Portable sleep breathing disorder monitoring equipment based on electroencephalogram, oxyhemoglobin saturation and electrocardio

Abstract

The invention discloses portable sleep breathing disorder monitoring equipment based on electroencephalogram, oxyhemoglobin saturation and electrocardio, and belongs to the technical field of sleep monitoring equipment. The monitoring equipment comprises a wireless sleep monitoring module and a sleep breathing disorder diagnosis module, the wireless sleep monitoring module is used for collecting EEG, SpO2 and ECG signals in the sleep process and transmitting the EEG, SpO2 and ECG signals to the sleep dyspnea diagnosis module in real time for signal processing and analysis. The equipment collects multiple modes directly related to accurate diagnosis of sleep breathing disorder, and huge deviation of evaluation of sleep breathing disorder caused by a single mode is avoided. A sensor for collecting forehead electroencephalogram and oxyhemoglobin saturation is integrated on the flexible circuit board, so that the integration level and portability of the device are higher. A multi-modal fusion algorithm is designed for multi-modal signals collected by equipment, and multi-task and accurate detection of automatic sleep staging and sleep breathing events is achieved.

Description

Technical Field

[0001] The invention relates to a portable sleep breathing disorder monitoring device based on EEG-blood oxygen saturation-ECG, which belongs to the technical field of sleep monitoring devices. Background Art

[0002] Obstructive sleep apnea (OSA) is a common sleep breathing disorder. Clinically, the gold standard for the diagnosis of sleep breathing disorders is polysomnography (PSG) (Huttunen R, Leppanen T, Duce B, et al. A Comparison of Signal Combinationsfor Deep Learning-Based Simultaneous Sleep Staging and Respiratory Event Detection[J]. IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, 2023, 70(5): 1704-1714). PSG can provide doctors with rich multimodal disease diagnostic information, which helps to accurately diagnose the disease. However, the use of PSG has some disadvantages for both medical staff and patients with sleep disorders: (1) It is complicated to wear and requires professional technicians to operate; (2) A dedicated sleep center or sleep monitoring room is required; (3) Manual analysis of PSG records is time-consuming, labor-intensive, and highly subjective; (4) Patients have a poor subjective sleep experience. These limitations lead to low diagnosis and treatment rates of OSA, further exacerbating the severity of the disease and increasing the social burden.

[0003] There are many commercial portable sleep monitoring devices that monitor the sleep process, such as the Zhaoguan sleep breathing initial screening device developed by Hangzhou Zhaoguan Sensing Technology Co., Ltd. and the finger pulse oximeter of Yuyue. Related patents include: an ECG sleep apnea detection method and system based on time series contrast learning "CN202410214466.5", a sleep apnea detection system and method based on millimeter wave radar "CN202410248483.0", etc., which collect one of the physiological signals such as chest and abdominal movement, finger pulse oxygen (Saturation of peripheral oxygen, SpO2), electrocardiogram (Electrocardiogram, ECG), etc., to detect respiratory events (mainly including apnea and hypopnea) during sleep, and then realize the initial screening of sleep breathing disorders. The clinical determination of the severity of sleep breathing disorders requires not only the number of respiratory events, but also the judgment of the patient's corresponding sleep state, and the accurate judgment of the sleep state requires the combination of multiple modalities such as electroencephalogram (Electroencephalogram, EEG). Compared with standard PSG, these devices mainly use a single modality and make rough judgments on sleep respiratory events through single indicators such as body movement and pulse oxygen, which cannot meet the clinical needs for precise disease diagnosis. Summary of the invention

[0004] Under the premise of ensuring monitoring accuracy, developing portable and comfortable sleep breathing disorder monitoring equipment can not only effectively solve the above problems, but also promote the rapid development of sleep medicine.

[0005] The technical solution adopted by the present invention is:

[0006] The wireless sleep monitoring module is used to collect EEG, SpO2, and ECG signals during sleep and transmit them in real time to the sleep breathing disorder diagnosis module for signal processing and analysis;

[0007] The module includes electrodes, analog quantity acquisition module, main control core, wireless transceiver, and power supply circuit;

[0008] The electrodes include a surface-mounted dry electrode, a forehead-mounted blood oxygen electrode and a bipolar patch electrode. The surface-mounted dry electrode and the forehead-mounted blood oxygen electrode are integrated in a forehead sensor, and the forehead sensor is attached to the skin at the forehead of the subject.

[0009] The surface-mount dry electrode uses a flexible film with n exposed electrodes, which transmit the EEG signal of the measured area to the wireless signal acquisition device through the surface circuit;

[0010] The forehead-mounted blood oxygen electrode uses a reflective optical method to measure the SpO2 signal;

[0011] Bipolar patch electrodes collect ECG signals from any two locations in the human chest to monitor ECG activity during sleep;

[0012] The analog quantity acquisition module has multiple analog quantity acquisition channels. After the acquired signal is amplified by a programmable gain amplifier, the signal is converted into a corresponding digital quantity using an analog-to-digital converter.

[0013] The main control core communicates with the analog acquisition module through the data communication bus, configures the acquisition parameters and reads the sampling values ​​of each channel;

[0014] The main control core and the wireless transceiver are in a chip package, and the main control core sends the collected sampling values ​​of each channel to the sleep breathing disorder diagnosis module in real time through the wireless transceiver; the power supply circuit includes a battery charging and discharging circuit and a system power management circuit;

[0015] b. Personalized sleep assessment module

[0016] The personalized sleep assessment module mainly performs personalized sleep quality assessment and sleep breathing disorder classification diagnosis by designing a multimodal fusion model, including signal preprocessing, model design, model training and testing, and sleep assessment. Specifically:

[0017] S1. EEG, SpO2, and ECG signals are used as the input of the model. Signal preprocessing consists of artifact removal, filtering, and data truncation.

[0018] S2. Build a multimodal fusion model:

[0019] In order to fully extract the features reflecting disease information in the three modalities, a model-level feature fusion algorithm based on a convolutional neural network is constructed. Compared with other multimodal fusion algorithms, this method can process signals of different lengths at the same time without destroying the original information of the data. The algorithm is mainly modular in design and includes 4 modules:

[0020] Module 1: The EEG signal is used as the input of this module. The module consists of three one-dimensional convolutional layers with different convolution kernel sizes, which are 64, 128, and 256 respectively;

[0021] Module 2: Since the order of magnitude of SpO2 and EEG signals differ greatly, the EEG features obtained in module 1 and the preprocessed SpO2 signals are first normalized through the batch normalization layer. Then, the two normalized signals are fused through the concatenation layer. Then, the fused features are input into the convolution layer with the same settings as module 1 for feature extraction.

[0022] Module 3: Similarly, the features obtained in module 2 and the ECG signal are normalized and fused, and the fused features are input into the long short-term memory network to extract the joint features of the three modalities.

[0023] Module 4: For sleep quality assessment, sleep staging is required to obtain the user's sleep structure. In order to reduce the impact of redundant information on the results of automatic sleep staging, a residual network is constructed, and only the EEG and ECG features in modules one and three are input. After fusion, they are input into a fully connected layer with 100 neurons, followed by a five-classification task of wakefulness, non-rapid eye movement 1 (N1) / 2 (N2) / 3 (N3), and rapid eye movement. For the diagnosis of sleep breathing disorders, abnormal sleep breathing event detection is required. The features extracted from module three are directly input into the fully connected layers with 100 and 2 neurons in turn to detect abnormal / normal sleep breathing events.

[0024] S3. Model training and testing. The optimization function of the model is "adam", and the functions used in the classification layer are "softmax" and "sigmoid". Set the number of model training iterations to 100, and retain the model weights of the iteration batch with the highest accuracy in the validation set to test the data in the test set;

[0025] S4. Sleep assessment. Based on the results of sleep staging and sleep respiratory event detection, the apnea-hypopnea index is calculated to monitor and automatically diagnose obstructive sleep apnea. The monitoring information of long-term users is dynamically analyzed to provide personalized and dynamic assessment results.

[0026] The sleep breathing disorder diagnosis module includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above steps S1-S4 are implemented.

[0027] A portable sleep breathing disorder monitoring device based on EEG-oxygen saturation-ECG includes a wireless acquisition device for EEG-oxygen saturation-ECG sleep signals and a personalized sleep assessment module.

[0028] Furthermore, the surface-mounted dry electrode adopts a flexible film and has n exposed electrodes, which can be attached to the skin of the forehead of the subject, and transmit the EEG signal of the measured area to the wireless signal acquisition device through the surface circuit. In addition, the present invention uses a reflective optical method to measure SpO2. Compared with the existing fingertip oximeters that use the principle of light absorption, the forehead-mounted oximeter designed by the present invention is easy to operate while ensuring measurement accuracy, has little impact on the user's sleep, and more importantly, can be integrated with EEG acquisition to improve the overall integration of the equipment. Then, ECG signals at any two positions of the human chest are collected through bipolar patch electrodes to monitor ECG activities during sleep; the analog quantity acquisition module has multiple analog quantity acquisition channels, and the collected signals are amplified by a programmable gain amplifier and then converted into corresponding digital quantities using an analog-to-digital converter; the main control core communicates with the analog quantity acquisition module through the data communication bus, configures the acquisition parameters (sampling rate, resolution, etc.) and reads the sampling values ​​of each channel; the main control core and the wireless transceiver are in a chip package, and the main control core sends the collected sampling values ​​of each channel to the measurement and control system in real time through the wireless transceiver; the power supply circuit includes a battery charging and discharging circuit and a system power management circuit to power the various functional modules of the acquisition device.

[0029] The beneficial effects of the present invention are as follows: In response to the problems existing in current medical and commercial sleep monitoring equipment, the present invention has developed a portable sleep breathing disorder monitoring device based on EEG-blood oxygen saturation-ECG, which mainly includes a wireless signal acquisition device and a personalized sleep assessment algorithm. The improvements of the device are as follows: (1) Collecting multiple modalities directly related to the accurate diagnosis of sleep breathing disorders to avoid the huge deviation caused by a single modality in the assessment of sleep breathing disorders; (2) Integrating sensors for collecting forehead EEG and blood oxygen saturation on a flexible circuit board to make the device more integrated and portable; (3) Designing a multimodal fusion deep learning algorithm for the multimodal signals collected by the device to achieve automatic sleep staging and sleep breathing event detection, thereby accurately diagnosing the severity of sleep breathing disorders. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of a portable sleep breathing disorder monitoring device based on EEG-blood oxygen saturation-ECG according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The technical solutions in the implementation of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention.

[0032] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Figure 1 FIG. 1 is a schematic diagram of a portable sleep breathing disorder monitoring device based on EEG-blood oxygen saturation-ECG according to an embodiment of the present invention. Figure 1 As shown in the figure, the device mainly includes a wireless sleep monitoring device and a personalized sleep assessment module, wherein the wireless sleep monitoring device is mainly composed of a signal acquisition sensor and a signal acquisition circuit. The signal acquisition sensor collects frontal EEG, head SpO2 and ECG signals during the subject's sleep. The signal acquisition module includes three parts: acquisition power supply, filter circuit and main control circuit. The personalized sleep assessment module is mainly implemented through a deep learning algorithm, which mainly includes four steps: signal preprocessing, model design, model training and testing, and sleep assessment.

[0034] A portable sleep breathing disorder monitoring device based on EEG-blood oxygen saturation-ECG includes a wireless sleep monitoring device of EEG-blood oxygen saturation-ECG sleep signals and an automatic diagnosis algorithm for sleep breathing disorders.

[0035] (1) Wireless sleep monitoring module

[0036] The wireless sleep monitoring module is used to collect EEG, SpO2, and ECG signals during sleep, and transmit them to the measurement and control system in real time for signal processing and analysis. The device includes a surface-mounted dry electrode, an analog quantity acquisition module, a main control core, a wireless transceiver, a power supply circuit, etc. The surface-mounted dry electrode adopts a flexible film and has n exposed electrodes, which can be attached to the skin at the forehead of the subject, and the EEG signal of the measured area is transmitted to the wireless signal acquisition device through the surface circuit. In addition, the present invention uses a reflective optical method to measure SpO2. Compared with the existing fingertip oximeters that use the principle of light absorption, the forehead oximeter designed by the present invention is easy to operate while ensuring the measurement accuracy, has little impact on the user's sleep, and more importantly, can be integrated with EEG acquisition to improve the overall integration of the equipment. Then, ECG signals at any two positions of the human chest are collected through bipolar patch electrodes to monitor ECG activities during sleep; the analog quantity acquisition module has multiple analog quantity acquisition channels, and the collected signals are amplified by a programmable gain amplifier and then converted into corresponding digital quantities using an analog-to-digital converter; the main control core communicates with the analog quantity acquisition module through the data communication bus, configures the acquisition parameters (sampling rate, resolution, etc.) and reads the sampling values ​​of each channel; the main control core and the wireless transceiver are in a chip package, and the main control core sends the collected sampling values ​​of each channel to the measurement and control system in real time through the wireless transceiver; the power supply circuit includes a battery charging and discharging circuit and a system power management circuit to power the various functional modules of the acquisition device.

[0037] (2) Personalized sleep assessment module

[0038] The personalized sleep assessment module mainly includes signal preprocessing, model design, model training and testing, and sleep assessment. First, the present invention uses three modal signals of EEG, SpO2, and ECG as the input of the model. Signal preprocessing consists of artifact removal, filtering, and data truncation. Secondly, a model-level multimodal fusion model is designed to modularly process and fuse the three modal signals collected by the device; module one extracts the features of the EEG signal, module two fuses the features extracted by module one with the SpO2 signal, and module three further fuses the features extracted by module two with the ECG signal, and finally inputs them into the fully connected layer for deep fusion of features. Then, the model is trained and tested to perform sleep staging and sleep respiratory event detection, and the model parameters are dynamically adjusted according to the information of long-term monitoring and evaluation of the user; finally, the apnea-hypopnea index is calculated according to the results of sleep staging and sleep respiratory event detection, so as to achieve personalized sleep quality assessment and graded diagnosis of obstructive sleep apnea.

[0039] The personalized assessment module mainly performs personalized sleep quality assessment and sleep breathing disorder graded diagnosis, which is achieved by designing a multimodal fusion model, including signal preprocessing, model design, model training and testing, and sleep assessment. This module includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the following steps S1-S4 are implemented; specifically:

[0040] S1. EEG, SpO2, and ECG signals are used as the input of the model. Signal preprocessing consists of artifact removal, filtering, and data truncation.

[0041] S2. Build a multimodal fusion model:

[0042] In order to fully extract the features reflecting disease information in the three modalities, a model-level feature fusion algorithm based on a convolutional neural network is constructed. Compared with other multimodal fusion algorithms, this method can process signals of different lengths at the same time without destroying the original information of the data. The algorithm is mainly modular in design and includes 4 modules:

[0043] Module 1: The EEG signal is used as the input of this module. The module consists of three one-dimensional convolutional layers with different convolution kernel sizes, which are 64, 128, and 256 respectively;

[0044] Module 2: Since the order of magnitude of SpO2 and EEG signals differ greatly, the EEG features obtained in module 1 and the preprocessed SpO2 signals are first normalized through the batch normalization layer. Then, the two normalized signals are fused through the concatenation layer. Then, the fused features are input into the convolution layer with the same settings as module 1 for feature extraction.

[0045] Module 3: Similarly, the features obtained in module 2 and the ECG signal are normalized and fused, and the fused features are input into the long short-term memory network to extract the joint features of the three modalities.

[0046] Module 4: For sleep quality assessment, sleep staging is required to obtain the user's sleep structure. In order to reduce the impact of redundant information on the results of automatic sleep staging, a residual network is constructed, and only the EEG and ECG features in modules one and three are input. After fusion, they are input into a fully connected layer with 100 neurons, followed by a five-classification task of wakefulness, non-rapid eye movement 1 (N1) / 2 (N2) / 3 (N3), and rapid eye movement. For the diagnosis of sleep breathing disorders, abnormal sleep breathing event detection is required. The features extracted from module three are directly input into the fully connected layers with 100 and 2 neurons in turn to detect abnormal / normal sleep breathing events.

[0047] S3. Model training and testing. The optimization function of the model is "adam", and the functions used in the classification layer are "softmax" and "sigmoid". Set the number of model training iterations to 100, and retain the model weights of the iteration batch with the highest accuracy in the validation set to test the data in the test set;

[0048] S4. Sleep assessment. Based on the results of sleep staging and sleep respiratory event detection, the apnea-hypopnea index is calculated to monitor and automatically diagnose obstructive sleep apnea. The monitoring information of long-term users is dynamically analyzed to provide personalized and dynamic assessment results.

Claims

1. A portable sleep apnea monitoring device based on EEG-blood oxygen saturation-ECG, characterized in that: The monitoring device includes a wireless sleep monitoring module and a personalized sleep assessment module; a. Wireless sleep monitoring module The wireless sleep monitoring module is used to collect EEG, SpO2, and ECG signals during sleep and transmit them in real time to the sleep breathing disorder diagnosis module for signal processing and analysis; The module includes electrodes, analog quantity acquisition module, main control core, wireless transceiver, and power supply circuit; The electrodes include a surface-mounted dry electrode, a forehead-mounted blood oxygen electrode and a bipolar patch electrode. The surface-mounted dry electrode and the forehead-mounted blood oxygen electrode are integrated in a forehead sensor, and the forehead sensor is attached to the skin at the forehead of the subject. The surface-mount dry electrode uses a flexible film with n exposed electrodes, which transmit the EEG signal of the measured area to the wireless signal acquisition device through the surface circuit; The forehead-mounted blood oxygen electrode uses a reflective optical method to measure the SpO2 signal; Bipolar patch electrodes collect ECG signals from any two locations in the human chest to monitor ECG activity during sleep; The analog quantity acquisition module has multiple analog quantity acquisition channels. After the acquired signal is amplified by a programmable gain amplifier, the signal is converted into a corresponding digital quantity using an analog-to-digital converter. The main control core communicates with the analog acquisition module through the data communication bus, configures the acquisition parameters and reads the sampling values ​​of each channel; The main control core and the wireless transceiver are in a chip package, and the main control core sends the collected sampling values ​​of each channel to the sleep breathing disorder diagnosis module in real time through the wireless transceiver; the power supply circuit includes a battery charging and discharging circuit and a system power management circuit; b. Personalized sleep assessment module The personalized sleep assessment module includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the following steps S1-S4 are implemented, specifically: S1. EEG, SpO2, and ECG signals are used as the input of the model. Signal preprocessing consists of artifact removal, filtering, and data truncation. S2. Build a multimodal fusion model: In order to extract the features reflecting disease information from the three modalities, a multimodal fusion model based on a convolutional neural network was constructed. The multimodal fusion model is modularly designed and includes four modules: Module 1: The EEG signal is used as the input of this module; Module 1 consists of three one-dimensional convolutional layers with different convolution kernel sizes, which are 64, 128, and 256 respectively; Module 2: First, the EEG features obtained in module 1 and the preprocessed SpO2 signals are normalized through the batch normalization layer; secondly, the two normalized signals are fused through the concatenation layer; then, the fused features are input into the convolution layer with the same settings as module 1 for feature extraction; Module 3: The features obtained in module 2 and the ECG signal are normalized and fused using the same method as module 2, and the fused features are input into the long short-term memory network to extract the joint features of the three modalities; Module 4: For sleep quality assessment, sleep staging is required to obtain the user's sleep structure; a residual network is constructed, and only the EEG and ECG features in modules 1 and 3 are input, and after fusion, they are input into the fully connected layer, followed by a five-classification task of awake period, non-rapid eye movement 1, non-rapid eye movement 2, non-rapid eye movement 3, and rapid eye movement period; For sleep breathing disorder monitoring, abnormal sleep breathing event detection is required; the features extracted by module three are directly input into the fully connected layer in sequence to detect abnormal / normal sleep breathing events; S3. Model training and testing; the optimization function of the model is "adam", and the functions used in the classification layer are "softmax" and "sigmoid" respectively; set the number of training iterations of the model, and retain the model weight of the iteration batch with the highest accuracy in the validation set to test the data in the test set; S4. Sleep assessment: Based on the results of sleep staging and sleep respiratory event detection, the apnea-hypopnea index is calculated to monitor obstructive sleep apnea. The monitoring information of long-term users is dynamically analyzed to provide them with personalized and dynamic assessment results.

Images