An automatic recognition system of artery and vein wave based on multi-channel neck detector

Abstract

The application discloses an automatic identification system of arteriovenous wave based on a multi-channel neck detector, relates to the technical field of physiological signal detection and processing, and realizes automatic identification of jugular venous wave and carotid arterial wave by acquiring a wide range of neck physiological data by using the multi-channel neck detector and based on time domain characteristics, frequency domain characteristics and a multi-channel fusion strategy. By using the natural physiological anatomical differences between channels, the signal differences of arteriovenous waves in different regions are captured, noise interference is stripped, multi-dimensional time domain and frequency domain characteristics are jointly extracted, and the multi-channel is selectively selected, so that the essential differences of arteriovenous waves are fully mined, the system can adapt to the complex physiological environment of the neck, individual differences and external interference, is safer compared with traditional invasive detection, and has a wider coverage compared with single-channel non-invasive detection.

Description

Technical Field

[0001] This invention relates to the field of physiological signal detection and processing technology, specifically to an automatic arteriovenous wave identification system based on a multi-channel neck detector. Background Technology

[0002] In clinical physiological monitoring, accurately acquiring and distinguishing between jugular venous waves and carotid artery waves is a crucial prerequisite for assessing the function of the human circulatory system and the metabolic state of brain tissue. Jugular venous waves reflect right ventricular function and venous return, while carotid artery waves are closely related to arterial blood pressure and cardiac pumping function. Accurate identification of both is of great significance in scenarios such as intensive care, cardiovascular disease diagnosis, and anesthesia management.

[0003] Currently, there are many limitations to the methods used in clinical acquisition of carotid artery and vein signals. Traditional invasive detection methods are high-risk, invasive, and difficult to implement continuous monitoring. Among non-invasive methods, photoplethysmography (PPG) is widely used, but existing technologies mostly use single-channel or small-area detection methods, which cannot fully cover the complex vascular distribution area of ​​the neck. Due to the complex anatomy of the neck, individual differences in arterial and venous locations, weak venous signals, and significant interference from arterial signals, single-point or small-area detection provides limited information, easily leading to signal instability and inaccurate classification. In addition, existing technologies lack a recognition system based on multi-channel, large-area coverage, making it difficult to utilize the natural differences in sensitivity between different channels to arteries or veins to improve recognition robustness.

[0004] Existing research has attempted to improve arteriovenous wave recognition by optimizing sensor optical paths and improving signal processing algorithms, but core problems remain: First, there is a lack of multi-channel detection devices capable of acquiring large-scale neck data, making it impossible to fully utilize the signal differences between different channels to improve recognition accuracy; second, the automatic arteriovenous wave discrimination methods lack robustness and are difficult to adapt to the complex physiological environment of the neck and individual differences. Currently, there is no mature technical solution for robust automatic recognition of jugular vein waves and carotid artery waves based on multi-channel, large-scale detection data for the neck region.

[0005] Therefore, there is an urgent need for a technical solution that can cover a large area of ​​the neck, acquire PPG signals from multiple channels, and combine time-frequency domain features with a multi-channel fusion discrimination mechanism to achieve automatic and reliable identification of carotid artery waves and jugular vein waves. Summary of the Invention

[0006] In view of the shortcomings of the prior art, the purpose of this invention is to provide an automatic arteriovenous wave identification system based on a multi-channel neck detector. Through multi-channel large-range acquisition, periodically stable signal preprocessing, multi-dimensional feature extraction and channel-level probabilistic fusion mechanism, it can achieve highly robust automatic identification of jugular vein waves and carotid artery waves.

[0007] This invention proposes an automatic arteriovenous wave identification system based on a multi-channel neck detector, the system comprising:

[0008] Multi-channel PPG data acquisition device: used to acquire the raw PPG signals of the corresponding detection area of ​​each channel;

[0009] Multi-channel signal preprocessing module: used to filter, denoise, and periodize the acquired raw PPG signal;

[0010] Multi-channel feature extraction module: extracts time-domain and frequency-domain features from the preprocessed features; the time-domain features include time parameters, peak-valley parameters, and waveform morphology parameters; the extraction of frequency-domain features includes calculating the dominant frequency feature, statistically analyzing harmonic features, and thus calculating the energy distribution features of each feature frequency;

[0011] The multi-channel data feature discrimination module distinguishes between the time-domain and frequency-domain features of the carotid artery and vein, and obtains the probability of arterial and venous waves for each channel.

[0012] The multi-channel fusion module based on the principle of maximizing probability selects the channel with the highest probability of arterial wave as the source of carotid artery wave and the channel with the highest probability of venous wave as the source of jugular venous wave.

[0013] Furthermore, the multi-channel PPG data acquisition device uses a flexible substrate that fits tightly against the skin of the neck, and forms multiple channels through several photosensitive detectors arranged in an array around the light source.

[0014] Furthermore, the denoising process uses a moving average filtering method to remove baseline drift in the PPG signal and a notch filter to remove power frequency interference.

[0015] Furthermore, the filtering process specifically involves: employing an infinite impulse response filter, using bidirectional filtering to obtain the AC components of each channel.

[0016] Furthermore, the period division is specifically as follows:

[0017] The pressure wave signal is generated by multiplying the AC component of the filtered light intensity signal by a negative one and reversing it. The first-order difference between adjacent sampling points of the AC signal is calculated, and the position where the difference changes from negative to positive is taken as the candidate valley value. The interquartile range method is used to screen unreasonable valley values. A search window is set near each candidate point to search for local smaller values ​​as valley values. In order to ensure the consistency of period division, when there are multiple smaller values, the one with the smaller time sequence in the search window is selected as the minimum valley value division period to ensure the consistency of period division.

[0018] Furthermore, the time-domain features specifically include:

[0019] Time parameters: Calculate the rise time from the beginning trough to the peak value, the fall time from the peak value to the end trough value, and the ratio of the two in a single period signal; at the same time, calculate the relative time of each peak and trough within the period.

[0020] Peak-valley parameters: Record the position and amplitude of the zero-crossing point of the first derivative in a single periodic signal, identify it as the peak value and valley value, calculate the maximum peak-valley difference, and calculate the sum of the peak amplitude and the product of the peak value and the valley amplitude and the product of the valley value and the valley value and the time.

[0021] Waveform morphology parameters: Calculate the maximum rising slope and maximum falling slope of the extracted waveform.

[0022] Furthermore, the process of obtaining the energy distribution characteristics specifically includes:

[0023] Calculate the peak frequency of the dominant frequency located near the heart rate in the frequency domain;

[0024] Statistical analysis of spectral peaks other than heart rate frequency, namely the amplitude and location of the second, third, and fourth harmonics;

[0025] Calculate the total signal energy within the filtered frequency band, and the ratio of the energy of each characteristic frequency band to the total signal energy. Calculate the power weighted ratio within the main frequency band as the harmonic index.

[0026] Furthermore, in the multi-channel data feature discrimination module, the discrimination of time-domain features includes: the ratio of carotid artery wave rise time to fall time is less than 1; the ratio of carotid vein wave rise time to fall time is greater than 1; the sum of the peak amplitude and the product of peak relative time of the carotid vein wave, and the sum of the trough amplitude and the product of trough relative time of the trough are both greater than those of the carotid artery wave; the maximum rise slope of the carotid artery wave is higher than that of the carotid vein wave, while the absolute value of the maximum fall slope of the carotid artery wave is lower than that of the carotid vein wave.

[0027] Furthermore, in the multi-channel data feature discrimination module, the discrimination of frequency domain features includes: the proportion of harmonic energy of the carotid artery wave to the total signal energy is greater than that of the jugular vein wave; and the harmonic index of the carotid artery is greater than that of the jugular vein.

[0028] Furthermore, the probabilities of arterial and venous waves in each channel are obtained by weighted summation based on the results of each discrimination rule in that channel.

[0029] The beneficial effects of this invention are as follows: This invention constructs a robust automatic arteriovenous wave recognition system by utilizing the natural physiological and anatomical differences between channels through a complete process of "multi-channel large-area data acquisition - multi-dimensional feature extraction - multi-feature discrimination - optimal channel selection". The multi-channel neck detector can comprehensively cover the complex vascular distribution area of ​​the neck, capture the differences in arteriovenous signals in different areas, and provide richer raw data support for recognition; the signal preprocessing stage effectively removes noise interference, ensuring the reliability of feature extraction; the joint extraction of multi-dimensional time-domain and frequency-domain features and the optimal selection of multiple channels fully explore the essential differences between arteriovenous waves, significantly reducing the risk of misjudgment caused by a single channel or single feature, enabling the system to adapt to the complex physiological environment of the neck, individual differences, and external interference, and achieve accurate automatic recognition of jugular vein waves and carotid artery waves, effectively solving the core problems of incomplete small-area detection coverage and insufficient recognition robustness in existing technologies.

[0030] The technical method employed in this invention possesses both unique advantages and broad application value. The multi-channel flexible detector design conforms to the neck skin, offering convenient operation and stable signal acquisition. Compared to traditional invasive detection methods, it is safer, and compared to single-channel non-invasive detection methods, it provides a wider coverage area. The signal processing algorithm, through an optimized combination of mature technologies such as filtering, noise reduction, period division, and FFT energy analysis, controls computational load while ensuring recognition accuracy, making it suitable for embedded device deployment and supporting continuous real-time monitoring. Furthermore, the arteriovenous waveforms identified by the system can be further used to calculate arteriovenous oxygen saturation and determine the non-invasive central venous pressure measurement site. Some time-domain features (peak-to-peak value, rise and fall time, etc.) can also indirectly assess right ventricular function, providing comprehensive physiological signal support for intensive care, cardiovascular disease diagnosis, and anesthesia management, with an applicability far exceeding existing single-function monitoring technologies. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the structure of a single light source with multiple photodetectors in this invention;

[0032] Figure 2 This is a schematic diagram of the structure of multiple light sources and multiple photodetectors in this invention;

[0033] Figure 3 This is a schematic diagram of the multi-channel neck PPG data acquisition and preprocessing process in this invention;

[0034] Figure 4 This is a comparison of measured jugular vein waves and carotid artery waves acquired through different channels in this invention;

[0035] Figure 5 This is a schematic diagram of the multi-channel feature discrimination process in this invention;

[0036] Figure 6This is a schematic diagram of the recognition results of the multi-channel fusion module in this invention. Detailed Implementation

[0037] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0038] This invention proposes an automatic arteriovenous wave identification system based on a multi-channel neck detector, specifically including a multi-channel PPG data acquisition device, a multi-channel signal preprocessing module, a multi-channel feature extraction module, a multi-channel data feature discrimination module, and a multi-channel fusion module based on the principle of maximum probability. Specific embodiments are as follows:

[0039] Multi-channel PPG data acquisition device:

[0040] A multi-channel neck detector acquires photoplethysmography (PPG) signals over a large area of ​​the subject's neck. The multiple channels are arranged in an array, ensuring that different channels naturally exhibit varying sensitivities to the carotid artery or jugular vein, due to their different locations, depths, and significant hemodynamic differences. The detector's specific design is as follows:

[0041] Detector Structure: Utilizing a flexible substrate, it fits snugly against the neck skin, ensuring stable signal acquisition. The detector employs one or more light sources and multiple photodetectors to form a multi-channel detection unit. The light source can be single-wavelength or multi-wavelength (e.g., red light, near-infrared light, wavelength range 400nm-1000nm). Multiple photodetectors are arranged around each light source, with a minimum distance of 20mm between the detectors and the light source. The multi-channel detection units are distributed in an array.

[0042] In this embodiment, as Figure 1 As shown, (a) illustrates the distribution structure of a single detection unit (light source and photodetector) used in this embodiment. The circular portion is the light-emitting device (LED), and the square portion is the photodetector (PD). Multiple detection units constitute a multi-channel acquisition array. The multi-channel neck detector uses a flexible FPC substrate with a thickness of 0.2 mm to ensure close contact with the neck skin. The overall size of the detector is 7cm × 7cm, covering a large area of ​​the neck. One light source (an LED with two wavelengths, 760nm and 850nm) is set up, and photodetectors (PDs) are arranged in a cross direction around the light source to form four detection channels. The distance between the photodetector and the light source is 25mm. A 0.5mm thick black silicone is pasted on one side of the light source and PD, and the other side is covered with black light-shielding tape to achieve light-shielding treatment. Medical pressure-sensitive adhesive is applied to the side of the detector that contacts the skin to form a double-sided adhesive layer to ensure stable contact.

[0043] Signal acquisition parameters: The signal sampling frequency is set to 100-500Hz, which can be adjusted according to actual monitoring needs. Multiple channels acquire signals simultaneously, with each channel independently obtaining the raw PPG signal for its corresponding detection area.

[0044] In this embodiment, healthy adult subjects were selected and kept relaxed, lying flat on a bed with their upper body angle adjusted to 15° to ensure fullness of the internal jugular vein. A multi-channel neck detector was applied to the right anterior side of the subject's neck, ensuring complete coverage of the densely vascularized area of ​​the neck. A data acquisition module was used to simultaneously acquire raw PPG signals from four channels at a sampling frequency of 200Hz for 20 minutes, and the data from each channel was stored in real time.

[0045] Furthermore, it should be noted that the structural form of the multi-channel neck detector described in this invention is not limited to the specific form described in the above embodiments. Without departing from the overall technical concept of this invention, the spatial arrangement, array structure, and external contour of the light source and photodetector can be adjusted and modified according to actual application requirements.

[0046] For example, such as Figure 1 As shown in (b), the detection unit can adopt a structure with the light source at the center and photosensitive detectors distributed radially or in a ring around it; for example... Figure 2 As shown in (a), the detector can adopt a regular array structure, with multiple sets of detection units arranged in a rectangular or grid pattern on a flexible substrate; and as shown in (a), the detector can adopt a regular array structure, with multiple sets of detection units arranged in a rectangular or grid pattern on a flexible substrate; Figure 2 As shown in (b), the detector can also be designed as a strip or linear arrangement, with multiple light sources and photosensitive detectors arranged along the anatomical structure of the neck.

[0047] Multi-channel signal preprocessing module:

[0048] The raw PPG signals acquired from multiple channels are preprocessed to eliminate noise interference and extract effective signal components for optimal performance. Preprocessing includes filtering, denoising, and period division. Figure 3 The complete process of multi-channel signal acquisition, noise reduction filtering, and period division is shown below:

[0049] Preferably, an infinite impulse response filter (such as a Butterworth filter) is used for filtering. The filter order is set to 4-8, and the cutoff frequency range is 0.5-10Hz. At the same time, a bidirectional filtering method can be used to avoid signal phase distortion and preserve the timing characteristics of the pulse signal.

[0050] As a preferred method, noise reduction can be achieved by using methods such as moving average filtering to remove baseline drift in the PPG signal, and by using a notch filter to remove power frequency interference.

[0051] As a preferred method, the period division involves extracting feature points (such as peaks and valleys) from the preprocessed PPG signal, determining the time interval between two adjacent feature points as a pulse cycle, and thus dividing the long-term signal into multiple independent single pulse cycle signals.

[0052] This embodiment specifically includes:

[0053] Preprocessing of the raw PPG signals from each channel:

[0054] Denoising was achieved by using a moving average filter to remove baseline drift, and a 50Hz notch filter was used to remove power frequency interference.

[0055] The filtering uses a fourth-order Butterworth bandpass filter with a cutoff frequency of 0.5-10Hz to obtain the AC components of each channel. The filtering process uses bidirectional filtering to avoid phase shift.

[0056] As a preferred method, the period division first multiplies the AC component of the light intensity signal by a negative one and inverts it to form a pressure wave signal. Then, the first-order difference between adjacent sampling points of the AC signal is calculated, and the position where the difference changes from negative to positive is taken as the candidate point of the valley value.

[0057] Unreasonable valley values ​​are filtered out using the interquartile range method (Q1-1.5×IQR is the lower threshold of the valley value, Q3+1.5×IQR is the upper threshold of the valley value, where Q1 is the lower quartile, Q3 is the upper quartile, and IQR is the interquartile range).

[0058] A search window is set near each candidate point (the window width is 0.5 times the sampling frequency points; for example, when the sampling frequency is 200Hz, the window width is 100 data points). Local smaller values ​​are searched as valley values. In order to ensure the consistency of period division, when there are multiple smaller values, the one with the smallest time sequence is selected as the minimum valley value to divide the period, thus ensuring the consistency of period division.

[0059] Multi-channel feature extraction module:

[0060] Physiologically, arterial waves originate from the heart's pumping action, have a steep rise, and are rich in harmonic energy; venous waves originate from atrial conduction, have a smoother waveform, and concentrated frequency domain energy. For the preprocessed signal of each channel, time-domain and frequency-domain feature extraction are performed. The measured characteristics of venous and arterial waves are as follows: Figure 4 As shown, (a) is the time-domain characteristic of the venous wave; (b) is the time-domain characteristic of the arterial wave; (c) is the power spectral density of the venous wave; and (d) is the power spectral density of the arterial wave.

[0061] The temporal feature extraction extracts temporal features within a single pulse cycle of each channel. The feature settings and extraction methods are as follows:

[0062] Time parameters: Calculate the rise time from the initial trough to the peak value, the fall time from the peak value to the final trough value, and the ratio of the two times for a single periodic signal. Also calculate the relative time of each peak and trough within the period.

[0063] Peak-valley parameters: Record the position and amplitude of the zero-crossing point of the first derivative in a single periodic signal, identify them as peak and valley values, and calculate the maximum peak-valley difference. Calculate the sum of the peak amplitude and the product of the peak amplitude and the valley amplitude and the product of the valley amplitude and the valley amplitude and the valley amplitude.

[0064] Waveform morphology parameters: Calculate the maximum rising slope and maximum falling slope of the extracted waveform.

[0065] The frequency domain feature extraction involves performing frequency domain analysis on signals from multiple consecutive pulse cycles in each channel to obtain frequency domain features, which are specifically as follows:

[0066] Dominant frequency characteristics: The peak value of the dominant frequency located near the heart rate in the frequency domain.

[0067] Harmonic characteristics: statistically analyze the peak values ​​of the spectrum outside the heart rate frequency, namely the amplitude and location of the second, third, and fourth harmonics.

[0068] Energy distribution characteristics: Calculate the total signal energy within the filter frequency band, and the ratio of the energy of each characteristic frequency band (main frequency band, second harmonic band, third harmonic band, fourth harmonic) to the total signal energy. Calculate the power weighting ratio within the four main frequency bands, i.e., the harmonic index HI, according to the weighting formula below.

[0069]

[0070] in The main frequency (usually heart rate) power, This is the second harmonic power. The third harmonic power, The power is the fourth harmonic. to For weighted parameters, satisfying .

[0071] The above features can be obtained through filtering, peak and valley detection, FFT energy analysis, interpolation fitting and other methods. The specific implementation is not limited to the technical means mentioned above.

[0072] In this embodiment: FFT (sampling points 4096) is performed on the signal of each channel for 20 consecutive cycles to obtain the power spectral density; the peak value of the dominant frequency (located in the 0.67-2Hz frequency band corresponding to the heart rate) is identified, and the amplitude and frequency of the dominant frequency are recorded; the amplitudes of the second harmonic (2 times the dominant frequency), the third harmonic (3 times the dominant frequency), and the fourth harmonic (4 times the dominant frequency) are statistically analyzed, and the ratio of the amplitude of each harmonic to the amplitude of the dominant frequency is calculated; the total signal energy in the 0.5-10Hz frequency band, as well as the ratio of the energy of the dominant frequency band (within ±0.1Hz), the second harmonic band (2 times the dominant frequency ±0.1Hz), and the third harmonic band (3 times the dominant frequency ±0.1Hz) to the total energy, are calculated, and the harmonic index HI is calculated.

[0073] Multi-channel data feature discrimination module:

[0074] like Figure 5 The illustrated process of feature extraction and multi-feature discrimination rules involves inputting the feature set of each channel into the discrimination module for independent classification of arterial and venous waves. Specific discrimination rules include:

[0075] Based on the correlation between features and arteriovenous waves, the system selects features with high discriminative power to establish the following feature discrimination logic, which discriminates features for each channel. The rule-based discrimination logic includes:

[0076] In the time domain, the ratio of the rise time to the fall time of the carotid artery wave is less than 1; the ratio of the rise time to the fall time of the jugular venous wave is greater than 1. The sum of the peak amplitude and the product of the peak relative time of the jugular venous wave, and the sum of the trough amplitude and the product of the trough relative time (negative values) are both greater than those of the carotid artery wave. The maximum rise slope of the carotid artery wave is higher than that of the jugular venous wave, while the maximum fall slope (absolute value) of the carotid artery wave is lower than that of the jugular venous wave.

[0077] In the frequency domain, the fourth harmonic energy of the carotid artery wave accounts for a larger proportion of the total signal energy than that of the jugular vein wave. The harmonic index of the carotid artery is greater than that of the jugular vein.

[0078] The channel discrimination module provides the probabilities of arterial and venous waves for each channel. By comparing the probabilities of multiple channels, the system selects the channel with the highest arterial wave probability as the source of the carotid artery wave and the channel with the highest venous wave probability as the source of the jugular venous wave from all the arterial and venous probabilities output by all channels. When two channels have the same probability, both channels are used as output.

[0079] Each judgment rule is quantified into a fixed probability, and the sum of the probabilities of all judgment rules is 1. In this embodiment, a probability weight is set for each rule, and the probability of each channel is calculated according to the following formula:

[0080]

[0081] in For the discrimination rules, This indicates that rule k determines channel i is more like an artery. This indicates that rule k determines channel i is more like a vein. As weight, .

[0082] The features and criteria used in the discrimination process can be adaptively adjusted based on individual physiological characteristics, or the robustness of the discrimination can be improved through a dynamic update mechanism (such as optimization based on historical discrimination results at regular intervals). The discrimination method is scalable, and discrimination rules can be added or adjusted according to actual application scenarios.

[0083] For example, in another embodiment, when identifying signals collected from the subject's neck, if the maximum upward slope of the arterial signal is greater than that of the venous signal in the identified 10-minute time period, but the relative difference percentage is lower than a preset threshold (set to 20% in this embodiment); and the maximum downward slope of the arterial signal is less than that of the venous signal, but the relative difference percentage is also lower than a preset threshold (set to 20% in this embodiment), then the discrimination of the above slope features in the current time period is considered to be reduced.

[0084] In this case, when calculating the arteriovenous probability of each channel after this period, the weights of the maximum upward slope and maximum downward slope rules should be appropriately reduced, while the weights of the remaining discrimination rules should be uniformly increased at the same time, so as to maintain the normalization constraint of each feature in the single-channel comprehensive discrimination.

[0085] Multi-channel fusion module based on the principle of maximizing probability:

[0086] By comparing the probability values ​​of arterial and venous waves output from multiple channels, the channel with the highest probability of arterial waves is selected as the source channel for arterial wave signals, and the channel with the highest probability of venous waves is selected as the source channel for venous wave signals.

[0087] like Figure 6 As shown, the single-source four-channel detector used in this embodiment acquired waveforms at different neck positions. The time-domain and frequency-domain characteristics of the corresponding channels were calculated using the method described above. Based on the discrimination rules, the rise-fall time ratio, the sum of the peak amplitude and peak relative time products, the sum of the trough amplitude and trough relative time products, the maximum rise slope, the maximum fall slope, the fourth harmonic energy proportion, and the harmonic index HI were used for discrimination. The rise-fall time ratio rule had a weight of 0.2, the fourth harmonic energy proportion rule had a weight of 0.2, the harmonic index rule had a weight of 0.2, and the weights of the other rules were 0.1. The recognition results were compared with the manually labeled results. The selected arterial and venous signal channels were consistent with the manually labeled results, demonstrating the effectiveness of this recognition system.

[0088] If a machine learning model (such as a random forest, support vector machine, or lightweight neural network) is used, the channel feature set can be used as input to train the model to probabilistically distinguish between arteries and veins. This machine learning model serves as an equivalent alternative to rule-based discrimination. In this invention, rule-based discrimination is the core approach; the machine learning model can be considered a further alternative and does not affect the core idea of ​​this invention.

[0089] This method utilizes the physiological and anatomical differences in the sensitivity of different channels to arteries or veins to achieve natural resistance to single-point deviations, individual differences, and noise.

[0090] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0091] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this application. This application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. An automatic arteriovenous wave identification system based on a multi-channel neck detector, characterized in that, The system includes: Multi-channel PPG data acquisition device: used to acquire the raw PPG signals of the corresponding detection area of ​​each channel; Multi-channel signal preprocessing module: used to filter, denoise, and periodize the acquired raw PPG signal; Multi-channel feature extraction module: Extracts time-domain and frequency-domain features from the preprocessed signal of each channel; the time-domain features include time parameters, peak-valley parameters, and waveform morphology parameters; the extraction of frequency-domain features includes calculating the dominant frequency feature, statistically analyzing harmonic features, and thus calculating the energy distribution features of each feature frequency; The multi-channel data feature discrimination module distinguishes between the time-domain and frequency-domain features of the carotid artery and vein, and obtains the probability of arterial and venous waves for each channel. The multi-channel fusion module based on the principle of maximizing probability selects the channel with the highest probability of arterial wave as the source of carotid artery wave and the channel with the highest probability of venous wave as the source of jugular venous wave.

2. The automatic arteriovenous wave identification system based on a multi-channel neck detector according to claim 1, characterized in that, The multi-channel PPG data acquisition device uses a flexible substrate that fits tightly against the skin of the neck, and forms multiple channels through several photosensitive detectors arranged in an array around the light source.

3. The automatic arteriovenous wave identification system based on a multi-channel neck detector according to claim 1, characterized in that, The denoising process uses a moving average filtering method to remove baseline drift in the PPG signal and a notch filter to remove power frequency interference.

4. The automatic arteriovenous wave identification system based on a multi-channel neck detector according to claim 1, characterized in that, The filtering process specifically involves using an infinite impulse response filter and employing bidirectional filtering to obtain the AC components of each channel.

5. The automatic arteriovenous wave identification system based on a multi-channel neck detector according to claim 1, characterized in that, The specific period division is as follows: The pressure wave signal is generated by multiplying the AC component of the filtered light intensity signal by a negative one and reversing it. The first-order difference between adjacent sampling points of the AC signal is calculated, and the position where the difference changes from negative to positive is taken as the candidate valley point. The interquartile range method is used to screen unreasonable valley values. A search window is set near each candidate point to search for local smaller values ​​as valley values. In order to ensure the consistency of period division, when there are multiple smaller values, the value with the smallest time sequence in the search window is selected as the minimum valley value to divide the period, thus ensuring the consistency of period division.

6. The automatic arteriovenous wave identification system based on a multi-channel neck detector according to claim 1, characterized in that, The time-domain features specifically include: Time parameters: Calculate the rise time from the beginning trough to the peak value, the fall time from the peak value to the end trough value, and the ratio of the two in a single period signal; at the same time, calculate the relative time of each peak and trough within the period. Peak-valley parameters: Record the position and amplitude of the zero-crossing point of the first derivative in a single periodic signal, identify it as the peak value and valley value, calculate the maximum peak-valley difference, and calculate the sum of the peak amplitude and the product of the peak value and the valley amplitude and the product of the valley value and the valley value and the time. Waveform morphology parameters: Calculate the maximum rising slope and maximum falling slope of the extracted waveform.

7. The automatic arteriovenous wave identification system based on a multi-channel neck detector according to claim 1, characterized in that, The process of obtaining the energy distribution characteristics specifically includes: Calculate the peak frequency of the dominant frequency located near the heart rate in the frequency domain; Statistical analysis of spectral peaks other than heart rate frequency, namely the amplitude and location of the second, third, and fourth harmonics; Calculate the total signal energy within the filtered frequency band, and the ratio of the energy of each characteristic frequency band to the total signal energy. Calculate the power weighted ratio within the main frequency band as the harmonic index.

8. The automatic arteriovenous wave identification system based on a multi-channel neck detector according to claim 6, characterized in that, In the multi-channel data feature discrimination module, the discrimination of time-domain features includes: the ratio of carotid artery wave rise time to fall time is less than 1; the ratio of carotid venous wave rise time to fall time is greater than 1; the sum of the peak amplitude and the product of peak relative time of the carotid venous wave, and the sum of the trough amplitude and the product of trough relative time of the trough are both greater than those of the carotid artery wave; the maximum rise slope of the carotid artery wave is higher than that of the carotid venous wave, while the absolute value of the maximum fall slope of the carotid artery wave is lower than that of the carotid venous wave.

9. The automatic arteriovenous wave identification system based on a multi-channel neck detector according to claim 7, characterized in that, In the multi-channel data feature discrimination module, the discrimination of frequency domain features includes: the proportion of harmonic energy of the carotid artery wave to the total signal energy is greater than that of the jugular vein wave; the harmonic index of the carotid artery is greater than that of the jugular vein.

10. The automatic arteriovenous wave identification system based on a multi-channel neck detector according to claim 1, characterized in that, The probabilities of arterial and venous waves in each channel are obtained by weighted summation of the results obtained from each discrimination rule in that channel. The weights of the maximum upward slope and maximum downward slope rules are adjusted and reduced according to the subjects, while the weights of the remaining discrimination rules are uniformly increased.

Images