Continuous blood pressure signal quality evaluation method, device, equipment and medium

Abstract

The invention discloses a continuous blood pressure signal quality evaluation method and device, equipment and a medium, and relates to the technical field of signal monitoring. The method comprises the steps that PPG signals and ACC signals synchronously collected by a local sensor are obtained; calculating a cross-correlation characteristic between the PPG signal and the ACC signal after preprocessing; constructing an artifact intensity index based on the cross-correlation characteristics, the preprocessed PPG signal and the preprocessed ACC signal; inputting the PPG signal, the ACC signal and the artifact intensity index into a preset quality evaluation model, performing fusion analysis on the PPG signal and the ACC signal based on the artifact intensity index to judge whether the PPG signal is used for blood pressure estimation or not, and outputting a corresponding quality evaluation result. According to the method, artifacts, pulled by actions, of waveforms are recognized through cross-modal PPG-ACC cross-correlation, heart rhythm stability and artifact intensity indexes are captured to provide artifact priori, complex artifacts can be accurately recognized, and quality evaluation of blood pressure signals can be achieved through low-cost wearable equipment.

Description

Technical Field

[0001] This invention relates to the field of signal monitoring technology, and in particular to a method, apparatus, equipment and medium for continuous blood pressure signal quality assessment. Background Technology

[0002] Photoplethysmo (PPG) is widely used in wearable devices for monitoring vital signs such as heart rate, blood oxygen, and blood pressure due to its simple hardware and low power consumption. However, PPG signals are susceptible to factors such as wearing tightness, ambient light, changes in contact pressure, and wrist movement. Especially in dynamic scenarios, signal quality deteriorates sharply, severely impacting the accuracy of continuous blood pressure estimation. Unlike high-cost wearable devices that utilize multiple optical paths or wavelengths, low-cost wearable devices mostly use only a single green LED (Light Emitting Diode), a single photodetector (PD), and a single analog front end (AFE). This structure in low-cost wearable devices is extremely sensitive to noise and motion, and traditional quality assessment methods are ineffective in real-world scenarios.

[0003] Current traditional quality assessment methods include fixed threshold detection based on accelerometers and simple rules based on signal-to-noise ratio or amplitude changes. However, they perform poorly in scenarios such as micro-tremor artifacts caused by typing, periodic movements caused by going up and down stairs, wrist muscle tension or sudden changes in contact pressure, and overall waveform drift caused by rapid arm swing. These complex artifacts cause the misjudgment rate of simple rule-based methods to soar, directly leading to a blood pressure estimation deviation of 10-20 mmHg. Continuous blood pressure estimation requires the extraction of features such as peak values, trough values, rising slope, reflected wave / dicrotic notch, and PPG pulse wave morphology indicators. Any morphological distortion will lead to distortion of the blood pressure model output. Therefore, signal quality assessment is essential in continuous blood pressure technology. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for continuous blood pressure signal quality assessment, capable of accurately judging waveform drift of PPG signals in motion scenarios and obtaining high-quality PPG signal quality assessment results. The specific solution is as follows:

[0005] In a first aspect, this application discloses a method for assessing the quality of continuous blood pressure signals, applied to wearable devices, comprising:

[0006] Acquire PPG and ACC signals synchronously collected by local sensors;

[0007] Calculate the cross-correlation characteristics between the preprocessed PPG signal and the ACC signal;

[0008] Based on the cross-correlation characteristics, the preprocessed PPG signal, and the ACC signal, an artifact intensity index is constructed.

[0009] The PPG signal, the ACC signal, and the artifact intensity index are input into a preset quality assessment model to perform a fusion analysis on the PPG signal and the ACC signal based on the artifact intensity index, so as to determine whether the PPG signal is used for blood pressure estimation and output the corresponding quality assessment results.

[0010] Optionally, after acquiring the PPG signal and ACC signal synchronously collected by the local sensor, the method further includes:

[0011] Write the PPG signal and the ACC signal into a circular buffer;

[0012] The PPG signal is extracted from the circular buffer and filtered to extract periodic consistency features and signal-to-noise ratio, so as to obtain the preprocessed PPG signal.

[0013] Optionally, the calculation of the cross-correlation characteristics between the preprocessed PPG signal and the ACC signal includes:

[0014] Within a preset time window, the maximum cross-correlation value of the preprocessed PPG signal and the ACC signal under multiple time delays is calculated to obtain the cross-correlation characteristics.

[0015] Optionally, the step of constructing the artifact intensity index based on the cross-correlation features, the preprocessed PPG signal, and the ACC signal includes:

[0016] The artifact intensity index is obtained by weighted summation based on the cross-correlation characteristics, the energy value of the ACC signal, the periodicity characteristics, and the reciprocal of the signal-to-noise ratio.

[0017] Optionally, the step of fusing and analyzing the PPG signal and the ACC signal based on the artifact intensity index to determine whether the PPG signal is used for blood pressure estimation and outputting the corresponding quality assessment result includes:

[0018] The convolutional neural network layer of the preset quality assessment model is used to perform convolution operations on the waveforms of the preprocessed PPG signal and the ACC signal to extract local morphological features of the PPG signal, including the slope of the rising edge of the pulse, the shape of the diabetic notch, and the sharpness of the peak.

[0019] The local morphological features are subjected to dilated convolution and residual connection operations through the temporal convolutional network layer of the preset quality assessment model to capture the contextual features of the PPG signal.

[0020] The contextual features are fused with the artifact intensity index, and a quality assessment result is generated by using the fully connected layer and activation function of the preset quality assessment model to determine whether the PPG signal is used for blood pressure estimation.

[0021] Optionally, PPG signals can be acquired, including:

[0022] A single PPG signal is acquired through reflection from a single green light source.

[0023] Optionally, the continuous blood pressure signal quality assessment method further includes:

[0024] If the quality assessment result of the current time window determines that the PPG signal is not used for blood pressure estimation, then a user prompt is triggered and the current continuous blood pressure estimation process is interrupted.

[0025] If PPG signals from multiple consecutive time windows are determined to be unsuitable for blood pressure estimation, a device wearing status check prompt message will be generated.

[0026] Secondly, this application discloses a continuous blood pressure signal quality assessment device for wearable devices, comprising:

[0027] The signal acquisition module is used to acquire the PPG and ACC signals synchronously collected by the local sensors.

[0028] The feature calculation module is used to calculate the cross-correlation features between the preprocessed PPG signal and the ACC signal;

[0029] An index construction module is used to construct an artifact intensity index based on the cross-correlation characteristics, the preprocessed PPG signal, and the ACC signal.

[0030] The quality assessment module is used to input the PPG signal, the ACC signal and the artifact intensity index into a preset quality assessment model, and to perform fusion analysis on the PPG signal and the ACC signal based on the artifact intensity index to determine whether the PPG signal is used for blood pressure estimation, and output the corresponding quality assessment results.

[0031] Thirdly, this application discloses an electronic device, including:

[0032] Memory, used to store computer programs;

[0033] A processor is configured to execute the computer program to implement the steps of the aforementioned disclosed continuous blood pressure signal quality assessment method.

[0034] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed continuous blood pressure signal quality assessment method.

[0035] As can be seen, this application discloses the following: acquiring PPG and ACC signals synchronously collected by a local sensor; calculating the cross-correlation characteristics between the preprocessed PPG signal and the ACC signal; constructing an artifact intensity index based on the cross-correlation characteristics, the preprocessed PPG signal, and the ACC signal; inputting the PPG signal, the ACC signal, and the artifact intensity index into a preset quality assessment model to perform fusion analysis on the PPG signal and the ACC signal based on the artifact intensity index, to determine whether the PPG signal is suitable for blood pressure estimation, and outputting the corresponding quality assessment result. Therefore, by identifying motion-induced artifacts in the waveform through cross-modal PPG-ACC cross-correlation, capturing cardiac rhythm stability, and using the artifact intensity index to provide artifact priors, complex artifacts can be accurately identified. This significantly improves the usable signal recognition rate, significantly reduces the artifact false reception rate, and enables quality assessment of blood pressure signals even with low-cost wearable devices. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0037] Figure 1 This is a flowchart of a continuous blood pressure signal quality assessment method disclosed in this application;

[0038] Figure 2 This is a time synchronization diagram of the synchronous acquisition of PPG signals and ACC signals disclosed in this application;

[0039] Figure 3 This is a diagram illustrating a signal preprocessing procedure disclosed in this application;

[0040] Figure 4 This is a diagram illustrating the process of constructing an artifact intensity index as disclosed in this application;

[0041] Figure 5 This is a schematic diagram of a pre-defined quality assessment model structure disclosed in this application;

[0042] Figure 6 This is a schematic diagram of the structure of a continuous blood pressure signal quality assessment device disclosed in this application;

[0043] Figure 7 This is a schematic diagram of a specific continuous blood pressure signal quality assessment device disclosed in this application;

[0044] Figure 8 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0045] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0046] Photoplethysmography (PPG) is widely used in wearable devices for monitoring vital signs such as heart rate, blood oxygen, and blood pressure due to its simple hardware and low power consumption. However, PPG signals are easily affected by factors such as wearing tightness, ambient light, changes in contact pressure, and wrist movement. Especially in dynamic scenarios, signal quality deteriorates sharply, severely impacting the accuracy of continuous blood pressure estimation. Unlike high-cost wearable devices that use multiple optical paths or wavelengths, low-cost wearable devices mostly use only a single green LED, a single photodetector (PD), and a single analog front-end (AFE). This structure of low-cost wearable devices is extremely sensitive to noise and motion, and traditional quality assessment methods cannot effectively address real-world scenarios.

[0047] Current traditional quality assessment methods include fixed threshold detection based on accelerometers and simple rules based on signal-to-noise ratio or amplitude changes. However, they perform poorly in scenarios such as micro-tremor artifacts caused by typing, periodic movements caused by going up and down stairs, wrist muscle tension or sudden changes in contact pressure, and overall waveform drift caused by rapid arm swing. These complex artifacts cause the misjudgment rate of simple rule-based methods to soar, directly leading to a blood pressure estimation deviation of 10-20 mmHg. Continuous blood pressure estimation requires the extraction of features such as peak values, trough values, rising slope, reflected wave / dicrotic notch, and PPG pulse wave morphology indicators. Any morphological distortion will lead to distortion of the blood pressure model output. Therefore, signal quality assessment is essential in continuous blood pressure technology.

[0048] Therefore, this invention provides a continuous blood pressure signal quality assessment scheme that can accurately determine the waveform drift of PPG signals in motion scenarios and obtain high-quality PPG signal quality assessment results.

[0049] like Figure 1 As shown, this invention discloses a method for assessing the quality of continuous blood pressure signals, applied to wearable devices, comprising:

[0050] Step S11: Acquire the PPG signal and ACC signal synchronously collected by the local sensor.

[0051] In this embodiment, a single-channel PPG signal formed by reflection from a single green light source is acquired. For example... Figure 2 As shown, the MCU (Microcontroller Unit) sets the drive current (1-10mA) through the LED driver, operating in pulse PWM (Pulse Width Modulation) mode to illuminate a 530-560nm green light source (20-30% duty cycle) to generate a modulated light signal that illuminates the skin area in contact with the wearable device, while reducing power consumption and skin thermal effects. Then, the photodetector (PD) receives the reflected or transmitted light signal (with a response band of 500-600nm) and converts the light intensity into a weak current signal (nA-μA level), completing the first conversion from light to electrical signal. Further, the current signal enters the analog front-end (AFE) for transimpedance amplification, converting the current signal into a voltage signal (gain 10k-100k), followed by hardware filtering. Specifically, it performs a high-pass filter (cutoff frequency 0.3-0.5Hz, to remove baseline drift) and an anti-aliasing low-pass filter (cutoff frequency <10Hz). This process amplifies the weak signal to a suitable sampling amplitude and initially suppresses key noise. The conditioned analog voltage signal, obtained after signal amplification and noise suppression, is then digitized using an ADC converter (12 or 16-bit resolution). The sampling rate can be configured to 25-250Hz. The digitized PPG signal is then directly written to a circular buffer via DMA (Data Memory Access). This yields a digitized PPG sequence, and the DMA and circular buffer mechanism ensure a continuous, non-loss-prone real-time data stream without consuming excessive CPU (Central Processing Unit) resources.

[0052] Meanwhile, during the synchronous acquisition of ACC (Accelerometer) signals, the sampling rate of the triaxial MEMS (Micro Electro Mechanical Systems) accelerometer is first set via the MCU. The MCU configures the operating parameters of the triaxial MEMS accelerometer, specifically the sampling rate and range, through the I²C (I-squared-C) or SPI (Serial Peripheral Interface) digital bus. To achieve synchronous acquisition, the sampling rate is set to match the PPG sampling rate, ranging from 25-250Hz. The range is selected based on the activity intensity of the wearable device, typically ±2g to ±8g. This ensures the sensor is ready and that its data output rate is on the same order of magnitude as the PPG acquisition. The specific acquisition steps include: the accelerometer is typically configured to generate a hardware interrupt when data is ready. In response to this interrupt, the MCU immediately reads the instantaneous digital acceleration values ​​of the three axes (X, Y, Z) via the I²C / SPI bus. Using interrupt-driven processing instead of polling ensures data can be read instantly, reducing latency and enabling high-precision time synchronization. The MCU's system clock adds a high-precision timestamp the data byte at the moment it reads it. Subsequently, the timestamped triaxial data is written to a circular buffer set for ACC. Marking each ACC data point with an absolute time provides the basis for millisecond-level alignment with PPG data. The acquisition of the ACC signal provides the wrist movement direction, amplitude, and frequency band, serving as the physical basis for artifact detection.

[0053] In this embodiment, the PPG signal and the ACC signal are written into a circular buffer; the PPG signal is extracted from the circular buffer and filtered to extract periodic consistency features and signal-to-noise ratio, resulting in a preprocessed PPG signal. It is understood that the circular buffer ensures that high-speed, continuous data streams are not lost and allows the ACC signal to match the PPG data buffering mechanism. After writing, the ACC signal and PPG signal are time-aligned. Specifically, the system resamples or interpolates the two signals based on the MCU timestamps carried by the PPG and ACC data respectively, ensuring they are synchronized on the time axis with a time error of <5ms. In this way, only after synchronization is achieved can the deformation of the PPG waveform be correlated with the instant of motion that caused the deformation.

[0054] like Figure 3 The signal preprocessing process shown is implemented in the MCU firmware and directly applies to the PPG / ACC sequence. The specific preprocessing steps for the original PPG signal include:

[0055] Debiasing: First-order IIR high-pass filter, cutoff frequency 0.3-0.5Hz, or sliding mean differential (window 0.5-1 second).

[0056] 0.5-5Hz bandpass filter: second-order or fourth-order IIR Butterworth filter, preserving the pulse band and suppressing baseline drift and high-frequency noise;

[0057] Normalization: The PPG signal after bandpass filtering is normalized to obtain the preprocessed PPG signal.

[0058] Then, feature extraction processing is performed on the preprocessed PPG signal, specifically including:

[0059] Peak detection: First-order derivative is used to find the zero-crossing point, second-order derivative is used to determine the peak shape, and peak width and peak amplitude are dynamically limited by thresholds.

[0060] Cycle consistency analysis: CV_cycle=σ(T_pp) / μ(T_pp), where T_pp is the peak-to-peak spacing.

[0061] Signal-to-noise ratio (SNR) estimation: SNR = E(0.5-3Hz) / E(>5Hz), calculated based on FFT or energy ratio.

[0062] The raw ACC signal is also preprocessed, and the preprocessing steps include:

[0063] ACC energy index extraction: triaxial module length The 1-10Hz bandpass filter is used to obtain the main frequency band of motion artifacts. The energy E_acc=Σa(t)² is calculated by filtering the 1-10Hz bandpass filter. High energy indicates large wrist movements.

[0064] In this way, the preprocessing module transforms complex real-time optical signals into structured features, making model inference more stable.

[0065] Step S12: Calculate the cross-correlation characteristics between the preprocessed PPG signal and the ACC signal.

[0066] In this embodiment, within a preset time window, the maximum cross-correlation value of the preprocessed PPG signal and the ACC signal under multiple time delays is calculated to obtain the cross-correlation characteristics. For example... Figure 4 As shown, the cross-correlation information is obtained by calculating the PPG-ACC cross-modal cross-correlation equation.

[0067] ;

[0068] in, This is represented as the cross-modal cross-correlation characteristic of PPG-ACC. Indicates the preset time window. This represents the preprocessed PPG signal at time t. Represented as The preprocessed ACC signal at the given time.

[0069] In this way, the PPG-ACC cross-modal cross-correlation equation is used to detect waveform artifacts caused by motion. If the PPG waveform deforms synchronously with wrist movement, the cross-correlation is high, and close to 0 at rest.

[0070] Step S13: Construct an artifact intensity index based on the cross-correlation features, the preprocessed PPG signal, and the ACC signal.

[0071] In this embodiment, the artifact intensity index is obtained by weighted summation based on the cross-correlation feature, the energy value of the ACC signal, the period consistency feature, and the reciprocal of the signal-to-noise ratio. It can be understood that the artifact intensity index is a comprehensive indicator of the cross-correlation feature, the energy value of the ACC signal, the period consistency feature, and the reciprocal of the signal-to-noise ratio, specifically expressed as follows:

[0072] ;

[0073] in, These are weighting coefficients fixed at the device end. This refers to the accelerometer energy, specifically the energy value of the ACC signal. This is the periodic variation coefficient, also known as the periodic consistency characteristic. This is the reciprocal of the signal-to-noise ratio, thus forming a single-value artifact index, providing a global noise prior for CNN / TCN.

[0074] Step S14: Input the PPG signal, the ACC signal and the artifact intensity index into a preset quality assessment model to perform fusion analysis on the PPG signal and the ACC signal based on the artifact intensity index, so as to determine whether the PPG signal is used for blood pressure estimation and output the corresponding quality assessment result.

[0075] In this embodiment, the waveforms of the preprocessed PPG signal and the ACC signal are convolved using the convolutional neural network layer of the preset quality assessment model to extract local morphological features of the PPG signal, including the slope of the pulse rise edge, the shape of the dicrotic notch, and the sharpness of the peak. The local morphological features are then subjected to dilated convolution and residual connection operations using the temporal convolutional network layer of the preset quality assessment model to capture the contextual features of the PPG signal. These contextual features are then fused with the artifact intensity index, and a quality assessment result for determining whether the PPG signal is suitable for blood pressure estimation is generated using the fully connected layer and activation function of the preset quality assessment model. Figure 5As shown, a pre-defined quality assessment model is first constructed and trained. This model consists of a three-stage structure: Stage 1 is a CNN (Convolutional Neural Network) for local morphological extraction; Stage 2 is a TCN (Temporal Convolutional Network) for temporal dependency modeling; and Stage 3 is a classification prediction layer for predicting and outputting classification results. Specifically, the evaluation process of the lightweight embedded model running in real-time on the MCU is as follows: It receives three types of structured inputs within a fixed time window (2-4 seconds): PPG sequence: preprocessed single-channel one-dimensional temporal data; ACC sequence: timestamp-aligned three-axis acceleration one-dimensional temporal data (3 channels); and handcrafted features: scalar features including artifact intensity index, signal-to-noise ratio, and periodicity consistency. This combines the original waveform with advanced prior knowledge (artifact index) to provide comprehensive information for the model. Then, the CNN layer performs local waveform morphology feature extraction. Specifically, the model first uses one-dimensional convolution (Conv1D) to scan the waveforms of PPG and ACC respectively. The first convolution (e.g., kernel=5) captures a wider range of morphological patterns, and the second convolution (e.g., kernel=3) focuses on local details. Each convolution is followed by batch normalization and ReLU activation function to accelerate training stability and introduce non-linearity. Finally, the CNN layer outputs a set of feature maps representing local morphological features, such as local patterns like "suspected dicrotic notch" and "rising slope" extracted from PPG, and "moment of motion abrupt change" extracted from ACC. It is important to note that the peak-to-peak spacing variability: PP_jitter = CV_cycle, is used to reflect the periodic disorder information caused by motion. Dicrotic notch integrity: The presence of the notch is detected using differentiation / slope. The presence of the notch indicates normal morphology, while the disappearance of the notch may indicate motion or pressure artifacts. Baseline drift amplitude: The rate of change of the DC component of PPG is detected, and the z-axis ACC is closely related to the DC drift. Then, the TCN's temporal dependency modeling step is entered. Specifically, the local features extracted by the CNN are input into the TCN temporal convolutional network. The TCN uses dilated convolutions (e.g., dilation=1, 2) to exponentially expand the receptive field without increasing the parameters, allowing observation of a longer time range. Residual connections ensure effective gradient propagation, prevent deep network degradation, and output high-level temporal features with long-range contextual information.This is used to determine whether the periodicity and rhythm of local features (such as a peak) are regular within a few seconds, thereby identifying artifacts where "individual waveforms are normal but the overall rhythm is chaotic." Then, it enters the third stage, a fully connected layer, for multimodal feature fusion and decision-making. Specifically, the high-level temporal features output by the TCN are globally averaged and pooled, compressed into a fixed-length feature vector, and then concatenated with hand-crafted features (especially the artifact index). The resulting composite feature vector is then passed through one or more fully connected layers. Here, the artifact index acts as a strong prior, influencing the decision weights. For example, a high artifact index might lead the model to favor a "not usable" judgment. Finally, a scalar value between 0 and 1 is output through the Sigmoid activation function. The binary classification result is output: typically with a threshold of 0.5, outputting 1 (usable) or 0 (not usable), where the Sigmoid output value itself can be used as a confidence score.

[0076] It's important to note that the model's operation is triggered within a real-time scheduling system. Specifically, the system reads data for a prepared time window from a double-buffered circular buffer approximately every 1 second. Classification is performed within the MCU's idle tasks or low-priority interrupts, and after obtaining the output results (0 / 1 and confidence level), the output results are passed to the output control module. Throughout the inference process, sampling interrupts (writing new data) always have the highest priority and are not blocked, thus ensuring no data loss.

[0077] In this embodiment, if the quality assessment result of the current time window determines that the PPG signal is not suitable for blood pressure estimation, a user prompt is triggered and the current continuous blood pressure estimation process is interrupted. If the PPG signals of multiple consecutive time windows are all determined to be unsuitable for blood pressure estimation, a device wearing status check prompt is generated. It is understood that when the model output determines that the blood pressure signal quality assessment result under the current time window is unusable, new prompts such as "Please remain still / re-measure" are displayed on the wearable device. If the wearable device has a vibration motor, it vibrates to remind the user, and the subsequent processing of blood pressure / heart rate variability is interrupted (to ensure safety). Furthermore, if the wearable device is connected to a mobile device such as a smartphone, relevant prompts are pushed through the wearable device app on the mobile device. When the quality assessment results of multiple consecutive time windows are all determined to be unusable, the user is reminded to check if the device is loose, and the LED driver is controlled to reduce brightness and power consumption.

[0078] As can be seen, this application discloses the following: acquiring PPG and ACC signals synchronously collected by a local sensor; calculating the cross-correlation characteristics between the preprocessed PPG signal and the ACC signal; constructing an artifact intensity index based on the cross-correlation characteristics, the preprocessed PPG signal, and the ACC signal; inputting the PPG signal, the ACC signal, and the artifact intensity index into a preset quality assessment model to perform fusion analysis on the PPG signal and the ACC signal based on the artifact intensity index, to determine whether the PPG signal is suitable for blood pressure estimation, and outputting the corresponding quality assessment result. Therefore, by identifying motion-induced artifacts in the waveform through cross-modal PPG-ACC cross-correlation, capturing cardiac rhythm stability, and using the artifact intensity index to provide artifact priors, complex artifacts can be accurately identified. This significantly improves the usable signal recognition rate, significantly reduces the artifact false reception rate, and enables quality assessment of blood pressure signals even with low-cost wearable devices.

[0079] like Figure 6 As shown, the present invention also discloses a continuous blood pressure signal quality assessment device for wearable devices, comprising:

[0080] Signal acquisition module 11 is used to acquire the PPG signal and ACC signal synchronously acquired by the local sensor;

[0081] Feature calculation module 12 is used to calculate the cross-correlation features between the preprocessed PPG signal and the ACC signal;

[0082] Index construction module 13 is used to construct an artifact intensity index based on the cross-correlation characteristics, the preprocessed PPG signal, and the ACC signal;

[0083] The quality assessment module 14 is used to input the PPG signal, the ACC signal and the artifact intensity index into a preset quality assessment model, and to perform fusion analysis on the PPG signal and the ACC signal based on the artifact intensity index, so as to determine whether the PPG signal is used for blood pressure estimation and output the corresponding quality assessment result.

[0084] like Figure 7 As shown, each module works collaboratively with an MCU as its core through an internal bus and interrupt mechanism. Specifically, the signal acquisition module 11 includes a PPG optical acquisition module, which specifically includes:

[0085] Green LED light source: wavelength 530-560nm, MCU sets the drive current (1-10mA) through LED driver, working mode is pulse lighting (20-30% duty cycle, reducing power consumption and skin heating).

[0086] Photodetector (PD): Response band 500-600nm, output is a weak current signal (nA-µA level).

[0087] Analog front end (AFE): Transimpedance amplification gain 10k-100k, hardware high-pass filter (0.3-0.5Hz), anti-aliasing low-pass filter (<10Hz).

[0088] ADC converter: 12 / 16-bit resolution, configurable sampling rate 25-250Hz, using the MCU ADC channel and DMA to continuously write sampled data to a circular buffer. The sampling frequency of 25-250Hz is chosen because the basic pattern can be identified at 25Hz or higher, the diapohyet features can be better distinguished at 64Hz or higher, and 250Hz or lower meets the processing capabilities of low-power devices.

[0089] The ACC sampling frequency is 25-250Hz: 1-10Hz is the main frequency band for motion artifacts, and it is sampled synchronously with PPG (MCU timestamp).

[0090] Light intensity changes are converted into digital PPG sequences, providing the basic input for subsequent DSP (Digital Signal Processing) and CNN / TCN models.

[0091] Signal acquisition module 11 includes an ACC acquisition module, which specifically includes:

[0092] Triaxial MEMS accelerometer: sampling rate 25-250Hz, synchronized with PPG; measurement range ±2g to ±8g; outputs x, y, z axis acceleration;

[0093] Data synchronization mechanism: ACC interrupt drives sampling, each sampling is accompanied by an MCU timestamp, and the MCU writes the ACC data into the ACC circular buffer in time sequence to ensure synchronization and alignment with PPG (error <5ms).

[0094] Providing the direction, amplitude, and frequency of wrist movement is the physical basis for artifact detection.

[0095] Cross-modal correlation, based on synchronized sampling data of PPG and ACC from specific hardware, detects motion coupling distortion (whose signal characteristics are strongly correlated with optical / electrical / spatial / device characteristics, such as optical path / structure / op-amp) generated by the hardware structure of a specific wearable device. Motion coupling distortion can specifically manifest as: wrist swing (x / y axis) causing overall waveform shift and periodic jitter, making peaks undetectable; changes in contact pressure (z axis) causing DC drift and waveform flattening, deforming the rising edge; ambient light causing increased high-frequency noise, leading to a decrease in SNR; and muscle tension causing morphological fragmentation, resulting in the disappearance of the dicrotic notch. All these noises are physically coupled to PPG-ACC; therefore, this invention uses cross-modal correlation as an indicator.

[0096] The MCU main control module includes a feature calculation module 12, which further includes a signal preprocessing submodule. This submodule is implemented in the MCU firmware and directly operates on the PPG / ACC sequence, including:

[0097] DC Removal: First-order IIR high-pass filter with a cutoff frequency of 0.3-0.5Hz, or sliding mean difference (window length 0.5-1 second). This method uses a first-order IIR high-pass filter or a sliding window mean difference (length 0.5-1.0 seconds) to remove long-period DC drift and avoid interference peak detection.

[0098] 0.5-5Hz bandpass filtering: Second-order or fourth-order IIR Butterworth filter, preserving the pulse frequency band and suppressing baseline drift and high-frequency noise. In this way, using second-order / fourth-order IIR Butterworth or low-order FIR (saving MCU computation) filters out low-frequency drift less than 0.5Hz and high-frequency noise greater than 5Hz, ensuring that the frequency band containing blood pressure / heart rate variability characteristics is preserved.

[0099] Peak detection: First-order derivative is used to find zero-crossing points, second-order derivative is used to determine peak shape, and peak width and amplitude are dynamically limited by thresholds. In this way, a combined judgment is adopted: first-order derivative zero-crossing, second-order derivative to determine peak sharpness, local maxima filtering, and shape constraints (peak width and spacing) are applied to output peak time series and peak spacing (T_pp).

[0100] Periodic consistency analysis: CV_cycle = σ(T_pp) / μ(T_pp), where T_pp is the peak-to-peak spacing. CV_cycle < 0.1 indicates high stability (usable), CV_cycle between 0.1 and 0.25 indicates moderate stability (determined by combining the artifact index), and CV_cycle > 0.25 indicates unstable stability (unusable).

[0101] Signal-to-noise ratio (SNR) estimation: SNR = E(0.5-3Hz) / E(>5Hz), calculated based on FFT or energy ratio. An SNR greater than 10dB is considered usable, an SNR between 6-10dB is considered gray (depending on CNN / TCN), and an SNR less than 6dB is considered unusable.

[0102] ACC motion energy extraction: three-axis module length A 1-10Hz bandpass filter is used to obtain the main frequency band of motion artifacts. An energy threshold is set; when the energy is greater than the threshold, it is considered high energy, indicating strong motion; when the energy is less than the threshold, it is considered low energy, indicating stillness. The main motion frequency band is obtained through a 1-10Hz bandpass filter; the higher the energy of the frequency band, the more severe the waveform distortion.

[0103] Transforming complex real-time optical signals into structured features makes model inference more stable.

[0104] Feature calculation module 12 extracts the correlation features between PPG and ACC for further construction of the artifact intensity index. Feature extraction includes ACC energy index extraction, specifically calculating the energy through a 1-10Hz bandpass filter; high energy indicates large wrist movements.

[0105] The feature calculation module 12 further extracts cross-correlation features. Specifically, it uses the cross-correlation feature calculation formula to extract cross-correlation features, which are used to detect waveform artifacts caused by motion. If the PPG waveform deforms synchronously with wrist movement, the cross-correlation is high, and close to 0 at rest. This is a capability that traditional signal quality assessment techniques do not possess.

[0106] The feature extraction process also includes the extraction of the following features of PPG morphological disruption indicators: peak-to-peak spacing variability: PP_jitter = CV_cycle, reflecting periodic disruption caused by motion; diabetic notch integrity: the presence of the notch is detected using differential / slope; the presence of the notch indicates normal morphology, while its disappearance may indicate motion or pressure artifacts; baseline drift amplitude: detecting the rate of change of the DC component of PPG, with the z-axis ACC closely related to DC drift. These features further reflect the intensity of artifacts.

[0107] The artifact intensity index is constructed by the index construction module 13 based on the features extracted above. The constructed artifact intensity index forms a single-value artifact index, which provides a global noise prior for CNN / TCN.

[0108] The quality assessment module 14 uses a lightweight embedded model that runs in real time on the MCU.

[0109] Input: PPG: 2-4 seconds, 25-250Hz sampling; ACC: three-axis, timestamp aligned; artificial features: SNR, CV, artifact index.

[0110] Model structure: Stage 1 - Local morphology extraction (CNN): Conv1D (kernel=5, channel=16), Conv1D (kernel=3, channel=32), batch normalization + ReLU activation, to extract the rising edge of the pulse, dichroic notch, valley morphology, and waveform sharpness;

[0111] Phase 2 - Time Dependency Modeling (TCN): 2-layer TCN (dilation=1,2) and residual connections to capture periodic patterns, rhythm stability, and long-range dependencies.

[0112] Stage 3 - Classification: Global average pooling, fully connected layer, Sigmoid output 0 / 1.

[0113] Output: 1 = Usable (stable morphology, high SNR, low artifact index); 0 = Unusable (motion, pressure changes, morphological disruption). A confidence score (0-1) can also be output. Specifically, in typing scenarios (micro-tremor artifacts): if the ACC high-frequency energy is moderate and the PPG-ACC correlation is strong, the model is considered unusable. In walking / running scenarios (periodic oscillations): if the ACC energy is high and the PPG periodicity is significantly disrupted, the model is considered unusable. In resting-state measurement scenarios: if the SNR is high, CV is low, and ACC energy is low, the model is considered usable. In sleep scenarios: if occasional turning over is short-term, the model is unusable; if the resting period is usable. In scenarios with unstable contact pressure: if the z-axis ACC changes significantly and the PPG baseline drifts significantly, the model is considered unusable.

[0114] The output module, when the model is deemed unusable, displays "Please remain still / Remeasure" on the wearable device; if the device has a vibration motor, it vibrates as a reminder; it interrupts subsequent processing of blood pressure / heart rate variability (to ensure safety); and if connected to a mobile phone, it pushes a notification. If the window determines the model is unusable N times consecutively, it reminds the user to check if the device is loose and controls the LED driver to reduce brightness and power consumption.

[0115] Storage module: used for storing model weights, sliding window storage (2-4 seconds), and DSP intermediate state storage.

[0116] Communication module: Bluetooth / Wi-Fi / NFC is used to upload signal quality level (available / unavailable) and confidence level to mobile app or cloud.

[0117] Furthermore, the overall operation process of the device is as follows: the device runs a quality assessment task in 1-second cycles. Both PPG and ACC are sampled and written to the circular buffer by MCU interrupt. Each round of tasks includes: (1) reading the PPG+ACC from the buffer for the most recent 2-4 seconds; (2) performing DSP preprocessing; (3) performing artifact recognition; (4) constructing the artifact index; (5) inputting CNN+TCN for binary classification; (6) outputting the signal quality level (1 or 0); (7) if it is 0, prompting the user to remeasure.

[0118] Therefore, it is evident that a single LED, single PD, single-channel dark current suppression AFE, ordinary triaxial accelerometer, and main control MCU are sufficient to achieve high-accuracy signal quality assessment. It is fully compatible with mainstream smartwatches / band hardware, achieving software-level enhancement without modifying the optical structure. It directly matches most existing wearable hardware. Furthermore, using a lightweight structure combining CNN and TCN, along with depthwise separable convolution, dilated convolution, 8-bit quantization, CMSIS-NN / RISC-V DSP optimization, interrupts, double buffering, and static memory reuse, it achieves embedded real-time inference. The inference latency is lower. Compared to most current signal quality assessments that require inference from the mobile CPU or cloud server, where the MCU cannot support LSTM (Long Short-Term Memory) / GRU (Gated Recurrent Unit) or large CNNs, this invention achieves embedded blood pressure signal quality control within the MCU. It can detect waveform morphology damage caused by movement traction, identify situations where ACC energy is low but PPG is severely deformed (pressure change), and identify situations where ACC energy is high but PPG is unaffected (e.g., wrist movement but tight strap). By identifying motion-induced artifacts in waveforms through cross-modal PPG-ACC cross-correlation, TCN captures cardiac rhythm stability, and the artifact intensity index provides artifact priors, enabling the device to accurately identify complex artifacts. This significantly improves the usable signal recognition rate and significantly reduces the artifact false reception rate, thus significantly enhancing the overall stability of backend algorithms such as continuous blood pressure, heart rate variability, and pulse wave propagation velocity.

[0119] Furthermore, embodiments of this application also disclose an electronic device, Figure 8 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0120] Figure 8This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the continuous blood pressure signal quality assessment method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0121] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0122] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0123] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0124] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system 221 can be Windows Server, Netware, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the continuous blood pressure signal quality assessment method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0125] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned continuous blood pressure signal quality assessment method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0126] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0127] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs (Compact Disc-Read Only Memory), or any other form of storage medium known in the art.

[0128] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0129] The solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method of continuous blood pressure signal quality assessment, characterized in that, Applications in wearable devices, including: Acquire PPG and ACC signals synchronously collected by local sensors; Calculate the cross-correlation characteristics between the preprocessed PPG signal and the ACC signal; Based on the cross-correlation characteristics, the preprocessed PPG signal, and the ACC signal, an artifact intensity index is constructed. The PPG signal, the ACC signal, and the artifact intensity index are input into a preset quality assessment model to perform a fusion analysis on the PPG signal and the ACC signal based on the artifact intensity index, in order to determine whether the PPG signal is used for blood pressure estimation and output the corresponding quality assessment results.

2. The continuous blood pressure signal quality assessment method of claim 1, wherein, After acquiring the PPG and ACC signals synchronously collected by the local sensor, the process further includes: Write the PPG signal and the ACC signal into a circular buffer; The PPG signal is extracted from the circular buffer and filtered to extract periodic consistency features and signal-to-noise ratio, so as to obtain the preprocessed PPG signal.

3. The continuous blood pressure signal quality assessment method of claim 1, wherein, The calculation of the cross-correlation characteristics between the preprocessed PPG signal and the ACC signal includes: Within a preset time window, the maximum cross-correlation value of the preprocessed PPG signal and the ACC signal under multiple time delays is calculated to obtain the cross-correlation characteristics.

4. The method for assessing the quality of continuous blood pressure signals according to claim 2, characterized in that, The construction of the artifact intensity index based on the cross-correlation characteristics, the preprocessed PPG signal, and the ACC signal includes: The artifact intensity index is obtained by weighted summation based on the cross-correlation characteristics, the energy value of the ACC signal, the periodicity characteristics, and the reciprocal of the signal-to-noise ratio.

5. The continuous blood pressure signal quality assessment method of claim 1, wherein, The method of fusing and analyzing the PPG signal and the ACC signal based on the artifact intensity index to determine whether the PPG signal is used for blood pressure estimation and outputting corresponding quality assessment results includes: The preprocessed PPG signal and the ACC signal waveforms are convolved through the convolutional neural network layer of the preset quality assessment model to extract local morphological features of the PPG signal, including the slope of the rising edge of the pulse, the shape of the diabetic notch, and the sharpness of the peak. The local morphological features are subjected to dilated convolution and residual connection operations through the temporal convolutional network layer of the preset quality assessment model to capture the contextual features of the PPG signal. The contextual features are fused with the artifact intensity index, and a quality assessment result is generated using the fully connected layer and activation function of the preset quality assessment model to determine whether the PPG signal is used for blood pressure estimation.

6. The continuous blood pressure signal quality assessment method of claim 1, wherein, Acquiring PPG signals, including: A single PPG signal is acquired through reflection from a single green light source.

7. The continuous blood pressure signal quality assessment method according to any one of claims 1 to 6, characterized in that, Also includes: If the quality assessment result of the current time window determines that the PPG signal is not used for blood pressure estimation, then a user prompt is triggered and the current continuous blood pressure estimation process is interrupted. If PPG signals from multiple consecutive time windows are determined to be unsuitable for blood pressure estimation, a device wearing status check prompt message will be generated.

8. A continuous blood pressure signal quality assessment apparatus, characterized by Applications in wearable devices, including: The signal acquisition module is used to acquire the PPG and ACC signals synchronously collected by the local sensors. The feature calculation module is used to calculate the cross-correlation features between the preprocessed PPG signal and the ACC signal; An index construction module is used to construct an artifact intensity index based on the cross-correlation characteristics, the preprocessed PPG signal, and the ACC signal. The quality assessment module is used to input the PPG signal, the ACC signal and the artifact intensity index into a preset quality assessment model, and to perform fusion analysis on the PPG signal and the ACC signal based on the artifact intensity index to determine whether the PPG signal is used for blood pressure estimation, and output the corresponding quality assessment results.

9. An electronic device, comprising: include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the continuous blood pressure signal quality assessment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the continuous blood pressure signal quality assessment method as described in any one of claims 1 to 7.

Images