An patch-type 18-lead electrocardio multiparameter monitoring diagnostic instrument3,4,5,14,15(Hereafter referred to as Patch-type Monitor)features microprocessor, memory, power supply, electrocardio signal and respiratory acquisition, ECG human identification, inertial sensors, bluetooth and so on modules, reusable defibrillation protective dry electrode arrays14, encapsulation3,4,5 of flat silicone flexible material shell suitable for pasting to human skin, as shown by the exploded schematics (Fig. 1a, 1b, 1c, 1d, 1e, 1f, 1g, Extended Data Fig. 1). The Patch-type Monitor is enclosed in a shell of ultra-small, ultra-thin and ultra-light, waterproof flat silicone flexible material suitable for pasting to the human skin15. The Patch-type Monitor has the functions of traditional 18-lead electrocardiographs16,17, multiparameter monitoring equipments16,18, Holter (18-lead ambulatory electrocardiogram systems)16,19, exercise plate electrocardiograph16 and defibrillator16,20 in five in one, and solves the major defect of iatrogenic missed diagnosis of right ventricular and posterior wall myocardial ischemia and myocardial infarction that cannot be detected by the gold standard 12-lead electrocardiogram equipments1,21,22 (Fig. 3a, 3b), and also allows a complete one-time wireless romete examination of the patient's left ventricle, right ventricle, and posterior wall of the left ventricle from different angles during resting, ambulatory, and motion. For example, the user can wear the Patch-type Monitor during bathing, sports, work, study, travel, daily life, sleep and hospitalization, even while running a marathon, and their life is completely unrestricted 3,4,5(Fig. 2a).
The ECG human identity recognition and inertial sensors (accelerometer, gyroscope, and magnetometer) module of the monitor can authenticate the user's human identity recognition23,24,25,26,27,28,29,,30,31,32,33,34, and can also accurately and comprehensively monitor and diagnose the electrocardiogram and breathing of the human body in the state of lie still, sleeping, sitting, walking, running, sports, up and down the hill, falling down, ascending and descending by elevator etc3,35,36.
In order to ensure that the patch-type single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG multi-parameter monitoring diagnostics series products3,4,5 (hereafter uniformly referred to as the Patch-type Multi-lead Monitors, Fig. 1h,1i,1j, 1k, Extended Data Fig. 1, and Supplementary Discussion 1, and Supplementary Videos 1 and 2, and 3 and 4, and 5 and 6,and 7 and 8) developed and manufactured by Shenzhen yASUN Technology Co., LTD. (hereafter referred to as yASUN) can have the defibrillation protective and defibrillation intervention functions in seconds, we designed and developed a patch-type automatic extracorporeal defibrillation monitor(hereafter referred to as Patch-type Defibrillator. Figure 4a, 4b, 4c, 4d, Extended Data Fig. 4).
Because the PhotoPlethysmoGraphy (PPG) signal contains a variety of physiological and pathological information of the heart and vascular system, it has a good application prospect in the noninvasive detection of human blood pressure, blood oxygen and blood flow, brain oxygen and muscle oxygen, blood glucose, pulse frequency, microcirculation, vascular resistance, respiratory rate, respiratory parameters, etc. We developed an head-mounted patch-type noninvasive blood pressure, blood oxygen, forehead temperature and glucose monitor3,4,5,37,38,39,40,41,42,43,44,45,46,47,48(hereafter referred to as Head-mounted Monitor. Figure 5a,5b,5c,5d, left, Extended Data Fig. 5) based on PPG signal and metabolic heat and blood flow. The Head-mounted Monitor has the subversive advantage of being suitable for the rapid real-time dynamic measurement of noninvasive blood pressure, blood oxygen, frontal temperature and blood glucose of local tissues of various parts for the human body.
In order to ensure that yASUN’s the Patch-type Multi-lead Monitors, the Patch-type Defibrillators and the Head-mounted Monitors, and the multiparameter monitoring and diagnosis devices of third-party companies can realize the wireless remote, long-term continuous, real-time dynamic monitoring, diagnosis and defibrillation intervention functions, we have designed and developed an new-type wireless remote connected ecosystem9,10,11,12 (Fig. 2a, 2b) for Global Wide Area Network based on B/S architecture (Browser-Server Mode) : “Wireless Remote ECG Multi-parameter Real-time Cluster Monitoring and AI Diagnosis and Intervention Cloud Platform Software System”( Supplementary Discussion 2, and Supplementary Videos 9 and 10 and 11).
The Patch-type Multi-lead Monitors, the Patch-type Defibrillator and the Head-mounted Monitors are connected to the multiparameter detection APP software system49,50,51(hereafter referred to as APP. Supplementary Videos 9) of mobile terminal (including smart phone, tablet, notebook, etc.) through Bluetooth Low Energy (BLE), which is responsible for hardware control, user interaction, multiparameter real-time monitoring and ECG diagnostic analysis results display. The mobile terminal’s APP can remotely transmit multi-parameter data to the new-type wireless remote connected ecosystem for real-time monitoring and diagnosis analysis through WIFI, 4G or 5G wireless communication.
To solve the key technical bottleneck problem of Bluetooth Low Energy (BLE) transmission bandwidth not meeting the transmission bandwidth of 18-lead and 12 lead electrocardiogram signals, we innovatively propose a ECG signal compression, encryption, transmission, and lossless reconstruction algorithm and mechanism52,53,54,55,562,57,58,59. The algorithm and mechanism significantly reduces the storage and transmission costs of electrocardio data while ensuring its absolute security, also provides an extremely low-cost new approach for high-precision, high sampling frequency, and high-dimensional data analysis of medical big data.
We have developed a set of artificial intelligence detection and diagnosis models60,61,~,75,76 to automatically classify 28 common cardiovascular health and disease categories, such as including sinus rhythm, supraventricular arrhythmia, ventricular arrhythmia, atrioventricular junction arrhythmia, preexcitation syndrome, heart block, pacemaker electrocardiogram, atrioventricular hypertrophy, myocardial infarction, myocardial ischemia, sleep apnea syndrome, etc. Arrhythmia types included cardiac arrest, sinus arrest, sinus tachycardia, sinus bradycardia, sinus arrhythmia, atrial premature beat, preatrial systolic bigeminal rhythm, atrial extrasystoles triad, atrial premature pairs, paroxysmal supraventricular tachycardia, atrial tachycardia, atrial flutter, atrial fibrillation, ventricular extrasystoles, ventricular extrasystoles bigeminy, ventricular extrasystoles triad Premature ventricular contraction pairs, paroxysmal ventricular tachycardia, VT, ventricular flutter, ventricular fibrillation, using patients' ECG data collected from the Patch-type Multi-lead Monitors, the Patch-type Defibrillators. We also developed non-invasive blood pressure, non-invasive blood oxygen, oxygen saturation, respiratory rate, body temperature and non-invasive blood glucose detection algorithms37,38,39,40,41,42,43,44,45,46,47,48. These technologies can realize wireless remote, long-term continuous real-time dynamic wearable cardiac function monitoring, diagnosis and defibrillation intervention, greatly improving the accuracy of ECG and multiparameter monitoring and diagnosis in various clinical application scenarios.
9 electrodes 18-lead synchronous electrocardiogram system and characterizations
The most commonly used traditional gold standard 12-lead electrocardiographs16,17, electrocardio multiparameter monitors16,18, ambulatory electrocardiogram systems (Holter)16,19, exercise tablet electrocardiographs16, and defibrillators16,20 have the following pain points:
-
These five types of devices are bulky and do not have wearable wireless remote long-term continuous, real-time dynamic cardiac function monitoring, diagnosis, and defibrillation intervention functions.
-
Although all five devices use a 12-lead electrocardiogram as the gold standard for routine examinations, these gold standard 12-lead devices all have the important defect of iatrogenic missed diagnosis that cannot detect right ventricular and posterior wall myocardial ischemia and myocardial infarction (Fig. 3a, 3b).
-
Each type of device can only provide services for specific single clinical application scenarios and medical service projects, and electrocardiogram machines can only perform short-term routine resting electrocardiogram examinations; The electrocardio multiparameter monitor can only perform resting multparameter monitoring; Holter and exercise tablet electrocardiograph can only perform resting, dynamic, and exercise electrocardiogram examinations, but do not have defibrillation protection or intervention functions.
In order to solve the pain points mentioned above for these five types of devices, we have designed and developed a wearable patch-type 9-electrode 18-lead ECG multiparameter monitoring diagnosis instrument3,4,5,14,15(Fig. 1a, 1b, 1c, 1d, 1e, 1f, 1g, Extended Data Fig. 1).
The Patch-type Monitor encapsulates modules such as microprocessors, memory, power supply, electrocardiosignal and respiratory acquisition, electrocardiogram human identity recognition, inertial sensors, Bluetooth, etc. in an ultra small, ultra-thin, and ultra-light waterproof flat silicone flexible material shell suitable for pasting to human skin, and has 2 reusable electrodes LL(Left Leg) and RL(Right Leg) on the host, 7 lead wire reusable electrodes RA (Right Arm), LA (Left Arm), and chest electrodes V5R, V1, V2, V5, V9 drawn forth from the host, which constitute a reusable 9 electrodes 18-lead synchronous ECG system1,3,4,5 (Fig. 1e, Extended Data Fig. 1 and Supplementary Video 9 and 10), and has the function of respiratory rate detection.
The Patch-type Monitor uses seven ECG signal acquisition channels CH1, CH2, CH3, CH4, CH5, CH6, CH7 in the ECG and respiratory acquisition module, as well as two electrodes LL (Left Leg) and RL (Right Leg) on the host, and seven lead wire electrodes RA (Right Arm), LA (Left Arm), and chest electrodes V5R, V1, V2, V5, V9 drawn forth from the host to form a 9 electrodes 18-lead synchronous electrocardiogram system, and collects bipolar limb lead I electrocardiogram data by connecting channel CH1, electrode LA, and electrode RA; collects bipolar limb lead II electrocardiogram data by connecting channel CH2, electrode LL, and electrode RA; collects unipolar chest lead V5R electrocardiogram data by connecting channel CH3, chest electrode V5R, and electrode WCT (Wilson Central Terminal); collects unipolar chest lead V1 electrocardiogram data by connecting channel CH4, chest electrode V1, and electrode WCT; collects unipolar chest lead V2 electrocardiogram data by connecting channel CH5, chest electrode V2, and electrode WCT; collects unipolar chest lead V5 electrocardiogram data by connecting channel CH6, chest electrode V5, and electrode WCT; collects unipolar chest lead V9 electrocardiogram data by connecting channel CH7, back electrode V9, and electrode WCT.
The limb leads are designed in Mason-Likar lead system, which can be derived
Bipolar limb lead Ⅲ = Ⅱ - Ⅰ,
Pressure lead aVR = - (Ⅰ + Ⅱ)/2,
Pressure lead aVL = Ⅰ- Ⅱ/2,
Pressure lead aVF = Ⅱ- Ⅰ/2.
Wilson unipolar lead system is used to design the chest lead. In Wilson unipolar lead system, V3, V4, V6, V7, V8, V3R and V4R are redundant leads, which can be reconstructed by unipolar lead system, using the ECG data of V1, V2, V5, V9, V5R and aVF leads. The ECG data of 7 leads V3, V4, V6, V7, V8, V3R, V4R can be deduced to form synchronize 12 chest leads. Therefore, an 18-lead synchronous electrocardiogram system with fewer 9 electrodes was derived1,21,22. We selecte V1, V2, V5, V5R, V9 and aVF leads as the initial leads of reconstruction, and deduce a complete 18-lead synchronous electrocardiogram system. The initial lead group S can be expressed as
S =(V 1 V 2 V 5 V 5R V 9 aVF).
The reconstructed chest lead unipolar system can be expressed as
E =(V 1 V 2 V 3 V 4 V 5 V 6 V 7 V 8 V 9 V 3R V 4R V 5R ),
with n sets of ECG samples where E is the real chest lead.
Suppose the derived chest lead is\(\hat {E}\), the linear model of \(\hat {E}\)and S can be expressed as
where, M is the conversion matrix and S is the reconstructed initial lead matrix. The conversion matrix M can be derived by minimizing the square error of and by the optimization program.
In this work, 140 volunteers were recruited for experiments. The volunteers were told the detailed procedure and potential risks. The experiment was conducted after volunteers signed an informed consent form and was approved by the Biomedical Ethics Committee of Shenzhen yASUN Technology Co LTD. The 30-minute 18-lead electrocardiogram 140 samples were collected using a traditional desktop 16-electrode 18-lead ECG machine, and a total of 168,762 heartbeats were used as the training set data. The 30-minute 18-lead electrocardiogram 42 samples were collected using yASUN patch-type 9-electrode 18-lead ECG multi-parameter monitor, with a total of 51,468 heartbeats as the test set data.
In accordance with the above method, a derived M can be expressed as
|
V1
|
V2
|
V3
|
V4
|
V5
|
V6
|
V1
|
1
|
0
|
0.251
|
0.235
|
0
|
-0.170
|
V2
|
0
|
1
|
0.247
|
0.149
|
0
|
0.047
|
V9
|
0
|
0
|
-0.235
|
0.064
|
0
|
0.166
|
V5
|
0
|
0
|
0.347
|
0.611
|
1
|
0.228
|
V5R
|
0
|
0
|
-0.044
|
0.103
|
0
|
-0.058
|
aVF
|
0
|
0
|
0.277
|
0.074
|
0
|
0.646
|
|
V7
|
V8
|
V9
|
V3R
|
V4R
|
V5R
|
V1
|
0.141
|
-0.553
|
0
|
0.166
|
0.556
|
0
|
V2
|
-0.122
|
0.054
|
0
|
0.119
|
-0.274
|
0
|
V9
|
-0.505
|
-0.008
|
1
|
0.543
|
-0.015
|
0
|
V5
|
-0.200
|
0.488
|
0
|
0.189
|
-0.240
|
0
|
V5R
|
0.448
|
0.191
|
0
|
0.181
|
0.103
|
1
|
aVF
|
0.722
|
0.494
|
0
|
-0.098
|
-0.051
|
0
|
In the table: The horizontal lead represents the lead derived from the reconstructed initial lead (longitudinal lead), where, the unit vector represents the actual collection results without participating in derivation, and the remaining numbers are reserved to 3 decimal places.
The errors of the 18-lead ECG (Fig. 3c) actually collected by the traditional desktop 16-electrode 18-lead ECG machine and the 18-lead ECG (Fig. 3d) derived by yASUN’s patch-type 9-electrode 18-lead ECG multi-parameter monitor can be calculated using Euclidean-distance formula:
The Euclidean distance is calculated using 500 data points corresponding to each other, and the obtained value is the error rate. The error rates (Fig. 3e) of ECG data for V3, V4, V6, V7, V8, V3R, and V4R leads derived by the above methods are:
|
V3
|
V4
|
V6
|
V7
|
V8
|
V3R
|
V4R
|
Error Rate
|
0.000631
|
0.000525
|
0.000179
|
0.000281
|
0.000269
|
0.00103
|
0.000328
|
Therefore, the 18-lead ECG derived from yASUN’s patch-type 9-electrode 18-lead ECG multi-parameter monitor has good accuracy. Meet the requirements of the national standards of the People's Republic of China and the IEC standards for ECG machine16,17, ECG multi-parameter monitor16,18, dynamic electrocardiograph (Holter)16,19, exercise plate electrocardiograph16 and external automatic defibrillation monitor16,20.
We also tried to reduce the number of reconstructed initial leads to three leads, V2, V9, and aVF, which were selected from completely independent directions. The whole 18-lead synchronous electrocardiogram system derived in this way requires only six electrodes V2, V9, LL(left leg), RL(right leg), RA (right arm) and LA (left arm). In practical tests, the error of the 18-lead synchronous ECG system with 6-electrode reconstruction is of the same order of magnitude as that of the 18-lead synchronous ECG system with 9-electrode reconstruction. A transformation matrix M trained by this method can be expressed as
|
V1
|
V2
|
V3
|
V4
|
V5
|
V6
|
V2
|
0.326
|
1
|
0.326
|
0.216
|
0.060
|
-0.016
|
V9
|
-0.148
|
0
|
-0.574
|
-0.398
|
-0.419
|
-0.045
|
aVF
|
0.791
|
0
|
0.857
|
0.947
|
0.998
|
0.758
|
|
V7
|
V8
|
V9
|
V3R
|
V4R
|
V5R
|
V2
|
-0.071
|
-0.0788
|
0
|
0.039
|
-0.105
|
-0.103
|
V9
|
-0.072
|
0.0907
|
1
|
-0.177
|
0.069
|
0.133
|
aVF
|
0.669
|
0.5343
|
0
|
0.445
|
0.169
|
0.257
|
In the table: The horizontal lead represents the lead derived from the reconstructed initial lead (longitudinal lead), where, the unit vector represents the actual collection results without participating in derivation, and the remaining numbers are reserved to 3 decimal places.
An patch-type automatic extracorporeal defibrillation monitor
The patch-type automatic extracorporeal defibrillation monitor (hereafter referred to as Patch-type Defibrillator) features a main control chip (including bluetooth transceiver unit), rechargeable battery unit, chest impedance detection unit, ECG detection unit, high-voltage charging unit, high-voltage monitoring unit, discharge unit, self-discharge unit, current monitoring unit, reusable defibrillation protective and defibrillation discharge dry electrodes. And an encapsulation of a flat silicone flexible material shell suitable for pasting onto human skin3,4,5,14,15, as shown in the breakdown diagram (Fig. 4a, 4b, 4c, 4d, Extended Data Fig. 4).
The Patch-type Defibrillator is mainly divided into the main control board (Fig. 4f, Extended Data Fig. 4) and the defibrillation board (Fig. 4e, Extended Data Fig. 4). The main control board and the defibrillation board are separated to isolate the high voltage circuit and the low voltage circuit. The main control board is the low voltage circuit, and the defibrillation board is the high voltage circuit. The thoracic impedance detection unit and the ECG detection unit on the main control board detect the patient's thoracic impedance and ECG signal through the same pair of defibrillation electrodes, and the main control chip (including bluetooth transceiver unit) is mainly responsible for collecting ECG and thoracic impedance data, processing data and controlling circuit operation, and transferring data via Bluetooth to mobile devices (including smartphones, tablets and laptops). The defibrillation board consists of two parts: charging and discharging. The charging part is mainly composed of an high-voltage charging unit, an high-voltage energy storage capacitor and an high-voltage monitoring unit. Its main function is to realize the fast charging of the defibrillator. The discharge part mainly consists of self-discharge unit, H-bridge discharge unit and current monitoring unit. Its main function is to realize the dual-phase wave discharge of the defibrillator. Compared with the traditional defibrillator, the current monitoring unit is added to the hardware of the patch-type automatic extracorporeal defibrillation monitor in this design. Its main function is to monitor the defibrillation current of the defibrillator, and at the same time, the high voltage monitoring unit is used to calculate the patient's chest impedance during shock defibrillation.
The software system of the defibrillator is mainly composed of functional modules such as system initialization, system self-test, circuit driving and control, data acquisition and data processing, and data transmission (Fig. 4g, Extended Data Fig. 4), and each functional module is composed of sub-modules. Each sub-module is a task, and the defibrillator needs to handle multiple tasks simultaneously when it is running. Therefore, the embedded real-time multitask operating system Zephyr RTOS was introduced in the design of the software system. The embedded real-time multitask operating system can dispatch each sub-task in the software system of the defibrillator in real time, reliably and accurately, so that the software system of the defibrillator can run smoothly and complete shock defibrillation.
The working flow of this defibrillator system (Fig. 4h, Extended Data Fig. 4) is as follows: If VF or VT is detected by one of the monitors of yASUN's patch-type single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG multiparameter monitoring and diagnostic instrument series products, the user should wear this defibrillator immediately. After the system is powered on and started up, the system is initialized firstly, and each function module is initialized to prepare for the self-test of the system. The system self-test mainly checks whether the ECG acquisition unit, chest impedance detection unit, high-voltage charging unit and discharge unit can work normally. If the user does not wear the yASUN’s monitor, the system self-test mainly checks whether the ECG acquisition unit, chest impedance detection unit, high voltage charging unit and discharge unit can work normally. When the system self-test passes, it checks whether the ECG and defibrillation electrodes are properly attached. It mainly detects whether the ECG electrodes fall off and whether the chest impedance is within the normal range of 20–200 Ω through the ECG acquisition unit and the chest impedance detection unit. After the ECG and defibrillation electrodes are attached, the rhythm analysis of the user's ECG data is firstly carried out by VF and VT algorithms. If VF and VT occur, the energy required for defibrillation is calculated, and automatic charging is prepared. The system immediately detects the patient's thoracic impedance, and the thoracic impedance at this time is the patient's thoracic impedance before shock defibrillation. The defibrillation current was calculated according to the set defibrillation energy and the patient's chest impedance, then the charging voltage was calculated according to the patient's chest impedance and the defibrillation current, and then the defibrillation pulse width was calculated according to the charging voltage, the patient's chest impedance and the set defibrillation energy. After these parameters are calculated, the system starts to charge the high voltage energy storage capacitor. After the charging is completed, the system is in discharge preparation state. Wait for discharge instructions from the system; When the discharge instruction is given, the system starts to discharge automatically. At the same time, the high voltage monitoring unit and the current monitoring unit are controlled to collect the voltage and current during shock defibrillation, and the defibrillation voltage at the beginning of discharge is divided by the defibrillation current to calculate the thoracic impedance of the patient. The thoracic impedance at this time is the thoracic impedance of the patient during shock defibrillation. The system calculates the defibrillation pulse width again according to the patient's chest impedance during shock defibrillation, and fine-adjusts the calculated defibrillation pulse width before shock defibrillation, so that the energy released by defibrillation can be more accurately controlled. When the discharge time reaches defibrillation pulse width, the discharge is complete.
Compared with the traditional defibrillator system, the main control process of this defibrillator system has three advantages. The first is that the calculation method of charging voltage is different. The charging voltage of the traditional defibrillator is calculated according to the set defibrillation energy, while the charging voltage of this system is calculated according to the set defibrillation energy and the patient's chest impedance detected before charging. Second, the discharge process is different. The combination of the current detection unit and the high voltage detection unit designed in this system can calculate the thoracic impedance of the patient during shock defibrillation during the discharge process, and fine-adjust the pulse width of defibrillation again according to the thoracic impedance during shock defibrillation. The third is to enable yASUN's patch-type single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG multiparameter monitoring and diagnostic instrument series products to cooperate with this defibrillator to have defibrillation intervention function in seconds.
Versatility of the multi-lead ECG multiparameter monitors
At present, when medical institutions at all levels provide users with physiological electrocardio multiparameter detection and early diagnosis and screening of cardiovascular health and diseases, the biggest pain point lies in the large number of users and low work efficiency. When a large number of users receive early diagnosis and screening, the traditional electrocardio multiparameter detection equipment needs users to queue up successively for detection. The detection time of cluster users is very long, and the gold standard 12-lead ECG detection data is only ten seconds. In fact, the longer the recording time of electrocardio multiparameter detection, the higher the detection rate of early cardiovascular disease diagnosis screening; When conducting 24-hour dynamic electrocardio multiparameter real-time monitoring for users in medical institutions at all levels, the pain point is that the existing electrocardio multiparameter acquisition equipment only has an single-parameter function, and the device is too large to be monitored and worn for a long time, and cannot provide services for all clinical application scenarios and medical service items. There is still short of inexpensive micro-wearable, wireless remote data transmission, long-term waterproof wearing and monitoring, and large data storage capacity, multifunction electrocardio multiparameter dynamic real-time monitoring medical instruments that are suitable for all clinical application scenarios and medical service items. These pain points have been broken through and solved in the series products of patch-type single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG multiparameter monitoring diagnosis instruments and their ecosystem developed by yASUN.
The international commercial five major categories of traditional electrocardiogram multiparameter monitoring and diagnostic equipment (multi-lead electrocardiogram machines, electrocardiogram multiparameter monitors, multi-lead dynamic electrocardiogram systems (Holter), exercise electrocardiographs, and automatic extracorporeal defibrillators) are planned, manufactured, and marketed according to five separate production lines. This is because the national standards of the People's Republic of China and the IEC standards for ECG machine16,17, ECG multi-parameter monitor16,18, dynamic electrocardiograph (Holter)16,19, exercise plate electrocardiograph16 and external automatic defibrillation monitor16,20 make special requirements for the basic safety and basic performance of each of the five major categories of medical electrical equipment. According to the basic safety and basic performance requirements of each of these five categories of medical electrical equipment, we propose a design and development method to integrate the basic safety and basic performance of these five categories of medical electrical equipment into one device.The device realizes the function of traditional 18-lead electrocardiogram machine, multi-parameter monitor, Holter (18-lead Holter recorder), motion electrocardiogram machine and automatic external defibrillation monitor, and thier application can cover all clinical application scenarios and medical service items13 of the heart and cardiovascular system.
To ensure that users of the defibrillator are not injured or burned by high-voltage electricity when using the monitor simultaneously, we have specially packaged a defibrillation effect protection circuit on 2 electrode LL and RL of the host, and on the button electrode base at the end of each lead wire led from the host (Fig. 1d). All yASUN’s single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG multiparameter monitoring diagnostic instrument series products3,4,5,14,15 have the functions of defibrillation protection and defibrillation intervention (Fig. 4g, 4h).
In recent years, biometrics technology has been widely studied and applied. Common biological features such as retina, iris, and palm lines are highly unique, but require expensive specialized equipment. Fingerprints, faces, and voices are also highly unique, but because these features are easily obtained, they can be easily faked. Many studies have confirmed that electrocardiogram is feasible as a biometric identification method23,24,25,26,27,28,29,,30,31,32,33,34. Using ECG signal to identify human individual has its unique advantages : (1) ECG signal is the only living body biological signal inherent in human body, which is not easy to be obtained by others and difficult to forge; (2) Since ECG signals are one-dimensional signals with low computing and storage overhead, relevant algorithms can be deployed in mobile devices such as wearable devices, smart phones, smartwatches and wristbands; (3) Wearable device can collect ECG, which is convenient and fast.
We propose an ECG-based personal recognition using a convolutional neural network23, a biometric identification method for ECG based on wavelet transform and piecewise correction24, a deep learning feature representation for electrocardiogram identification25,26, and a practical human authentication method based on piecewise corrected Electrocardiogram27, and a ECG identification based on neural networks28. The identification accuracy of these ECG identification methods has reached 97.7%, 98.5%, 98.49%, 96.6% and 99.6% respectively.
The ECG multiparameter monitoring diagnostic instrument series products produced by us can automatically verify and identify the user's personal identity through the ECG human identity identification module23,24,25,26,27,28,29,,30,31,32,33,34, realizing an ECG real name identity authentication solution for wireless remote multiparameter cluster monitoring and real-time diagnosis and intervention of multi-lead electrocardiogram in the cloud platform ecosystem, providing functions such as ECG living body detection and identity certificate ECG authentication. It can directly connect to authoritative data sources such as public security, operators, bank cards, and provide a complete set of integrated and maintenance solutions such as APP, H5, cloud services. It can be widely applied in scenarios such as finance, insurance, healthcare, government affairs, etc., ensuring the business operation of thousands of enterprises.
In order to solve the pain point problem of traditional ambulatory electrocardiogram systems (Holters), which only records and comprehensively analyzes the user's electrocardiogram during a certain period of time, but cannot accurately monitor and diagnose the user's electrocardiogram in what state of human posture, through the Patch-type Monitor's nine-axis inertial sensor (including three-axis accelerometer, three-axis gyroscope and three-axis magnetometer) measurement unit can automatically and quickly obtain the user's body in different posture (including Walking Forward, Walking Left, Walking Right, Walking Upstairs,Walking Downstairs, Running Forward, Jumping Up, Sitting, Standing, Sleeping, Elevator Up, and Elevator Down)3,35,36, which has very important medical statistics and clinical significance. It provides a very effective means for human individual cardiovascular health and disease monitoring and diagnosing.
Correctly identifying user’s activities is very significant when using the monitor simultaneously. Almost all feature extraction methods based on inertial sensors are directly based on acceleration and angular velocity. However, we studied user’s 12 activities (including Walking Forward, Walking Left, Walking Right, Walking Upstairs,Walking Downstairs, Running Forward, Jumping Up, Sitting, Standing, Sleeping, Elevator Up, and Elevator Down) and found that some activities (e.g., Elevator Up and Elevator Down) have no difference in acceleration and angular velocity35. Terefore, we believe that for these activities, any feature extraction method based on acceleration and angular velocity is difficult to achieve good results. After analyzing the difference of these indistinguishable movements, we propose several new features to improve accuracy of recognition. we have proposed four features on each axis of the accelerometer that have signifcantly improved the distinction between the two types of movements, elevator up, and elevator down. In the experiment, we found that the combination of frequency-domain features and time-domain features does not signifcantly improve the distinction of activities. The two kinds of features are two diferent aspects of acceleration and angular velocity, and there is no essential diference. From the experimental results, the time-domain features are better than the frequency-domain features and can more fully refect the diferences between diferent activities. Our custom features are not another response to acceleration,but instead, these features can be used to distinguish movements that difer in the speed of movement. In particular, it is of great signifcance to distinguish between movements that do not have a signifcant diference in acceleration and angular velocity but have a signifcant diference in speed. In the experiment, for the sake of convenience, we assumed that the component of the gravitational acceleration remains unchanged, which is obviously not in line with the actual situation. Next, related personnel may consider introducing some basic theories of motion analysis in order to accurately calculate the components of the gravitational acceleration and thus more accurately calculate the features we introduce. We believe that when the characteristics of velocity and displacement are introduced, we can make great breakthroughs in the existing problems of human motion classification, significantly improve the accuracy of activities without acceleration and angular velocity differences, and to a certain extent get rid of some artificial dilemmas of features that cannot accurately identify kinematic acceleration and angular velocity.
We apply the limit gradient Enhancement algorithm (XGBoost) to the human motion classification of the monitor’s users36. In this study, we compared the performance of XGBoost with other machine learning methods such as Support Vector Machines (SVM), Naive Bayes (NB), k-Nearest Neighbor (k-NN), and Random Forest (RF). The experimental results show that XGBoost achieves good results in user activity classification based on the inertial sensor of the monitor.
We capture the user's activity signals using the monitor's nine-axis inertial sensor measurement unit, which integrates a three-axis accelerometer, three-axis gyroscope and three-axis magnetometer. In this experiment, signals were collected for user’s 12 activities (including walking forward, walking left, walking right, walking upstairs, walking downstairs, running forward, jumping up, sitting, standing, sleeping, getting on the elevator, getting off the elevator). There were 158 subjects, aged 21–49 years, height 160-185cm and weight 43-80kg.
We compared SVM, k-NN, NB, and XGBoost. We divided the data collected by the measurement unit of the nine-axis inertial sensor of the monitor into a two-second time window, randomly selected 70% of the data as the training set and the remaining 30% as the test set. Table 1 shows the results of our experiment.
Table 2. Comparison of XGBoost and SVM, k-NN, NB.
Model
|
Accuracy
|
SVM
|
89.58%
|
k-NN
|
58.37%
|
BN
|
68.89%
|
XGBoost
|
95.68%
|
As can be seen from the above table, XGBoost is completely superior to SVM, k-NN and NB.
An head-mounted patch-type noninvasive blood pressure, blood oxygen, frontal temperature and blood glucose monitor
The head-mounted patch-type noninvasive blood pressure, blood oxygen, frontal temperature and blood glucose monitor (hereafter referred to as head-mounted monitor) features a main control chip (including Bluetooth transceiver unit), rechargeable battery power unit, reflective photoelectric volume pulse wave sensor measuring unit, forehead non-contact infrared temperature and humidity sensor measurement unit, thermal radiation temperature sensor measuring unit, environmental temperature and humidity sensor measuring unit, and the encapsulation of an ultra-small, ultra-thin and ultra-light, flat silicone flexible material shell suitable for pasting onto human skin3,4,5,15, as shown in the decomposition diagram (Fig. 5a, 5b, 5c, left, extended data in Fig. 5).
The reflex photoelectric volume pulse wave sensor measuring unit of the head-mounted monitor is used to collect the photoelectric volume pulse wave signals (including some physiological and pathological information of systemic circulation system, respiratory system, heart and vascular system, etc.) in the local tissues of the human body, and the user's blood pressure, blood oxygen and pulse frequency are measured based on the photoelectric method37,38,39,40,41,42,43,44,45,46,47,48. The working principle of the photoelectric measuring method (Fig. 5e) is that when light of a certain wavelength is irradiated on the surface of human skin, the light beam will be transmitted to the photodetector through reflection mode. In this process, due to the absorption attenuation of skin muscles and blood detected by the detector, the light intensity detected by the detector will decrease. The absorption of light by skin, muscle and tissue is constant in the whole blood circulation. The blood volume inside the skin exhibits pulsatile changes under the action of the heart. When the heart contracts, the peripheral blood volume is the largest, the light absorption is also the largest, and the light intensity is the smallest. When the heart diastole, on the contrary, the light intensity detected is the highest, so the light intensity received by the light detector also exhibits pulsating changes. The change of volume pulse blood flow can be obtained by converting the light intensity change signal into electrical signal. It can be seen that volume pulse blood flow contains many important physiological and pathological information of cardiovascular system such as blood flow. Meanwhile, volumetric pulse blood flow mainly exists in microarterioles, capillaries and other microvessels in peripheral blood vessels. Therefore, volumetric pulse blood flow also contains rich physiological and pathological information of microcirculation, which is an important source of information for us to study the human systemic circulation system and cardiovascular system.
1. Measurement of noninvasive blood pressure
We formulated a hybrid model37 for blood pressure estimation from a PhotoPlethysmoGraphy (PPG) signal based on Mean Impact Value (MIV) and Genetic Algorithm-Back Propagation (GA-BP) Neural Network (BPNN). The MIV method is used to evaluate the input variable of BP neural network and simplify the neural network model. First, we analyzed the effect of each feature of 21 characteristic parameters which are extracted from PPG signals, and removed 8 characteristic parameters of each feature that have little impact on the system, 13 parameters were selected as the input variable for BP neural network from 21 characteristic parameters, thereby reducing the redundancy of the system. Second, in order to solve the problem of low prediction accuracy caused by the initial value of the neural network extracted randomly, we use the GA to optimize the initial parameters of the neural network for a better performance. In addition, In order to overcome the problem that BP neural network is easy to fall into the local minimum, we use GA algorithm to optimize the initial weights and thresholds of BP neural networks and then establish the GA-BP model to predict blood pressure. The YASUN Vital Signs Dataset was used for training the BPNN, and more than 18900 heartbeat PPG signals were selected from 176 cases for analysis. The maximum error margin acceptable to the American National Standards is 5 ± 8 mmHg. The result of this algorithm is that 4.891 ± 5.926 mmHg for systolic blood pressure and 3.936 ± 4.819 mmHg for diastolic blood pressure, which meets the requirement of the American National Standards.
We also provided a novel method of estimating Blood Pressure (BP) from PhotoPlethysmoGram (PPG) signal38. The first 15 points of the Discrete Cosine Transform (DCT) sequence of the PPG signal are trained as inputs of the Back-Propagation Neural Network (BPNN), the Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP) extracted from the Arterial Blood Pressure (ABP) signal which is orresponding to the PPG signal are used as outputs of the BPNN. Combining the idea of AdaBoost algorithm, 10 BPNN with different initial values are chosen as “weak predictors” to form a “strong predictor” for predicting blood pressure. First, the algorithm performs DCT to transform a single heartbeat PPG segment. Using the concentrated property of the energy DCT transform, the low-frequency part of the DCT transform sequence is selected as the function for training BPNN. Then, combined with AdaBoost algorithm, 10 BPNNS with different initial weights and thresholds are selected as weak predictors to form strong predictors. In this work, we selected 26,890 heartbeat PPG signals from the YASUN Vital Signs Dataset as the training set and the test set, among which 20,000 heartbeat PPG signals were randomly selected as the training set and the remaining 6890 heartbeat PPG signals as the test set. The test results of this algorithm on these data are as follows: the systolic blood pressure is 2.2408 ± 3.8542 mmHg, and the diastolic blood pressure is 2.1006 ± 2.2482 mmHg, which meets the requirements of the American national standard of 5 ± 8 mmHg.
2. Measurement of noninvasive oxygen saturation
Since oxygenated hemoglobin in the blood (HbO2) and deaeration hemoglobin (Hb) in the red light and infrared area (600 ~ 1000 nm) has a unique absorption spectrum, it is a simple and effective test method to use PPG signal to detect the blood oxygen status in local tissues of human body. The photoelectric method for measuring pulse oxygen saturation (SPO2) is based on the Lambert-Beer rule. It takes advantage of the difference in light absorption coefficient between oxygenated hemoglobin and deoxygenated hemoglobin. In the red region (600 ~ 700 nm), the absorption difference of HbO2 and Hb is very large. However, the absorption difference is small in the infrared spectrum range (800 ~ 1000nm). When oxygen saturation changes, the relative Hb concentration of HbO2 changes. There is a good linear relationship between oxygen saturation and the relative light intensity of photodetector at 660nm and 940nm44,45.
Blood oxygen saturation: SpO2 = A + BR× 100%
Where, coefficient A and B represent constant coefficients of equation fitting, R represents the maximum intensity of red light and infrared light and the maximum variation of light intensity. SPO2 can be obtained by calculating coefficient R.
According to the above formula, 660nm red light and 940nm near-infrared light sources were selected as incident light sources for detecting SPO2, and the reflected photoelectric volume pulse wave sensor measuring unit of the head-mounted monitor was used to collect volume pulse wave signals in the user's forehead or local tissue.
3. Measurement of forehead temperature
The forehead temperature of the user is measured by the forehead non-contact infrared temperature and humidity sensor measuring unit of the head-mounted monitor3,4,5,7. The working principle of medical infrared forehead temperature measurement is generally: all objects in nature whose temperature is higher than absolute zero (-273.15℃) will emit infrared radiation, and the energy of the infrared radiation is positively correlated with temperature. Using this relationship, medical infrared frontal temperature measuring unit receives infrared radiation from the forehead of the measured human body through the infrared temperature sensor, corrects the temperature difference between the forehead and the reference body parts with appropriate algorithm, outputs and displays the reference body parts' temperature.
4. Measurement of noninvasive blood glucose concentration
Under normal circumstances, glucose is the main energy material. In the case of adequate oxygen supply, the vast majority of human tissue cells obtain energy through the aerobic oxidation of glucose. Therefore, more than 80% of the energy produced by metabolism is finally released in the form of heat energy, which is released through skin surface radiation, conduction, convection, skin surface evaporation and respiration. Since each heat dissipation component in the local position of the human body has a certain relationship with the heat dissipation component of the whole body, the metabolic heat and blood flow of the human body can be measured through the local tissues of the human body, and the glucose concentration can be estimated by using the obtained metabolic heat and oxygen content. This head-mounted monitor collects the photoelectric volume pulse wave signal, temperature, humidity, and thermal radiation temperature of the user's forehead or local tissue of human body, temperature and humidity of the surrounding environment for the user's forehead or the local tissue through using the reflected photoelectric volume pulse wave sensor measuring unit, temperature and humidity sensor measuring unit and thermal radiation temperature sensor measuring unit. These datas are used to estimate blood sugar levels by calculating the metabolic rate, oxygen saturation, pulse rate and blood flow velocity of the user's forehead or local tissue46,47,48.
We quote a non-invasive blood glucose concentration detection method47 based on conservation of energy metabolism to test and verify the effectiveness and accuracy of the head-mounted monitor in detecting user’s blood glucose. According to the method in References47, 40 volunteers were recruited for the experiment. The volunteers were told the detailed procedure and potential risks. The experiment was conducted after volunteers signed an informed consent form and was approved by the Biomedical Ethics Committee of Shenzhen yASUN Technology Co LTD. The experimental time points were 30 min before meal, 30 min, 60 min, 90 min and 120 min after meal. The entire experiment was conducted in three time periods for two consecutive days: breakfast, lunch, and dinner. The amount of data measured and the end time of data testing depends on the individual situation of each volunteer, and some individuals may not be able to participate in the collection of all data for two days. A total of 862 samples were collected. The maximum reference blood glucose concentration was 11 mmol/L and the minimum was 3.1 mmol/L. To compare the performance of quartic Multivariate Polynomial Regression (MPR4) model and Back Propagation Neural Network (BPNN) model47, 862 samples were divided into two groups by hold-out method. 603 samples were randomly selected as the training set for model calibration, and the remaining 259 samples were used as the test set for model validation and performance evaluation. Clarke’s error grid analysis is considered the clinical gold standard for assessing the accuracy of determination of blood glucose concentration and is used to assess the consistency between predicted and reference blood glucose values48. Clarke error grid analysis was performed on the results of blood glucose concentration predicted by MPR4 and BPNN models (Fig. 5f, 5g). It can be seen from the analysis results in Fig. 5f and 5g that 95.01% and 95.71% of the predicted results of MPR4 model and BPNN model are distributed in area A, respectively, which meet the accuracy of clinical requirements. BPNN model has better predictive performance than MPR4 model.
The Head-mounted Monitor has the subversive advantage of being suitable for the rapid real-time dynamic measurement of noninvasive blood pressure, blood oxygen, frontal temperature and blood glucose of local tissues of various parts for the human body. In particular, the rapid real-time dynamic detection of noninvasive blood glucose perfectly avoids the risk of using high cost, wound trauma, psychological pressure and blood infection caused by long-term and frequent invasive blood glucose detection.
An new-type wireless remote connected ecosystem
We have designed and developed an new-type wireless remote connected ecosystem9,10,11,12 (Fig, 2a, 2b) for Global Wide Area Network based on B/S architecture (Browser-Server Mode) : “Wireless Remote ECG Multi-parameter Real-time Cluster Monitoring and AI Diagnosis and Intervention Cloud Platform Software System”(Supplementary Discussion 2, and Supplementary Videos 9 and 10 and 11). The platform can provide global medical institutions and families and individuals with quality service of a six-in-one for “Screening, First Aid, Monitoring, Diagnosis and Treatment, Rehabilitation and Health Care” of the cardiovascular health and disease whole diagnosis and treatment cycle (Fig, 2c), and its application can cover all clinical application scenarios and medical service items13 of the heart and cardiovascular system. It breaks the technical bottleneck of only applying to local area networks based on the C/S (Client-Server Mode) architecture commonly used in entire world multiparameter monitoring and diagnostic software systems, high concurrency patient users can dynamically expand capacity without restrictions, truly achieves dynamic multiparameter cluster real-time monitoring and electrocardiogram diagnostic analysis across countries, regions, hospitals, wards, departments, and families worldwide, as well as fastly and automaticly online generates and publices diagnostic reports(the platform’s the diagnostic and interpretation time of 24-hour 18-lead electrocardiogram is within 10 seconds). And it can implement monitoring, diagnosis, and intervention for cardiovascular disease patients without leaving their homes, as well as timely online drug delivery.
We have built a new pattern of diagnosis and treatment service for cardiovascular health and disease with wide clinical application, high intelligence and strong scalability. With innovative system architecture and business models, combined with the construction of a graded diagnosis and treatment system for common chronic diseases and frequently-occurring diseases in the region, we have breakthrough solved the problem of insufficient multi-parameter monitoring equipments and diagnosis treatment services in primary medical institutions. Expand the existing traditional monitoring, diagnosis, and treatment services from large hospitals to grassroots medical institutions, community health centers, township health centers, and individual families.
1. Multiparameter detection APP software system
yASUN’s series products of Patch-type single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG multiparameter monitoring diagnostics devices, patch-type defibrillators and head-mounted monitors, and multiparameter monitoring diagnosis devices of third-party companies are connected to multi-parameter detection APP software system49,50,51(Supplementary Videos 5) of mobile terminals (including smart phones, tablets, laptops, etc.) through Bluetooth Low Energy(BLE). Responsible for hardware control of the devices, user interaction, as well as display of real-time multi-parameter monitoring and ECG diagnostic analysis results.
The multiparameter detection APP can be run on mobile terminals (including Smart Phone, Tablet, Notebook, etc.) that have Bluetooth, WIFI, 4G, or 5G wireless communication functions and storage space greater than 2G. It is suitable for operating system platforms across Android 6.0, Huawei HarmonyOS 2.0, and Apple iOS 12.0.
The functional realization process of yASUN’s Patch-type Monitors, Patch-type Defibrillators and Head-mounted Monitors, and ECG multiparameter devices of third-party companies are as follows: the Patch-type Monitors, Patch-type Defibrillators and Head-mounted Monitors, and ECG multiparameter devices of third-party companies collect single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG, respiration, blood pressure, blood oxygen, body temperature and blood glucose datas from the user's body surface and store them in TF card, and simultaneously transmit these multi-parameter datas to the APP on mobile devices (including smartphones, tablets, laptops, etc.) through Bluetooth; The APP displays real-time single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG, respiratory, blood pressure, blood oxygen, body temperature, and blood glucose datas; The APP stores these multi-parameter datas and remotely transmits thier to the cloud server through WIFI, 4G or 5G wireless communication for real-time monitoring and diagnostic analysis. The APP can receive and display the diagnostic analysis results and alarm informations returned by the server cluster cloud. If VF or VT is detected by the yASUN’s patch-type single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG monitors worn by the user, the user should immediately wear the Patch-type Defibrillator. At this time, the patch-type monitor worn by the user and the defibrillator communicate with each other through Bluetooth, and the ECG defibrillation monitoring mode and defibrillation workflow are simultaneously switched into (Fig. 4h). It makes yASUN’s patch-type single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG multiparameter monitor series products cooperate with the patch-type defibrillator to become a lifesaving instrument with defibrillation intervention function in seconds.
2. An algorithm and mechanism for compression, encryption, transmission and lossless reconstruction of ECG multi-parameter signals .
In order to reduce the amount of ECG multi-parameter datas transmitted wirelessly by Bluetooth of 18-lead and 12-lead ECG multi-parameter monitors, completely solve the key technical bottleneck problem that BLE transmission bandwidth cannot meet the transmission bandwidth of 18-lead and 12-lead ECG multi-parameter signals, and extend thier monitoring endurance. We design and develop an innovative ECG multi-parameter sampling, compression, encryption, transmission, lossless reconstruction method and system52,53,54,55,562,57,58,59. First, yASUN’s patch-type multi-lead monitors, patch-type defibrillators, and head-mounted monitors (If the ECG multiparameter sampling signals comes from a third-party company’s ECG multiparameter devices, the ECG multi-parameter sampling signals can be converted into yASUN’s ECG multi-parameter sampling signals standard format58,59) to collect ECG multi-parameter signals with lower precision and lower frequency52,53. Then, the sampled signals is compressed and encrypted before it is sent by Bluetooth, which ensures that the amount of ECG multi-parameter data transmitted wirelessly through Bluetooth is greatly reduced under the safe condition54,55,56,57. A lossless reconstruction mechanism for ECG multi-parameter sampled signals is designed on the mobile terminal and the cloud: That is, the compressed and encrypted ECG multi-parameter signals sampled with low precision and low frequency are reconstructed lossless into original ECG multi-parameter data with high precision and high sampling frequency by using our unique patented technology52,53. Due to the avoidance of high precision, high frequency and high speed sampling, on the one hand, the method and system significantly reduce the cost of ECG multi-parameter datas storage and transmission under the condition of ensuring the absolute safety of datas; on the other hand, it also provides a new way of extremely low cost for high precision, high sampling frequency and high dimensional data analysis of ECG multiparameter big datas. The ECG multiparameter big datas system of the new-type wireless remote connected ecosystem for yASUN’s ECG multi-parameter cluster monitoring and ECG real-time and comprehensive analysis is shown in Fig. 2d.
3. yASUN General Computer ECG Multi-parameter Information Standard Communication Protocol
It is well known that almost all new-type ECG multi-parameter monitoring diagnostic equipments are digital rather than analog, and recording, waveform interpretation and communication are assisted by various processors. These miniaturized computers can be connected to each other and to other computer devices for long-term storage and data exchange. At present, although some countries in the world and some international organizations have formulated ECG multi-parameter data and communication standards, such as SCP-ECG in Europe, MFER in Japan, HL7 and DICOM in the United States, ISHNE of International Society for Holter Electrocardiography and Noninvasive Electrocardiography, ICE 62D/60601-2-53/Ed.1: Standard Communications Protocol for Computer Assisted Electrocardiography, but there is no unified ECG multi-parameter data and communication standard in the world. Manufacturers of ECG multi-parameter monitoring and diagnostic equipments in different countries in the world generally use different technical means to collect ECG multi-parameter datas, and their definition of the ECG multi-parameter data format are not the same, which inevitably leads to the communication, exchange, sharing and processing methods of ECG multi-parameter datas are not compatible. Therefore, the ECG multi-parameter informations of different ECG multi-parameter equipments and their information processing computer systems in the world cannot communicate, exchange, share and process with each other. However, different technical approaches for users limit the range of options for using these ECG multi-parameter devices, and at the same time, hinder the development of cardiovascular science as a whole. Therefore, yASUN developed the General Computer ECG Multi-parameter Information Standard Communication Protocol.
The yASUN General Computer ECG Multi-parameter Information Standard Communication Protocol stipulates the content and the ECG multi-parameter data standard format of the information exchanged between ECG multi-parameter equipments and the computerized ECG multi-parameter management systems, as well as with other computer systems capable of storing ECG multi-parameter informations. It can establish logical links for ECG multi-parameter datas communication between any two such systems in a standard form. The ECG multi-parameter big data system of yASUN's new-type wireless remote connected ecosystem is shown in Fig. 2d. The main technical contents are as follows:
-
Collect the rest, ambulatory and motion ECG multi-parameter signals with low accuracy and frequency, and compress, encrypt, and encode these datas. To ensure the integrity of the rest, ambulatory and motion ECG multi-parameter datas within the specified error limits, and to provide quality assurance for the processing and electronic exchange of the rest, ambulatory and motion ECG multi-parameter datas.
-
The basic information content of the communication between ECG multi-parameter equipments and computers. It includes patient identification data, ECG multi-parameter datas, reference heartbeats, waveform interpretation, measurement results, diagnosis and the diagnostic codes customized by the manufacturer.
-
The standard format of the data exchange informations. Communication contents (including patient’s identification datas, ECG multi-parameter datas, reference heartbeats, waveform interpretation, measurement results, diagnostic informations, etc.) are uniformly organized into yASUN ECG multi-parameter data standard format in communication packets: Firstly, the ECG multi-parameter data format collected according to different standards such as European SCP-ECG, Japanese MFER, American HL7 and DICOM, ISHNE and ICE is converted into yASUN ECG multi-parameter data standard format. Then, yASUN ECG multi-parameter signals compression, encryption, transmission and lossless reconstruction methods are used to reconstruct the compressed and encrypted ECG multi-parameter signals with low precision and frequency into the original ECG multi-parameter datas with high precision and high sampling frequency.
-
Control and query the structure of messages, message queue control. The format of the message header requested and transmitted at the application layer between ECG multi-parameter devices, the state order, queue control, and method protocol required for data transmission.
-
Wireless communication protocol for ECG multi-parameter equipments and computers. Using Bluetooth, WIFI, or 4G, or 5G, Internet communication ports, from the ECG multi-parameter equipments to the computers communication protocols for the transmission of ECG multi-parameter datas: Bluetooth transmission protocol, TCP based WebSocket transmission protocol, HTTPS based transmission protocol, etc.
-
Low-level communication protocols for ECG multi-parameter equipments and computers. Using RS232 communication port, UART communication port, USB communication port, etc. from the ECG multi-parameter devices to the computers to transmit the ECG multi-parameter data of the low-level communication protocols.
4. Construction of yASUN Vital Signs ECG Multiparameter Database
According to the requirements of yASUN General Computer ECG Multi-parameter Information Standard Communication Protocol, we have built the yASUN Vital Signs ECG Multiparameter Database, whose ECG multiparameter data standard format is built according to the rules for ECG multiparameters big data system of the new-type wireless remote connected ecosystem for yASUN ECG multiparameter cluster monitoring and real-time comprehensive analysis, as shown in Fig. 2d. Partial data samples of the ECG Multi-parameter Database (including 30 minutes and 24 hours single-lead, 7-lead, 8-lead, 12-lead and 18-lead ECG sample datas, pulse wave sample datas, blood pressure, blood oxygen, body temperature, blood glucose sample datas, etc.) were collected by the yASUN’s patch-type single-lead, 7-lead, 8- lead, 12- lead and 18-lead ECG multiparameter monitor series products, patch-type defibrillators and head-mouted patch-type blood pressure, blood oxygen, body temperature, and blood glucose monitors, as well as the third-party company’s multi-parameter monitoring diagnostic equipments; Another part of the data samples of the ECG Multi-parameter Database are selected from the CSE database, the AHA database, the MIT database, the NST database, the CU database, the PhysioNet database, the Physikalisch-Technische Bundesanstalt’s database, and University of Queensland vital signs Database, etc., and these sample datas were converted into yASUN’s ECG multiparameter data standard format. So far, yASUN vital signs ECG multiparameter database collected 2984,632 single-lead, 7-lead, 8- lead, 12- lead and 18-lead ECG sample data, pulse wave sample data, blood pressure, blood oxygen, body temperature, blood sugar and other desensitization sample datas from 1232,862 patients.
AI detection and diagnosis models for Cardiovascular health and diseases
We have developed a complete set of AI detection and diagnosis models60,61,~,75,76 to automatically classify 28 common cardiovascular health and disease categories using the single-lead, 7-lead, 8- lead, 12- lead and 18-lead ECGs from patients who weared the patch-type single-lead, 7-lead, 8- lead, 12- lead and 18-lead ECG monitors, and the patch-type defibrillators. These 28 common cardiovascular health and disease categories are sinus rhythm (including normal sinus rhythm, sinus bradycardia, sinus tachycardia, sinus arrhythmia and sinus arrest), supraventricular arrhythmia, ventricular arrhythmia, atrioventricular junction arrhythmia, preexcitation syndrome, heart block, pacemaker electrocardiogram, atrioventricular hypertrophy, myocardial infarction, myocardial ischemia, sleep apnea syndrome, etc. Arrhythmia types included cardiac arrest, sinus arrest, sinus tachycardia, sinus bradycardia, sinus arrhythmia, atrial premature beat, preatrial systolic bigeminal rhythm, atrial extrasystoles triad, atrial premature pairs, paroxysmal supraventricular tachycardia, atrial tachycardia, atrial flutter, atrial fibrillation, ventricular extrasystoles, ventricular extrasystoles bigeminy, ventricular extrasystoles triad Premature ventricular contraction pairs, paroxysmal ventricular tachycardia, VT, ventricular flutter, ventricular fibrillation.
The complete set of AI detection and diagnosis models for cardiovascular health and diseases proposed and developed by us are described as follows:
We have proposed a multi-label learning algorithm for ECGs detection60, an adaptive algorithm for premature ventricular contraction recognition based on multiple template matching61,62, an ECG synthesis detection algorithm based on wavelet transform63, a deep learning arrhythmia detection method based on the multi-input and single-output joint optimization model64, an electrocardiogram beat classification method based on deep belief networks65, a SVM algorithm model optimized by kernel function to classify the electrocardiogram66, a method based on SVM for classification of arrhythmias67, an SE-ResNet model algorithm without hand-crafted features extraction68, an atrial fibrillation detection algorithm based on EEMD and XGBoost69, an algorithm for Detecting Atrial Fibrillation Using RR Intervals70, a ventricular fibrillation detection by an improved time domain algorithm combined with SVM71, an optimal template matching discriminant algorithm for complete bundle branch block72, an automatic detection algorithm for myocardial ischemia and myocardial infarction based on the heartbeat-attention mechanism73, an automatic detection algorithm for myocardial ischemia and myocardial infarction based on the dominant T-wave cluster74, a heart failure detection method based on one-dimensional CNNs75, a method based on fixed number of consecutive ECG heartbeats via bidirectional RNN combined with attention mechanism to detect CHF76.
We have deployed and implanted these AI detection and diagnostic models for cardiovascular health and disease into yASUN’s new-type wireless remote connected ecosystem. In this work, from the yASUN Vital Signs ECG Multi-parameter Dataset, we respectively selected 100,000 haertbeats’ ECG sample datas of single-lead, 7-lead, 8-lead, 12-lead and 18-lead of the above 28 common cardiovascular health and disease categories as the training set and test set, in which 70% haertbeats’ ECG sample datas were randomly selected as the training set, 30% of the haertbeats’ ECG sample datas was used as the test set. The selected training sets were used to conduct model iteration training of the AI detection and diagnosis models for cardiovascular health and disease. These algorithms were more than 95% accurate in automatically classifying 28 common cardiovascular health and disease categories from ECG samples in the test set.