitation. Although the patient may be lying on a slope, in most scenarios where chest compressions are required, the patient lies on a horizontal hard ground or on a bed with a hard backboard. Therefore, in an embodiment of the present application, it can be considered that the direction in which the patient's sternum is compressed and deformed is perpendicular to the horizontal plane. When the rescuer is performing chest compressions, the inertial information of the rescuer's hand movement and the forced movement of the patient's sternum is the same.

As mentioned above, most of the existing chest compression assist devices are placed directly on the patient's chest, or designed as mobile phone applications, and the phone is clamped between the rescuer's left and right hands. The former will introduce a hard shell to interfere with the rescuer's sense of touch, causing discomfort and affecting the rescuer's judgment of the integrity of the ribs and sternum and the accurate force application position; the latter, clamping the smartphone between the hands is also prone to accidental touches, interfering with the rescuer's pressure application method.

In order to solve the above-mentioned problems existing in the existing chest compression assisting devices, in the present application, a chest compression compliance detection system is provided.

Referring to Figure 1, the system is composed of a central control processing module 1-1, an inertial measurement module 1-2, a voice interaction module 1-3, a touch display module 1-4, a power module 1-5, a communication module 1-6, a data storage module 1-7, a shell and a rubber band 1-8. The central control processing module uses a low-power embedded microprocessor that can be single-core or multi-core, responsible for the logic control, data calculation, and algorithm deployment of the entire system, and communicates data with other modules such as the touch display module and the voice interaction module. The inertial measurement module is an integrated unit composed of an accelerometer, a gyroscope, and a magnetometer, which can measure the three-axis inertial data of the system attachment, including acceleration, angular velocity, and attitude quaternion. The inertial measurement module itself has a data preprocessing function, and transmits the real-time collected inertial data to the central control processing module through a data transmission bus. The voice interaction module is composed of a voice interaction controller, a microphone, and a micro speaker. The voice interaction controller can deploy the required voice interaction instructions and voice models in the cardiopulmonary resuscitation scenario. The voice interaction module senses the voice instructions issued by the rescuer in the environment through the microphone, and gives the rescuer guidance and real-time feedback correction through the micro speaker. The touch display module is composed of a micro touch screen, a touch drive and a display drive circuit. Before the start of chest compression, the auxiliary system can be configured through the touch display module. After the chest compression is completed, the compression index detection statistics can be viewed through the touch display module. Preferably, the touch display module can also cooperate with the voice interaction module to provide audio-visual feedback when applying chest compression. The ground power module is composed of a power conversion module, a battery and a charging module, which is responsible for the power supply and discharge of the entire system. The communication module includes wired communication and wireless communication, which is used to transmit the node data to the network device or PC device; the data storage module is used to store the necessary information such as the collected IMU data and the state data of the chest compression process, and an external storage device such as a TF memory card can be selected; the shell is made by 3D printing, injection molding, or mold processing, and the shell reserves the necessary human-computer interaction interface and interface to the outside, and the aforementioned modules are sealed and installed in the shell. The rubber strap is used to connect the shell and the rescuer's wrist.

Referring to Figure 2, the overall structure diagram of the wristband auxiliary system of the present application is described. The system as a whole is composed of an interactive touch screen 2-1, a housing 2-2, an integrated circuit board 2-3, and a rubber strap 2-4. Referring to Figure 3, the use scenario of the present embodiment is described. The patient or dummy model 3-1 lies flat on the horizontal ground, and the rescuer 3-2 wears the auxiliary device 3-3 proposed in the present application on the wrist. In clinical or training scenarios, chest compressions are performed without causing tactile or visual interference to the rescuer, and the rescuer does not need to change the original chest compression habits or general operating steps.

The embodiment of the present application also provides a method for detecting compliance of multi-dimensional indicators for chest compression. Referring to FIG4 , the framework structure of the algorithm (method) driven by the task is described, wherein the method is converted into a program through a programming language and then runs in the aforementioned auxiliary system. Since the algorithm needs to run in real time in a dual-core microcontroller to solve the target of the real-time collected data, the program needs to maintain a dynamically updated data window and perform calculations according to the following algorithm steps:

S1: The wristband IMU receives your data collection and segmentation. The microcontroller collects the data output by the IMU through the UART communication interface, segments the data and adds it to the data window.

S2: Use the attitude quaternion to transform the acceleration of the load system;

S3: Select the acceleration in the vertical direction after coordinate transformation, perform trapezoidal numerical integration calculation, and then input the calculation result into a 0.3Hz Butterworth high-pass filter to remove the trend term to obtain the velocity in the vertical direction;

S4: dividing the velocity in the vertical direction into velocity curve segments of the pressing stage and the rebound stage;

S5: performing trapezoidal numerical integration on the segmented curves of the compression and rebound stages to obtain the depths of the compression and rebound stages;

S6: According to the fixed sampling rate and the number of data points of the speed curve obtained by segmentation in S4, the time consumed by pressing and rebounding can be obtained;

S7: Install a linear displacement sensor in the chest cavity of the dummy model to collect the actual sternum depression distance caused by the compression of the sternum of the dummy;

S8: Obtain a continuous compression displacement curve, and identify the curve features based on peak detection and frequency domain analysis;

S9: Obtain the reference depth, frequency and time during compression and rebound;

S10: Compare the solution values of S9, S6, and S5 with the reference values, and use machine learning methods to correct errors.

To obtain more accurate information such as compression depth and frequency;

S11: The pitch angle is used as a feature vector and input into the LSTM network for sequence recognition to detect arm compliance.

The following is an analysis of the algorithm in necessary details and theoretical methods:

This embodiment takes into account that the device is worn on the rescuer's arm. During chest compression, the arm posture changes. As one embodiment, it is necessary to consider the situation that the rescuer's arm does not meet the standard vertical compression, and the posture of the device when worn is also uncertain. Therefore, it is necessary to solve the three-axis acceleration of the coordinate transformation to the geographic coordinate system based on the posture quaternion for the acceleration collected, based on the following algorithm:

p _body =[0,a _bx ,a _by ,a _bz ] ^T

p _world =q·p _body ·q ^-1 =q·p _body ·q*/||q|| ²

Among them, a _bx, a _by , a _bz are the three-axis accelerations in the IMU carrier coordinate system, and p _body is a virtual quaternion that describes the acceleration in the carrier system. q is the attitude quaternion in the world coordinate system measured by the IMU, and p _world is the acceleration quaternion in the world coordinate system. Since it is also a virtual quaternion, taking its imaginary part can get the three-axis acceleration in the world coordinate system. The three axes are northeast, northeast, and celestial.

In terms of solving the pressing depth, it is based on the following formula:

Where a' is the bias after subtracting gravity acceleration (caused by the incomplete calibration and low accuracy of low-cost IMU), a _noise is the wide-band measurement noise, and A(t), V(t) and D(t) are the ideal acceleration, velocity and displacement caused by motion.

First, we perform a trapezoidal integration on the Z-axis acceleration after coordinate transformation to obtain the velocity signal, and introduce a 4th-order 0.3Hz Butterworth high-pass filter to suppress the drift trend term. After integration, the broadband noise in the acceleration and the error introduced by the measurement accuracy will still be mostly retained in the velocity solution value v. When the integration is performed again, the error will continue to accumulate with the length of the acceleration sequence. After one integration and filtering, the numerical sign of the velocity changes due to the change in the direction of movement from pressing to rebounding. Therefore, we divide the obtained velocity curve into pressing and rebounding stages, and the obtained velocity sequence is also of unequal length. Perform trapezoidal numerical integration again in a small interval to solve the motion displacement of the pressing and rebounding process in each cycle. Since the sampling rate and the length of the velocity sequence obtained by the above division are known, the time and velocity of the pressing and rebounding process can be solved. Since there is an error between the solved value and the true value, in the existing method, a 1Hz high-pass filter is used to correct the solved pressing displacement error, which suppresses the low-frequency noise component in the displacement signal to a certain extent; in the above analysis, we pointed out that the solved error includes wide-band noise and the error introduced by the low precision of the low-cost IMU, which makes there is a certain error between the solved value and the true value obtained by analyzing the linear displacement sensor, but the two are positively correlated.

To this end, we introduce a machine learning regression algorithm to correct the error. As an optional implementation, a linear regression model can be used to correct the error between the compression depth solution value and the reference value. Consider a set of data sets {( _xi , _yi ), i = 1, 2, ..., n}, where _xi is the reference depth and _yi is the solution depth. The linear regression model is based on the following form:

y＝x ^T β+ε

Among them, ε～N(0,σ ² ), error variable σ ² and parameter vector β need to be determined based on the existing data set through training and iteration.

In this embodiment, in order to achieve the normative identification of the elbow and body posture of the pressing arm, we still focus the input features of the algorithm on the data of the wristband IMU. As mentioned above, when the rescuer applies chest compression, we believe that the chest of the rescuer and the rescued have the same inertial data. On the other hand, when the rescuer performs chest compression, some non-normative chest compression behaviors appear, such as elbow bending and pressing, arm tilting and pressing, and longitudinal pitching and pressing caused by kneeling far away. These non-normative pressing postures and actions will eventually be connected through the wrist and act on the patient's sternum through the base of the palm. Therefore, the information of different modes such as angular velocity, acceleration and posture angle obtained by the IMU sensor at the wrist can be used as an effective observation input to evaluate the normativeness of the rescuer's arm.

Referring to Figure 5, when the IMU node is worn on the rescuer's wrist, the pitch angle is continuously observed. In Figure 4, this section of the pitch angle curve is divided into three parts: segment 1, segment 2, and segment 3. Segment 1 describes the arm pressing vertically 30 times in a relatively standard posture, segment 2 describes the arm pressing 30 times with an inclined position, and segment 3 describes the elbow pressing 30 times. From the pitch angle characteristics of Figure 4, the human eye can easily distinguish the differences between these curves.

Therefore, in the method described in this embodiment, an effective algorithm needs to be designed to be used in real time to identify the standard and non-standard categories shown in the aforementioned pitch angle continuous curve.

As an optional implementation method for identifying the above-mentioned normativeness, a neural network is used to classify the categories of the sequence, and LSTM can be used to process time series classification, regression and prediction tasks. The LSTM network is a recursive neural network that can learn the long-term dependencies of the time steps of sequence data. In this embodiment, it is necessary to classify the category to which the pitch angle sampling sequence of fixed-length sampling belongs. Since the mapping of time value and value or value and category constructed by the general neural network cannot meet the classification of the analogy of the sampling sequence in this application, LSTM can take the point set within a certain historical moment of the current sampling point as input, and mine the category of the features of the sequence composed of these continuous point sets. During the model training process, the identification effect is optimized by adjusting the number of neurons and observing the residual performance of the validation set.

As a method for identifying the rescuer's arm compression and posture norm in this embodiment, preferably, in addition to the pitch angle as the algorithm input, the inertial information of other modes such as the acceleration and angular velocity of the horizontal orthogonal axis can be used as the input of the identification algorithm to enhance the robustness of the algorithm.

As mentioned above, the method is forwarded into program code through a programming language, compiled and downloaded to the wrist-worn device. All calculation, reasoning, storage, and communication tasks are completed independently in the wearable node, without the need for external computers to provide additional computing power support.

Referring to Figure 6, the error distribution between the solution value and the reference value of the compression depth after error correction is described. An experiment was designed to invite 30 participants to collect more than 4,000 compression experiments. After solving the compression depth and the reference depth through the aforementioned algorithm steps, the error correction model was trained by using the Gaussian process regression model through 5-fold cross validation to predict the compression depth after error correction. The consistency between the extreme value and the true value was displayed by the Bland-Altman diagram. The upper and lower limits of 95% consistency in Figure 6 were 2.9449mm and -2.9439mm respectively.

7 , a flowchart of the auxiliary system working when the auxiliary system of the present application is used in training or clinical scenarios is described:

S1: The rescuer configures the device in advance to put it in a standby state; the rescuer can set the system's feedback method, feedback frequency, monitoring indicators, learning mode or training mode and other related parameter indicators through the touch interaction module;

S2: The rescuer wakes up the device through voice commands. The wake-up command words can be customized according to the rescuer's habits. Preferably, the rescuer can wake up the device through multiple methods, such as one or more of touch, button and other methods.

S3: The device guides the rescuer to perform compressions at a standard compression speed through audio, breathing effects of screen brightness, and vibration information;

S4-1 to S4-4: After the rescuer completes the voice start, chest compressions are started, and the auxiliary system starts to work, counting the number of compressions, solving the compression depth and frequency; solving the compression and rebound depth and time consumption of each cycle, evaluating whether there is impact compression, and whether the compression and rebound time consumption are balanced; solving the standardization of the posture when applying pressure with the arm and elbow;

S5: Voice, image feedback and guidance correction. When executing steps S4-1 to S4-4, the auxiliary system feeds back necessary guidance information through the voice interaction module and the touch display screen;

S6: During the execution of steps S4-1 to S4-4, the auxiliary system will count the number of compressions. According to the AHA guidelines, compressions and ventilations are performed in a ratio of 30:2. Therefore, when the system counts 30 compressions, subsequent steps such as ventilation and defibrillation are guided through voice or graphical interfaces.

S7: When the rescuer triggers the system to start a new round of compression assistance monitoring through voice commands, otherwise, the current real-time assistance system is exited.

S8: Record historical data and provide comprehensive and multi-dimensional evaluation of the chest compression emergency treatment process.

S9: When the rescuer issues a voice command during the compression process, that is, the patient recovers spontaneous circulation, the system will exit real-time auxiliary monitoring and go to S8.

The above description is only a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent change made based on the technical essence of the present invention still falls within the scope of protection required by the present invention.

Claims (7)

1. The utility model provides a cardiopulmonary resuscitation chest presses compliance detecting system, comprises central control processing module (1-1), inertial measurement module (1-2), voice interaction module (1-3), touch-control display module (1-4), power module (1-5), communication module (1-6) and data storage module (1-7), its characterized in that:

the central control processing module (1-1) adopts a low-power consumption embedded microprocessor, is responsible for logic control, data operation and algorithm deployment of the whole system, and performs data communication with the inertia measuring module (1-2), the voice interaction module (1-3), the touch display module (1-4), the power module (1-5), the communication module (1-6) and the data storage module (1-7);

The inertial measurement module (1-2) is an integrated unit composed of an accelerometer, a gyroscope and a magnetometer, and can measure triaxial inertial data of a system attachment body, wherein the triaxial inertial data comprises acceleration, angular velocity and attitude quaternion, the inertial measurement module (1-2) has a data preprocessing function and transmits inertial data acquired in real time to the central control processing module (1-1) through a data transmission bus, and the voice interaction module (1-3) is composed of a voice interaction controller, a microphone and a micro loudspeaker;

the voice interaction controller deploys a voice interaction instruction and a voice model in a required cardiopulmonary resuscitation scene, and the voice interaction module perceives the voice instruction sent by a rescuer in the environment through a microphone and gives the rescuer guidance comments and real-time feedback correction through a micro-speaker;

The touch display module (1-4) is composed of a miniature touch screen and a touch driving and display driving circuit, an auxiliary system is configured through the touch display module (1-4) before chest compression begins, and after chest compression is completed, compression index detection statistical information is checked through the touch display module (1-4);

The power supply module (1-5) consists of a power supply conversion module, a battery and a charging module and is responsible for supplying and discharging of the whole system;

The communication module (1-6) comprises wired communication and wireless communication and is used for transmitting data of the nodes to network equipment or PC equipment; the data storage module is used for storing the IMU data for implementing acquisition and the related information of the chest compression process state data; the method for detecting the cardiopulmonary resuscitation chest compression compliance detection system comprises the following specific steps of:

S1: the data acquisition and segmentation of the wrist strap IMU node are carried out, and the microcontroller acquires the data output by the IMU through the UART communication interface, segments the data and adds the segmented data into a data window;

S2: carrying out coordinate transformation on the acceleration under the obtained carrier system by using the attitude quaternion;

S3: selecting acceleration in the vertical direction after coordinate transformation, performing trapezoidal value integral calculation, and then inputting a calculation result into a 0.3Hz Baud Wo Sigao pass filter to remove trend items, so as to obtain speed in the vertical direction;

s4: dividing the speed in the vertical direction into a speed curve section of a pressing stage and a rebound stage;

S5: detecting the pressing depth, and performing trapezoidal value integration on the segmented curve in the pressing and rebound stage to obtain the depth of the pressing and rebound stage;

s6: according to the fixed sampling rate and the number of data points of the speed curve obtained by segmentation in the step S4, the time consumed by pressing and rebound can be solved;

s7: a linear displacement sensor is arranged in the chest of the dummy model, and the sternum depression distance actually generated when the sternum of the dummy is pressed is acquired;

S8: obtaining a continuous pressing displacement curve, and carrying out curve characteristic identification based on peak detection and frequency domain analysis;

s9: according to the curve characteristics, solving the reference depth, frequency and time spent in the pressing and rebound processes;

S10: comparing the solution values of S9, S6 and S5 with a reference value, and performing error correction by using a machine learning method to obtain more accurate pressing depth and frequency information;

s11: inputting a longitudinal rocking angle as a characteristic vector into a long-short-term memory artificial neural network LSTM, performing sequence identification, and detecting arm compliance through an arm pressing verticality detection method and an elbow bending detection method;

The arm pressing perpendicularity detection method and the elbow bending detection method of the step S11 are as follows:

In order to identify the compliance of arms and elbows of a rescuer when the rescuer applies pressure, wearing an IMU node on the wrist of the rescuer, continuously observing a pitching angle, and recording standard pressing data for a certain period of time, wherein the included angle between the arms and the horizontal plane is a threshold value for judging whether the arms and the horizontal plane are in compliance or not; arm inclination pressing data, namely an included angle between the arm and the horizontal plane is between lambda ₁ and lambda ₂, and lambda ₁ and lambda ₂ are thresholds for judging whether compliance exists or not; elbow bending press data, i.e. the elbow remains unlocked during the press, elbow angle bending changes and the elbow applies force, the data set obtained above will be used as feature observation and model training data;

Classifying the categories of the model training data sequence by using LSTM, wherein the LSTM can take a point set in a certain moment of the current sampling point history as input, and mine the categories of the characteristics of the sequence formed by the continuous point sets;

The LSTM neural network structure consists of a sequence input layer, a LSTM layer, a full connection layer, a softmax layer and a classification layer; the input layer receives an input sequence as a feature vector, when the pitch angle is used as the feature vector, the number of the input layer is set to be 1, the lstm layer is set to be the number of hidden layer neurons, the output mode is set to be sequence output, and a section of class to which the sequence belongs is output through the full connection layer, the softmax layer and the classification layer;

Training was performed using adam algorithm, calculating an exponential moving average of gradient m _l, an exponential moving average of gradient square v _l:

where β ₁、β₂ is the gradient descent factor and the mean square gradient descent factor, m _l-1 is the exponential moving average of the gradient over the last time step, v _l-1 is the exponential moving average of the gradient square over the last time step, Is the gradient of the current time step;

Next, using m _l and v _l solved by the above equation, adam updates the network parameters using a moving average:

θ _l is the parameter of the current time step, α is the learning rate, ε is a constant;

Before the model training process, the number of neurons and corresponding solver parameters in the algorithm formula are adjusted, the numerical value of an MSE of a verification set and an MSE graph in an iterative process are observed, when the MSE is not converged any more, the identification effect is optimal, and the MSE measures the same descriptive statistics between a predicted value and a true value label:

Wherein n is the number of feature vectors, y _i and Respectively a predicted value and a true value.

2. The cardiopulmonary resuscitation chest compression compliance detection system of claim 1, wherein: the auxiliary system is matched with the shell and the rubber binding belt, the shell is formed by 3D printing, injection molding or die processing, a necessary man-machine interaction interface and an interface are reserved for the shell, the modules are installed in the shell in a sealing mode, and the rubber binding belt is used for connecting the shell and the wrist of a rescuer.

3. The cardiopulmonary resuscitation chest compression compliance detection system of claim 1, wherein: the low-power consumption embedded microprocessor is a single-core or multi-core processor.

4. The cardiopulmonary resuscitation chest compression compliance detection system of claim 1, wherein: the touch display module (1-4) is matched with the voice interaction module to perform audio-visual feedback when chest compression is applied.

5. The cardiopulmonary resuscitation chest compression compliance detection system of claim 1, wherein: the data storage module selects the TF memory card as external storage equipment.

6. The cardiopulmonary resuscitation chest compression compliance detection system of claim 1, wherein:

the method for transforming the coordinates of the acceleration under the obtained carrier system by using the attitude quaternion in the step S2 is as follows:

For the acceleration for implementing acquisition, based on the gesture quaternion, the triaxial acceleration of coordinate transformation to a geographic coordinate system is solved, and the method is based on the following algorithm:

p_body＝[0,a_bx,a_by,a_bz]^T

p_world＝q·p_body·q^–1＝q·p_body·q*/||q||²

Wherein a _bx,a_by,a_bz is the triaxial acceleration under the IMU carrier coordinate system, p _body is a virtual quaternion, the acceleration under the carrier system is described, q is the attitude quaternion under the world coordinate system obtained by IMU measurement, p _world is the acceleration quaternion under the world coordinate system, and because the acceleration quaternion is also a virtual quaternion, the triaxial acceleration under the world coordinate system can be obtained by taking the imaginary part of the acceleration quaternion, and the triaxial acceleration is northeast day.

7. The method of cardiopulmonary resuscitation chest compression compliance detection system according to claim 1, wherein:

The method for detecting the pressing depth in the pressing stage in the step S5 is as follows: based on the following formula:

Wherein a represents the acceleration of the IMU in the vertical direction under the geographic coordinate system, V represents the speed generated by a, a' is the offset after subtracting the gravitational acceleration, a _noise is the wide-band measurement noise, the acceleration, the speed and the displacement of the arm motion are respectively represented by A (t), V (t) and D (t), V _noise,d_noise represents the noise components in the speed and displacement analysis, C _d is a displacement constant term, and C _v is a speed constant term; and (3) carrying out trapezoidal numerical integration again in the small section of the speed curve section of the pressing stage and the rebound stage, and solving to obtain a motion displacement Distance of the pressing and rebound process in each period, wherein the sampling rate Sample and the speed sequence Length obtained by the segmentation are known, so that the time t and the speed v of the pressing and rebound process and the average frequency f of each minute are obtained by solving:

t＝Length/Sample

v＝Distance/t

Where N is the cumulative number of compressions, t ₁ is the time spent in the compression process, t ₂ is the time spent in the rebound process, and a machine learning regression algorithm is introduced to correct the compression depth error, a linear regression model is applied to correct the error between the compression depth solution and the reference value, considering a set of datasets { (x _i,y_i), i=1, 2, …, N }, where x _i is the reference depth, y _i is the solution depth, the linear regression model is based on the following form:

y＝x^Tβ+ε

Where ε N (0, σ ²), the error variable σ ² and the parameter vector β need to be based on existing datasets, their parameters are determined by training and iteration.