Low perfusion oximetry method based on dynamic light compensation

Abstract

The present application relates to a low perfusion blood oxygen saturation detection method based on dynamic light compensation, a combined feature relationship based on multi-wavelength reflection intensity and fluctuation amplitude is constructed, the optical sparsity of tissue, that is, the optical representation of the density of red blood cells in the tissue, is dynamically judged; the current perfusion degree is deduced through the morphological characteristics of the reflection waveform; a perfusion optical index is established; the time series modeling method is used to analyze the change trend of the light reflection dynamics in the past period of time; the LED driving strategy is dynamically adjusted to form a light compensation behavior similar to physiological rhythm in three modes of step-by-step climbing, pulse stimulation and cycle maintenance; each light compensation action is regarded as a physiological stimulation; the dynamic morphological similarity discrimination is introduced to quantify whether each light compensation truly causes tissue response, and to distinguish real physiological fluctuations from noise disturbances; for low-response light compensation, the light compensation intensity is feedbackly reduced; otherwise, for high-response light compensation, the key deformation characteristics are extracted for signal reconstruction, which can enhance the signal recognition ability in a low perfusion environment.

Description

Technical Field

[0001] This invention relates to a method for detecting low perfusion oxygen saturation, specifically a method for detecting low perfusion oxygen saturation based on dynamic light compensation. Background Technology

[0002] Although Chinese patent CN103027690A proposes a method for measuring low-perfusion oxygen saturation based on autocorrelation modeling, which improves the measurement accuracy in weak signal environments to some extent and has the advantages of signal enhancement and noise suppression, its method still has many structural deficiencies and fundamental drawbacks when analyzed from multiple dimensions such as practical application scenarios, technical logic depth, physiological response reliability, system adaptability, and energy efficiency control. In particular, its inherent limitations gradually become apparent when facing dynamic perfusion states, complex physiological fluctuations, and individualized responses, making it difficult to meet the current development needs of intelligent wearables and continuous health monitoring.

[0003] First, this patented method is essentially still a static modeling + signal post-processing approach, lacking a multi-dimensional signal quality control and correction framework centered on pattern recognition, pattern classification, and pattern clustering. Its key technical route relies on traditional filtering and autocorrelation enhancement of PPG signals, estimating the AC component through modeling parameters to obtain blood oxygen estimates. However, the entire process lacks dynamic adjustment and control of the input light source and real-time signal pattern recognition to determine waveform reliability. In other words, it cannot actively increase luminous flux to improve reflection quality when the signal is weak; it can only rely on the enhancement of the signal's inherent characteristics. Therefore, under extremely low perfusion conditions, when the original signal is close to the noise threshold, the waveform amplitude is extremely low, or the autocorrelation is already poor, the modeling parameters of this method are prone to drift or misjudgment, leading to severe estimation errors. Secondly, the autocorrelation method relies on the periodic structural stability of the signal for processing, which is effective under certain perfusion conditions. However, under low perfusion, cardiac waveforms often exhibit atypical morphological changes, such as irregular peaks, plateaued waveforms, and disordered micro-motion structures. In such cases, without signal pattern classification and object recognition mechanisms, the autocorrelation function is prone to spurious peaks or weak periodic matching, affecting the accurate extraction of AC components and the interpretation of physiological significance. Furthermore, this method does not involve structural analysis of waveform morphology, nor does it utilize object recognition or pattern clustering based on physiological priors to determine tissue reflex characteristics. It only performs signal enhancement from a statistical perspective, lacking a deep understanding of physiological mechanisms and response judgment. Especially in the peripheral tissue regions where the capillary system dominates reflexes, the microscopic morphological characteristics of the waveform are far more diagnostically valuable than the overall energy intensity, which this patent completely fails to address, making it a typical signal-leading method rather than a physiological response-leading method. Thirdly, the existing technology lacks a dynamic feedback control mechanism.

[0004] Modern blood oxygen monitoring systems, especially in wearable environments, increasingly rely on a closed-loop feedback system between light output and tissue response. This means the system should be able to distinguish between effective physiological waveforms and artifact noise in real time through signal pattern recognition and object recognition, and automatically adjust parameters such as light source power, wavelength, and duty cycle according to the current physiological state of the tissue to achieve precise supplemental lighting and energy consumption optimization. However, the light source parameters in this patent are fixed, failing to adapt to changes in skin thickness, blood vessel distribution, and ambient brightness, and also unable to distinguish whether the current signal enhancement truly originates from increased blood flow or is merely accidental noise interference. Fourth, this method's processing heavily relies on the stability of the modeling parameters. In actual wearing scenarios, factors such as movement, temperature, or nerve tension can cause drastic fluctuations in the reflected signal. Without a quality stratification and screening mechanism based on pattern clustering, processing all data together will lead to misestimation of blood oxygen levels in high-noise areas. Fifth, the scheme does not propose any form of signal weighting or data credibility scoring system, lacks intelligent evaluation and weight allocation of the quality of multiple waveform segments, and ignores the core issue of significant differences in waveform quality under low perfusion conditions. In modern personalized health monitoring scenarios, especially for people with chronic diseases, their signal status is often in flux. Without the support of signal pattern recognition, pattern classification, and pattern clustering, it is impossible to prioritize the use of high-quality waveforms and eliminate inferior waveforms, which ultimately greatly limits the feasibility and practical value of this technology in clinical and consumer applications. Summary of the Invention

[0005] The purpose of this invention is to provide a method for detecting low perfusion oxygen saturation based on dynamic light compensation, thereby addressing some of the drawbacks and shortcomings pointed out in the background art.

[0006] The present invention addresses the aforementioned technical problems by employing the following technical solution: a method for detecting low perfusion oxygen saturation based on dynamic optical compensation, comprising: constructing a combined characteristic relationship based on multi-wavelength reflection intensity and fluctuation amplitude to dynamically determine tissue optical sparsity, i.e., the optical characterization of red blood cell density in tissue; inferring the current perfusion level through the morphological characteristics of the reflection waveform, including pulse edge slope, peak-to-valley ratio, and micro-fluctuation cycle; thereby establishing perfusion optical indicators;

[0007] The dynamic trend of light reflection over a period of time was analyzed by using time series modeling methods including gated recursive units (GRU) or sliding window variable-length exponential filtering; and the LED driving strategy was dynamically adjusted to form a physiological rhythm-like supplemental lighting behavior with three modes: step-by-step climbing, pulse stimulation, and periodic maintenance.

[0008] Each illumination action is considered a physiological stimulus. Within the time windows before, during, and after illumination, the amplitude, phase, and morphological response differences of the three light segments are calculated. Dynamic morphological similarity discrimination is introduced to quantify whether each illumination truly causes a tissue response and to distinguish between real physiological fluctuations and noise disturbances. For low-response illumination, the illumination intensity is reduced in a feedback manner; conversely, for high-response illumination, key deformation features are extracted for signal reconstruction.

[0009] Furthermore, the method for establishing perfusion optical parameters includes:

[0010] S1. Collect multi-wavelength light reflection signals from the target detection area and obtain pulse photoplethysmography waveform;

[0011] S2. Extracting morphological features from the reflected waveform, including:

[0012] a) The slope of the rising edge of the pulse.

[0013] b) The amplitude ratio of peak to trough pulse values

[0014] c) The repeatability and similarity of micro-fluctuation cycles;

[0015] S3. Based on the waveform morphology characteristics, construct an optical perfusion status index to determine whether the current tissue is in a low perfusion state; when the perfusion status index is lower than a preset threshold, trigger the dynamic light compensation control module to adjust the incident light power.

[0016] S4. During the execution of light compensation, continuously monitor the changes in waveform morphology, judge the effect of light compensation, and adjust the light output accordingly; calculate the blood oxygen saturation estimate based on the reflected signal after light compensation optimization.

[0017] Furthermore, the extraction of the pulse edge slope includes analyzing the time and amplitude change rate of the rising edge of the photoplethysmography waveform to determine the blood propulsion speed and tissue photoresponse speed.

[0018] Furthermore, the amplitude ratio of the peak value to the trough value is used to represent the magnitude of tissue volume change within the cardiac cycle; the smaller the amplitude, the weaker the perfusion. The micro-fluctuation cycle is a secondary fluctuation outside the main pulse waveform, used to determine capillary reflux or vascular elasticity in the tissue.

[0019] Furthermore, the optical perfusion status index is a comprehensive score calculated based on the weighted sum of multiple waveform features, used to reflect the real-time tissue perfusion quality; the dynamic light compensation control module includes a light source driver and a feedback controller, used to dynamically adjust the power and pulse rhythm of the LED light source according to the optical perfusion status index;

[0020] In current technologies, blood oxygenation detection heavily relies on the stability of tissue reflection signals, especially under low perfusion conditions. The photoplethysmography fluctuations caused by tissue blood flow are extremely weak, resulting in signals close to the noise lower limit, making it difficult to establish an effective ratio and easily causing blood oxygenation estimation errors. At the same time, traditional dynamic light compensation methods are usually triggered by static thresholds and only perform linear light compensation based on photoplethysmography intensity, lacking physiological response correlation and failing to accurately identify whether hemodynamic feedback has truly occurred. Based on three key temporal characteristics of the photoplethysmography waveform: rising edge slope S(t), waveform amplitude ratio V(t), and micro-fluctuation periodic stability Δ(t), a dynamic comprehensive scoring function is constructed to characterize the real-time perfusion state response capability of tissues.

[0021] Its comprehensive optical perfusion index function is defined as follows:

[0022]

[0023] in:

[0024] Π(t): Represents the optical perfusion status index score at time point t; ε: Historical time variable within the integration window; τ: Length of the integration review window, representing the memory length of the perfusion response; S(ξ): Waveform rise slope, reflecting cardiac conduction efficiency (waveform rise rate per unit time); V(ξ): Waveform peak-to-valley amplitude ratio, representing the ability of a unit heartbeat to cause volumetric blood flow change; Δ(E): Waveform micro-periodic variation, reflecting the stability of the local blood flow fine-tuning structure; the smaller the value, the more regular the fluctuation; Λ: Normalization constant, used to limit the function result to the [0,1] interval, improving the stability of the algorithm; log(1+V 2 ·Δ): Nonlinear coupling term, used to emphasize the enhancement contribution of high-amplitude waveforms to the perfusion state in the presence of regular micro-fluctuations; e -Δ The more stable the micro-fluctuations (the smaller Δ), the larger this term is, indicating higher perfusion quality;

[0025] When the scoring function Π(t) shows that the perfusion state is below the threshold (e.g., Π(t) < θ, where θ is an empirically set threshold), the system initiates the light compensation control process.

[0026] The light compensation control module includes:

[0027] Light source driver: used to adjust the power and current pulse width of the LED light-emitting unit;

[0028] Feedback controller: continuously receives Π(t) and incorporates its time trend To determine whether the tissue produces a substantial response to supplemental lighting;

[0029] The controller dynamically adjusts based on the following strategies:

[0030] If Π(t) is in a decreasing trend, continue to increase the pulse intensity;

[0031] If the fluctuation of Π(t) does not change significantly, maintain the current rhythm;

[0032] If Π(t) increases significantly, reduce the light intensity to save energy.

[0033] Furthermore, the blood oxygen saturation calculation process introduces response weights based on the changes in waveform morphology before and after light compensation, which are used to correct the estimation bias of the traditional ratio method.

[0034] Furthermore, the method for calculating the differences in amplitude, phase, and morphology of the capacitance-recorded waveforms of the three light segments includes:

[0035] Dynamic light compensation signals are sent to the detected tissue at a preset rhythm; each light compensation action is divided into three time windows: before light compensation, during light compensation, and after light compensation; multi-dimensional features of amplitude, phase, and morphology are extracted from the photoplethysmography waveforms collected in each time window.

[0036] The characteristics of the three waveforms are compared to determine the differences in the physiological response of tissues to supplemental lighting. The waveforms with physiological response characteristics within the supplemental lighting cycle are used for blood oxygen saturation estimation, while the remaining waveforms are downweighted or discarded.

[0037] Furthermore, the amplitude characteristics include the amplitude variation between the peak and trough of the pulse waveform, used to determine whether the supplemental light causes enhanced tissue reflection; the phase characteristics include the advance or delay of the pulse wave start time relative to the previous cycle reference point, used to assess the tissue's response speed to the supplemental light.

[0038] Furthermore, the morphological features include changes in the contour of the pulse waveform, the slope of the rising edge, and the appearance and disappearance of micro-wave structures, which are used to determine whether changes have occurred in the tissue microcirculation. If the waveform morphology differences in the three time windows are significant, it is judged as effective physiological response supplementary lighting; otherwise, it is considered as ineffective supplementary lighting, and the intensity or cycle of supplementary lighting is reduced.

[0039] This invention utilizes a method to dynamically adjust LED light compensation strategies, achieving synchronized control with the actual response of microcirculation. This method accurately identifies whether tissues exhibit a physiological response to supplemental lighting by dividing time windows, extracting morphological features, and constructing an integral difference function. Each time dynamic LED supplemental lighting occurs, the system divides the photoplethysmography waveform reflection waveform into:

[0040] Front window for supplemental lighting: Represents the microcirculation output under natural conditions;

[0041] Window during supplemental lighting: Observe the immediate changes in waveform under supplemental lighting stimulation;

[0042] Window after supplemental lighting: Observe the trend of morphological recovery after supplemental lighting is removed;

[0043] In each window, the system extracts the following three types of morphological features:

[0044] Pulse wave profile characteristics, including waveform symmetry, acromion formation, and rise duration;

[0045] The slope of the rising edge reflects the speed at which blood flows into the tissue;

[0046] The appearance, disappearance, and periodic changes of micro-fluctuation structures characterize the activity of the capillary system.

[0047] By constructing the following original integral function, the differences in the morphology of the three waveform segments are quantified, thereby determining whether the supplemental lighting induces physiological changes in tissue microcirculation:

[0048]

[0049] in:

[0050] ΔΨ: Defined waveform response morphology difference index, used to measure the significance of physiological response; t1, t3: Represent the start time of the window before illumination and the end time of the window after illumination, respectively; T: Any time point within the integration interval; χ s (τ): The waveform composite morphology function within the window during supplementary lighting, which includes the fusion of contour factor, slope factor, and micro-fluctuation factor; χ b (τ), χ r (τ): Composite waveform morphology values ​​at corresponding time points before and after supplemental lighting; Ω(τ): Physiological reliability weighting function, used to amplify rhythmic responses and suppress the influence of noise disturbances; |·|: Represents the absolute morphological shift between the three segments, indicating the net change in the waveform caused by supplemental lighting;

[0051] This function captures the waveform structure changes of tissue during supplemental lighting by integrating in the time domain, and uses the average waveform before and after supplemental lighting as a reference to assess whether there is significant deformation in the middle section. When ΔA exceeds the system's preset response threshold ε, the supplemental lighting behavior is determined to be effective physiological response supplemental lighting; otherwise, it is considered ineffective supplemental lighting or optical over-supplementation, and the system will reduce the LED intensity or shorten the duration of supplemental lighting in the next supplemental lighting cycle.

[0052] Furthermore, the response differences of the three waveforms are used to construct a response reliability score, which is used to weight different waveforms in the blood oxygen saturation calculation.

[0053] The beneficial effects of this invention are as follows: Under conditions of insufficient tissue blood perfusion (such as peripheral circulatory disorders, low temperature environments, and chronic diseases), traditional PPG signals have extremely weak amplitudes and low signal-to-noise ratios, easily leading to fluctuations or misjudgments in blood oxygen estimation. This invention achieves dynamic light compensation by real-time judgment of tissue perfusion status and adaptive control of LED light source intensity and pulse rhythm, thereby significantly enhancing the ability to identify weak reflected signals and ensuring stable acquisition of usable waveforms even under low perfusion conditions. Unlike traditional passive illumination methods that use static thresholds or fixed light power, this invention introduces a real-time response evaluation model based on multi-dimensional features such as waveform morphology, amplitude, and phase. By analyzing the differences in three waveform segments before, during, and after illumination, it determines whether the tissue has produced a physiological response. Illumination continues only after being identified as an "effective response," possessing physiological feedback closed-loop control capabilities, significantly improving the system's intelligence and energy efficiency.

[0054] The reliability of each waveform segment is dynamically evaluated, and its weight in the blood oxygenation estimation process is adjusted accordingly, effectively avoiding interference from invalid waveforms. This waveform quality-based weighting strategy can significantly improve the stability and accuracy of blood oxygenation estimation even under conditions of severe perfusion fluctuations and significant quality differences between waveforms. This method not only evaluates the effect of a single supplemental lighting session but also incorporates the tissue's temporal response to continuous supplemental lighting, forming a physiological-like supplemental lighting rhythm with progressively increasing, periodically maintained, or decreasing rhythms. This makes the LED output more closely match the actual tissue condition, avoiding excessive light exposure that could cause interference or discomfort to the tissue. Attached Figure Description

[0055] Figure 1 This is a functional relationship diagram of the dynamic light-compensated low-perfusion blood oxygen detection of the present invention.

[0056] Figure 2 This is a flowchart of the dynamic optical compensation control based on time series analysis of the present invention.

[0057] Figure 3 This is a diagram showing the relationship between the intelligent supplementary lighting function based on three-segment waveform analysis of the present invention.

[0058] Figure 4 This is a schematic diagram of the home blood oxygen monitoring and intelligent light compensation system for Mr. Wang under hypoperfusion conditions in Example 1 of this embodiment.

[0059] Figure 5 This is a schematic diagram of the structure of Mr. Wang's layered dynamic supplemental light blood oxygen monitoring system in Embodiment 2. Detailed Implementation

[0060] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0061] Combined with appendix Figure 1This invention relates to a low-perfusion oxygen saturation detection method based on dynamic optical compensation. It utilizes the physical and morphological characteristics of the tissue's reflected light waveform to infer the local tissue's blood perfusion status, guiding dynamic optical compensation behavior. Even under extremely weak perfusion conditions, it can accurately estimate blood oxygen saturation. This method does not rely on external physical sensors such as pressure or skin temperature. Instead, it uses multi-wavelength optical reflection characteristics and PPG waveform details as a foundation, constructing a fusion data model to estimate tissue optical sparsity, thereby indirectly characterizing the dynamic changes in erythrocyte density distribution, i.e., tissue perfusion level. The tissue's optical response is transformed into a perfusion status perception index, which then drives optical compensation behavior, forming a physiologically meaningful closed-loop control system. Data acquisition is based on a three-wavelength optical detection system, typically including three LED channels for red light (660nm), near-infrared light (880-940nm), and a reference wavelength (e.g., 530nm or 590nm), and a photodiode (PD) to simultaneously acquire the reflection intensity of each wavelength on the target tissue (e.g., fingertip). The device continuously acquires the reflected light intensity at various wavelengths at a high sampling rate (e.g., 100Hz to 250Hz) and records the timestamp-aligned PPG waveform.

[0062] The data preprocessing stage first performs noise suppression and baseline drift correction, mainly using wavelet transform and adaptive filtering methods to remove motion artifacts and ambient light interference. Then, within each sampling period, three core morphological features are extracted: first, the pulse edge slope, i.e., the slope of the rising segment of the PPG waveform, representing the dynamic characteristics of blood entering peripheral tissues; second, the peak-to-trough ratio, i.e., the AC / DC amplitude ratio, reflecting the amplitude of local volume change within each cardiac cycle; and third, micro-fluctuation periodic features, used to reflect the micro-rhythm of capillary blood flow in tissues, whose changes reflect whether the local perfusion system has stable microcirculatory activity. All features are normalized and a sliding window is constructed in the time dimension to form a structured input vector. The model part adopts a lightweight neural network structure to support the operating efficiency of embedded platforms (such as MCUs). The model input is a structured feature sequence within a time-series window (e.g., the past 5 seconds or 10 heartbeat cycles), containing the aforementioned multi-wavelength reflection intensity and morphological features. The output is a scalar, the perfusion optical score index, ranging from 0 to 1, representing the optical sparsity of the current tissue—a lower value indicates sparser blood signals in the reflected light and weaker perfusion. The model structure employs a three-layer temporal perceptron (e.g., GRU or simplified LSTM) superimposed with a multilayer perceptron (MLP) output layer. Model training uses a dataset with real perfusion labels, which can be obtained from clinical standard perfusion instruments or PI sensors. These labels are then paired with the acquired reflection waveforms to construct supervised learning samples. The dataset includes various perfusion scenarios under different temperatures, pressures, and pathological conditions to enhance the model's generalization ability. The training loss function is mean squared error plus a regularization term, introducing a weighting mechanism to enhance sensitivity to low perfusion states. Iterative training is performed using the Adam optimizer. The final model can output an infusion score per second in practical applications, which is used to determine in real time whether to enter the light compensation state and to control parameters such as LED driving current and pulse width to dynamically adjust the luminous intensity.

[0063] Combined with appendix Figure 2As shown, by introducing a time-series modeling mechanism, the problem of delayed and inaccurate supplemental lighting caused by weak or unstable fluctuations in light reflection signals in traditional blood oxygenation detection equipment under conditions of insufficient perfusion is solved. Time-series analysis methods such as gated recursive units (GRUs) or sliding window variable-length exponential filtering are used to model the dynamic trend of light reflection signals over a period of time. This allows for real-time prediction of the potential response trend of tissues to light compensation, and based on this, the driving strategy of the LED light source is dynamically adjusted to achieve physiologically rhythmic light compensation behavior. This ensures that physiologically meaningful PPG waveforms can be continuously acquired and accurate blood oxygenation estimates can be completed even under low perfusion conditions. Data acquisition relies on a multi-channel optical detection system, which includes red light, near-infrared light, and an LED and photodetector (PD) with a reference wavelength channel. The light source and detector are installed in a wearable device, continuously recording the reflected light intensity of each channel at a high sampling frequency (e.g., 200Hz). Simultaneously acquired signals include dynamic LED supplemental lighting control signals (actual luminous power), ambient light intensity, and heart rhythm timestamps, which are used to establish a dynamic correspondence with the reflected waveform. In the data preprocessing stage, signal denoising is first performed using adaptive bandpass filtering and wavelet packet decomposition to remove motion artifacts and baseline drift. Then, a time-series window is constructed for each sampling point (e.g., approximately 1000 data points from the past 5 seconds). The rate of change of light reflection intensity, local extremum density, and short-term average amplitude changes within the window are extracted in a structured manner. These are then combined with previously output LED light intensity and control commands as covariates to form a time-series input tensor. A sliding window approach is used to construct a time-sliding input stream to feed the time-series model to learn the light reflection trend. One implementation method is to use a gated recurrent unit (GRU) neural network, which has better memory capacity and lower computational complexity than traditional RNNs, making it suitable for embedded platforms. The GRU input is a T×F dimensional tensor, where T is the time window step size and F is the feature dimension of each time point. The output is a high-order state estimate of the light reflection trend at the current moment. Another approach is an adaptive trend-tracking model based on variable-length exponential filtering. This model dynamically adjusts the exponential decay factor at each time point according to the historical rate of change, achieving a weighted response to local fluctuations. This is suitable for modeling the response of tissues sensitive to frequency changes. Regardless of the model output, the final result will be an intermediate quantity describing the trend of tissue light reflection changes, called the dynamic reflection trend response score, which is used to guide subsequent supplemental lighting.In terms of supplemental lighting control, the response score is used to drive the LED output strategy to dynamically switch between three states: when the score shows a negative change in tissue reflectance (i.e., the light reflectance signal continues to decrease or the fluctuation disappears), the system enters a gradual ramp-up mode, that is, the LED driving current is increased in a gentle linear increment; when the score shows short-term strong fluctuations without a periodic pattern, the system judges that the tissue is in a stage of lag or compression, and switches to a pulse stimulation mode, that is, using short-term strong light stimulation with a low duty cycle and high amplitude to try to induce a tissue response; when the score is stable and the light reflectance pattern is rhythmically output, the system enters a periodic maintenance mode, that is, outputting light compensation behavior with a constant frequency and power to maintain a balance between energy efficiency and tissue adaptability. This three-stage supplemental lighting mechanism can be regarded as a photophysiological response mode that simulates the rhythm of real microcirculation, enabling light compensation and tissue blood flow to form a feedback-like synergistic mechanism.

[0064] Combined with appendix Figure 3As shown, each LED illumination event is treated as a controllable micro-photophysiological stimulus to observe whether the tissue produces a substantial hemodynamic response under light stimulation, thereby determining whether the illumination event has physiological significance and can be used for signal enhancement or blood oxygenation estimation. Starting from signal acquisition, a three-segment waveform analysis mechanism is constructed: at each illumination event, a baseline window before illumination begins, a stimulation window during illumination, and a recovery window after illumination ends are defined. Within each window, three key features of the tissue reflection waveform are acquired and calculated: waveform amplitude change, phase change, and morphological structure change. This quantifies whether the illumination event causes real tissue blood flow changes, rather than merely producing physical light enhancement or system noise disturbance. Data acquisition is based on a PPG (photoplethysmography) system, simultaneously acquiring reflected light signals in red, near-infrared, and a reference wavelength channel at a sampling frequency of 125-250Hz. In the wearable device, a timestamp is added each time illumination is triggered to mark the three data segments before, during, and after illumination. When extracting features in each window, the waveform is first preprocessed for denoising, including baseline correction based on sliding minimum filtering, artifact removal by wavelet packet decomposition, and extraction of effective components of the pulse main frequency band (approximately 0.5-5Hz) using a bandpass filter. Subsequently, three types of waveform indicators are extracted in the three windows: (1) amplitude features are the mean and standard deviation of the pulse amplitude in the current window, which measures the strength of blood flow fluctuations; (2) phase features are the time offset of the current period waveform relative to the previous pulse reference point, reflecting local blood flow delay or advance; (3) morphological features include secondary morphological factors such as rising slope, shoulder peak ratio, waveform width, and bimodal structure, which are used to determine whether blood flow changes actually occur at the microstructural level. In order to systematically quantify the differences between these three waveforms and make an effectiveness judgment, this method introduces a model module called the dynamic morphological similarity discriminator. This module determines whether the supplemental lighting has caused meaningful physiological changes by calculating the distance between the stimulation window and the morphological features of the preceding and following windows. Its core is to construct a similarity scoring function, and denote the feature vectors of the three windows as F. b F s ,F r By defining the morphological difference vector Δ s =F s -(F b +F r The difference is then mapped to a normalized score Ψ using a nonlinear mapping function. A higher score indicates a true response in the waveform. This function can be implemented as a simple feedforward neural network or heuristic rule logic. Its training data comes from multiple real-world supplementary lighting events, and is labeled manually or semi-manually to determine whether the supplementary lighting is effective, i.e., whether a physiological response occurs, thus forming supervised learning samples.

[0065] During model training, the input consists of waveform feature vectors extracted within three time windows, labeled with whether the illumination triggered a valid waveform change (1 or 0). The network structure employs a two-layer fully connected neural network, using either Swish or Tanh activation functions to enhance the perception of small-scale nonlinear changes. Weighted cross-entropy is used as the loss function to improve the model's accuracy in identifying low-response events, and the Adam optimizer is used for iteration. The training dataset includes illumination response behaviors under various environments, bonding, and irrigation conditions. Precision, recall, and AUC curves are evaluated on the test set to ensure the model maintains real-time performance and accuracy after deployment on edge devices. During runtime, the DMSE model immediately calculates a response score Ψ after each illumination. If the score is lower than a set threshold θ, the model is not considered active. l If the system determines the light source to be low-response supplemental lighting, it will feedback-wise reduce the LED light intensity and shorten the duration, entering energy-saving mode; if the score is higher than the set threshold θ, the system will adjust accordingly. h If the signal is positive, it is determined to be high-response supplemental lighting. The system will record the key waveform changes caused by this supplemental lighting, such as a sharp increase in slope and enhancement of microstructure, as the target structural features of the subsequent signal enhancement module, so as to promote the physiological reconstruction of PPG signal at the structural level. Even if the perfusion is insufficient, it can restore the waveform input with physiological rhythm.

[0066] Example 1:

[0067] Combined with appendix Figure 4 As shown, a user, Mr. Wang, a 65-year-old diabetic patient, experienced typical hypoperfusion symptoms such as cold fingertips and insufficient perfusion due to long-term peripheral microcirculation dysfunction. He used the technology based on this invention at home. The system executes step S1, which involves collecting multi-wavelength reflection signals from the target detection site (the tip of the left index finger). The LED in the device emits pulsed light at 660nm (red light), 880nm (near-infrared), and 530nm (reference green light), with each wavelength emitting alternately at a frequency of 50Hz. Simultaneously, a coaxial photodetector collects the reflection intensity signal at a sampling frequency of 250Hz, continuously recording for 20 seconds to form the initial data segment, and extracting the PPG waveform. Preliminary data shows that the reflection intensity fluctuation range in the near-infrared band is ±0.15V, the red light fluctuation is ±0.10V, while the green light almost presents a low-amplitude flat waveform with fluctuations not exceeding ±0.03V, reflecting the possibility of initially low perfusion.

[0068] In step S2, the system enters the waveform morphology feature extraction stage. First, the edge slope of the rising edge of the pulse waveform is analyzed. The average rising slope of the 3rd, 4th, and 5th cardiac cycles in the near-infrared band is measured to be 0.012V / ms, significantly slower than the normal value (e.g., >0.025V / ms). Second, the peak-to-trough amplitude ratio of each cardiac cycle is analyzed, revealing an average amplitude of 0.09V, while the baseline DC value is 0.95V, resulting in an AC / DC ratio of approximately 0.095, lower than the conventionally set low perfusion warning threshold of 0.12. Finally, the similarity of micro-fluctuation cycles is analyzed. Through wavelet packet decomposition and zero-crossing density analysis, the system determines that the fine fluctuation cycle of the PPG waveform is structurally unstable over the past 5 cardiac cycles, with an average period variance of 0.14s. 2 The result indicates an irregular and fluctuating state. After summarizing the three indicators, the system constructs a feature vector [0.012, 0.095, 0.14], which is then fed into a pre-trained perfusion status scoring model (this model is a two-layer feedforward neural network, trained with 20,000 data samples). The perfusion optical index value output by the model is 0.36, which is far below the normal perfusion threshold of 0.70, and is therefore judged as severely under-perfusion.

[0069] In step S3, the system triggers the dynamic light compensation module. After reading the perfusion index value, the light source controller linearly increases the current drive of the LED red light and near-infrared channels from the original 8mA to 16mA, and adopts a dual-strategy output method of gradual increase + pulse intervention. That is, it increases by 1mA per second for 3 seconds, and then introduces a short pulse with an intensity of 22mA, a duty cycle of 30%, and a duration of 600ms to induce tissue photoreaction. During this process, the system continues to execute S4, monitoring the changes in waveform morphology during the supplemental lighting and judging the supplemental lighting effect in real time.

[0070] Based on real-time data acquisition, the system detected that during the 6th-8th cardiac cycle, the near-infrared waveform's rising slope increased to 0.026 V / ms, returning to normal levels; the AC / DC ratio increased to 0.13; the similarity of micro-fluctuation cycles significantly improved; and the cycle variance decreased to 0.06 s. 2 These morphological changes indicate a clear physiological response of the tissue to supplemental lighting. The system inputs the before-and-after morphological differences into the morphological similarity discriminator (DMSE) module, which calculates a dynamic response index of 0.82, far exceeding the preset threshold of 0.60, confirming high-response supplemental lighting. Based on this, the system decides to maintain the current LED output level and uses the current waveform for blood oxygen saturation calculation. The blood oxygen estimation model uses a ratio method with a correction coefficient for calculation, taking the AC / DC ratio of the three cycles after supplemental lighting for R calculation, resulting in a corrected R value of 0.41. After substituting into the calibration curve regression model, the output SpO2 is 93%, consistent with the patient's actual historical blood oxygen saturation data. Simultaneously, the system indicates perfusion recovery, indicating that the system has entered a stable supplemental lighting mode.

[0071] The slope of the pulse edge directly reflects the speed and intensity of the tissue's response to light after blood is propelled to the detection site by the heartbeat within the capillaries during each cardiac cycle. Therefore, it is used in this invention as an important signal for judging microcirculation efficiency. In Mr. Wang's device, the slope extraction process begins with the acquisition of the original PPG waveform. This waveform is obtained by illuminating the fingertip with light emitted from a multi-wavelength LED light source (660nm, 880nm, 530nm), and the reflected signal is collected by a photodiode to form a continuous waveform. The system records the reflection intensity at a sampling rate of 250Hz, forming an independent time series on each optical channel. The original waveform first undergoes noise suppression filtering, including a bandpass filter (0.5-8Hz) to remove motion and baseline drift interference, and smoothing sharp abrupt changes through wavelet transform or local weighted regression filtering, making the extracted rising edge more physiologically meaningful. Subsequently, the system executes the pulse recognition logic, using the zero-crossing point and derivative extremum method to lock the start and peak positions of each cardiac cycle. For each detected pulse cycle, the system extracts the rising segment from the trough to the peak as the analysis interval. This segment of the waveform typically exhibits a non-linear rising pattern. Instead of using traditional linear fitting methods, the system divides this rising segment into equally spaced time windows (e.g., every 4 ms) and calculates the instantaneous slope value within each window. This is achieved by dividing the change in reflection intensity ΔI by the time increment Δt to obtain the waveform's growth rate per unit time. To adapt to the situation of small waveform amplitude and slow change under low perfusion, this invention introduces a weighted time-amplitude response rate (WTVR) algorithm: within each micro-window, the current slope value is compared with the average value of the previous window to form an acceleration factor. If this acceleration value remains positive, it indicates that the blood propulsion speed is still increasing, and the tissue is gradually responding to the supplemental light behavior; conversely, if the acceleration is negative or close to zero, it can be preliminarily judged that blood propulsion is limited or the response is weak. The system detected that in the 5th cardiac cycle, the rising phase lasted 96 ms, with a maximum instantaneous slope of 0.013 V / ms and an average slope of 0.009 V / ms. Compared with the individual baseline model (average 0.021 V / ms under normal conditions), the system determined that the slope at the edge was slowing down, indicating sluggish blood propulsion in local tissues. This slope information was input into the optical perfusion model as part of the three-dimensional feature vector, and compared with the simultaneously extracted amplitude ratio (e.g., 0.095) and micro-fluctuation period similarity (e.g., period variance of 0.14 s). 2Together, these elements constitute the perfusion status assessment vector. The perfusion identification model is a shallow neural network consisting of three fully connected layers: the input layer is a 3D waveform feature vector; the first hidden layer contains 16 nodes with ReLU activation; the second layer has 8 nodes and introduces Dropout to enhance generalization; and the output layer is a single-node Sigmoid mapping, outputting a normalized perfusion score. Training data was collected from real clinical settings, including over 20,000 PPG signal segments in normal and low perfusion states. Each data segment was labeled with its actual perfusion status (based on perfusion index (PI) measurement or human expert interpretation). The model was trained using supervised learning, employing a weighted binary cross-entropy loss function, and the Adam optimizer was used for 100 iterations to converge to the optimal parameters. The final deployed model can run on an MCU platform with a latency of less than 30ms. Therefore, during Mr. Wang's wearing process, when the system judges that the slope continues to decrease based on edge slope extraction, and the perfusion score of the vector formed by the other two features is lower than the set threshold (e.g., <0.5), the system immediately activates the dynamic light compensation mechanism and continuously tracks the degree of slope recovery after supplemental light. If the slope rises back to above 0.021V / ms, the system judges that the tissue response is obvious and the supplemental light is effective. The subsequent waveform can be used as the basis for blood oxygen estimation.

[0072] The system further relies on two key waveform features—the amplitude ratio of peak to trough pulse values ​​and the periodic characteristics of micro-fluctuations—to comprehensively assess the microcirculation status of the fingertips and drive subsequent light compensation and blood oxygenation estimation strategies. During the raw data acquisition phase, the multi-wavelength LEDs in the device operate sequentially in red (660nm), near-infrared (880nm), and reference green (530nm) light. The photodetector acquires the reflected signal of each wavelength at a sampling rate of 250Hz to form the raw PPG waveform. In the 20-second data segment acquired by Mr. Wang, the reflected signal of the near-infrared channel exhibited basic periodic fluctuations, but the overall amplitude was low. The system preprocessed this waveform segment, including three steps: first, using a bandpass filter (0.5-5Hz) to remove low-frequency drift and high-frequency noise; second, using Empirical Mode Decomposition (EMD) to separate the main wave component from secondary micro-fluctuations; and third, segmenting and normalizing each cardiac cycle to ensure consistent feature extraction across different amplitude scales.

[0073] In extracting the peak-to-valley amplitude ratio feature, the system first detects the start point and peak point of each cardiac cycle, quickly locating them using the zero-crossing method combined with the first derivative extreme points. Then, it measures the amplitude difference ΔA between the peak value (P) and the previous trough value (V) of each cycle, normalizes it with the average DC level (DC) of that cycle, and calculates the AC / DC ratio. For Mr. Wang's actual data, in the near-infrared waveforms of cycles 5 to 8, the average ΔA was 0.085V, the average DC was 0.96V, and the AC / DC ratio was 0.088. This ratio is far lower than the typical value under perfusion conditions in healthy individuals (usually 0.12-0.25). Based on an empirical threshold, the system judges that the volumetric fluctuation amplitude is weak, reflecting a reduced local perfusion level. In the dynamic perfusion model, this ratio is normalized to a feature value f1, used as one of the inputs for perfusion status indicators. The system then analyzed the micro-fluctuation periodic characteristics of the PPG waveform, specifically the secondary low-amplitude, high-frequency periodic fluctuations remaining in the waveform after removing the main cardiac frequency component. These fluctuations are typically caused by non-cardiac mechanisms such as capillary reflux, venous filling, and changes in vascular elasticity. The system used wavelet packet decomposition (WPD) to decompose the PPG signal to level 4, selecting wavelet components in the 1.5-3.5Hz frequency band as the main components of the micro-fluctuations. Autocorrelation analysis was then used to detect periodic repeating structures. Mr. Wang's data showed irregular micro-fluctuations in this analysis, with significant variations between periods; the system measured the period variance to be 0.15s. 2The correlation peak value dropped to 0.42, far below the typical 0.75-0.9 range under high perfusion conditions. This indicates weakened microcirculation activity in the fingertip capillaries and insignificant elastic response of the vessel wall. The system combined the two indicators f1 (amplitude ratio) and f2 (micro-fluctuation periodic stability) with the previously extracted edge slope f3 to form a three-dimensional feature vector [f1, f2, f3] = [0.088, 0.15, 0.009], which was then input into the optical perfusion state model. The model structure is a three-layer feedforward neural network with 3 nodes in the input layer, 16 neurons in the first hidden layer (with ReLU activation function), 8 neurons in the second hidden layer, and a perfusion score output by a Sigmoid-normalized node in the output layer. A lower score indicates poorer perfusion. The model was trained using approximately 30,000 labeled data segments collected from the hospital. The training data included real PPG waveform segments under different perfusion states. Manual labels, provided by professional physicians, indicated the presence of effective tissue blood flow perfusion (based on the perfusion index PI). The model training employed a weighted cross-entropy loss function, used the Adam optimizer, and underwent 150 training epochs, ultimately achieving a test accuracy of 92.4%. After edge optimization and compression, it can be deployed to low-power MCU devices for real-time operation. In Mr. Wang's case, the model output a perfusion score of 0.38, below the supplemental lighting activation threshold of 0.5. The system immediately activated the LED dynamic light compensation mechanism, increasing the near-infrared LED's drive current from 8mA to 18mA, employing a periodic strong pulse stimulation mode. During the supplemental lighting process, the system continued to monitor new PPG data, finding that the peak-to-valley ratio increased to 0.12 in cycles 9 to 11, and the variance of the micro-fluctuation cycle decreased to 0.07s. 2 The autocorrelation increased to 0.66, indicating a significant enhancement in tissue blood flow fluctuations and microcirculation. The perfusion score also rebounded to 0.64, confirming the effectiveness of supplemental lighting. The updated PPG waveform was then used for subsequent blood oxygen estimation, resulting in an SpO2 of 94%, an improvement of 3% compared to before supplemental lighting.

[0074] During Mr. Wang's continuous monitoring, in order to overcome the problem of unstable blood oxygen estimation caused by insufficient blood perfusion in his fingertip tissue, real-time sampling and continuous integration modeling of three temporal physiological characteristics of PPG waveform—rise slope S(t), peak-to-valley amplitude ratio V(t), and micro-fluctuation periodic stability Δ(t)—were used to form an optical perfusion status scoring function Π(t). Based on this score, the supplementary lighting behavior of the LED light source was controlled, so that the light output was highly matched with the tissue physiological response, ensuring that high-quality PPG signal acquisition and accurate SpO2 estimation were maintained even when perfusion was insufficient.

[0075] During data acquisition, the system refreshes the waveform status every 100ms, with an integration window of T = 6s, meaning the value of Π(t) at each time t is determined by the data from the previous 6 seconds. The device Mr. Wang was wearing recorded the following waveform characteristic data segment (a simplified time series is used for clarity): Within the integration interval [4s, 10s] at time t0 = 10s, the following mean sequence was acquired: S(ξ) = 0.012V / ms, V(ξ) = 0.093, Δ(ε) = 0.16s. 2 Substituting into the scoring function model:

[0076]

[0077] Numerically, the integral can be performed using the equally spaced numerical integration method (with an integration interval of 0.5 s). The system accumulates the integral results within the function point by point, obtaining six integral terms, each with a value of approximately 0.0071, 0.0070, 0.0072, 0.0073, 0.0071, and 0.0072, for a total integral of approximately 0.0429. Setting the normalization constant Λ = 0.2, the final result is:

[0078]

[0079] The perfusion threshold θ is set to [0.4, 0.6], and the system defaults to the median θ = 0.5. Since Π(t0) < 0, it is determined that Mr. Wang's current perfusion status is poor, and the system immediately triggers the dynamic light compensation control module.

[0080] After receiving this determination result, the light source driver, which had initially set the LED near-infrared channel drive current to 10mA, then followed a step-by-step increment + feedback observation strategy, performed three incremental supplementary lighting operations, increasing the current by 1.5mA every 0.5 seconds, until reaching 15mA. Simultaneously, the system recorded the waveform changes during the supplementary lighting period and continuously calculated Π(t) to observe its response trend. For example, recalculating Π(t) 2 seconds after the supplemental lighting (t = 12s) yields an integral value of 0.31, which is approximately 0.1 higher than the initial state, and the slope is... A positive value indicates that the supplemental lighting is effective. Based on the control strategy, the system maintains a constant LED current at this point, entering a stable output state.

[0081] If Π(t) is found to stagnate or fluctuate during subsequent operation, and floats back and forth within ±0.01 in the short term, that is... The system maintains the current light intensity and pulse frequency unchanged and activates the microscopic thermal management module to limit the LED temperature rise. If monitoring continues and Π(t) is found to be continuously rising, reaching 0.56 at the 15th second, which is higher than the threshold and continues to rise, the system determines that tissue perfusion has significantly improved. Therefore, according to the energy-saving strategy, the LED current is reduced from 15mA to 12mA, and the duty cycle is adjusted from 40% to 30% to save power consumption and prevent tissue overexposure.

[0082] Furthermore, during dynamic supplemental lighting, the system records the physiological response data of tissues to light behavior and extracts their morphological structures in the high-response band, including the time of acromion appearance, the emergence of second-order waveforms, and steep increases in slope, to guide the subsequent signal enhancement module in selecting signal reconstruction templates. When a supplemental lighting behavior is confirmed to have caused a structural change, the system marks it as an effective response supplemental lighting segment, prioritizes its inclusion in the SpO2 estimation algorithm input, and eliminates low-response or non-response segments to further improve the estimation reliability.

[0083] Throughout the process, the comprehensive scoring function Π(t) provides a dynamic, continuous, and interpretable means of perfusion assessment, which is completely different from the traditional method based on single-point AC / DC ratio hard triggering. This invention integrates three physiological levels of indicators—upward momentum, volumetric response, and microcirculation rhythm—through a nonlinear integral structure, enabling the device to respond to real physiological changes. Simultaneously, its control logic, through continuous trend judgment feedback, makes the LED light source both energy-efficient and physiologically adaptable, truly constituting a dynamic physiological response-driven light compensation system. This is particularly suitable for individuals like Mr. Wang who are chronically underperfused, achieving medical-grade precision for continuous home blood oxygen monitoring.

[0084] During Mr. Wang's continuous use, the accuracy of blood oxygen saturation estimation under low perfusion conditions was further improved. Based on the traditional ratio method, a response weighting mechanism was introduced. By analyzing the degree of morphological change in the photoplethysmography waveform before and after supplemental lighting, the physiological response intensity of the tissue to light stimulation was quantified. This response level was then used as a weighting factor to correct the output results in the traditional ratio method, thereby offsetting signal fluctuations or abnormal ratio shifts caused by insufficient perfusion. During data acquisition, the device continuously acquired reflected light signals at two wavelengths, 660nm and 880nm, at a sampling rate of 250Hz. The light source control module precisely recorded the timestamps of LED supplemental lighting start and end, enabling the system to accurately divide the PPG waveforms before (e.g., the first 2 seconds), during (e.g., 1-3 seconds), and after (e.g., the last 3 seconds) supplemental lighting. Subsequently, the system preprocessed these waveform segments, including first-order difference detrending, wavelet packet denoising, and amplitude normalization, to ensure that each waveform segment had the same data scale and comparability. Next, the system extracts multi-dimensional features such as slope, amplitude, and morphological structure (e.g., the appearance of a shoulder peak, bimodal structure) from the waveforms before and after illumination, and calculates the morphological similarity changes between the waveforms, including waveform mean square error, changes in morphological gradient direction, and frequency domain energy redistribution. Then, a trained lightweight model is used to map these differences into response weight values ​​between 0 and 1. This model is a small neural network consisting of two fully connected hidden layers. The input is the feature difference vector of the waveforms before and after illumination, and the output is the response weight value. During training, approximately 10,000 sets of manually labeled data are used. The data labels are set based on whether the tissue shows a significant response (e.g., by empirical perfusion indicators or by human experts to determine whether it is a true response band label). The loss function is weighted binary cross-entropy, and the optimizer is Adam. Data augmentation strategies are incorporated during training to improve the model's generalization ability at low signal-to-noise ratios. During operation, in one of Mr. Wang's tests, after the supplemental lighting was completed, the system calculated that the near-infrared waveform's upward slope increased by 38%, the peak-to-valley amplitude increased by 22%, the probability of acromion appearance increased to 80%, and the morphological similarity index increased from 0.42 to 0.74. The model's output response weight was 0.82, indicating a significant tissue response to supplemental lighting stimulation. Traditional ratio methods typically estimate SpO2 using only the AC / DC ratio of red and near-infrared waveforms. However, under low perfusion conditions, this ratio can be distorted or underestimated due to insufficient light absorption or waveform flattening. This traditional ratio R is multiplied by the response weight and then fed into a correction function for adjustment. A higher weight indicates more reliable waveform quality and a result closer to the true blood oxygenation level; conversely, a lower weight corresponds to a larger ratio deviation and a stronger correction.Taking this case as an example, the traditional ratio method estimated SpO2 to be 91%, while after response weight correction, the system estimated it to be 93.4%, which is very close to Mr. Wang's blood oxygen value under normal conditions and consistent with the 93% measured by the finger-clip pulse oximeter. The error was controlled within ±0.5%, which is far better than the conventional solution under the no-response correction strategy.

[0085] Example 2:

[0086] Combined with appendix Figure 5 As shown in Example 1, during Mr. Wang's long-term fingertip blood oxygen monitoring, his chronic diabetes combined with peripheral microvascular sclerosis caused frequent fluctuations in local finger perfusion, with most of the time spent in a state of low perfusion. The dynamic light compensation control system embedded in the device periodically emits light stimulation at a preset rhythm, triggering LED supplemental lighting every 30 seconds. This behavior is divided into three time windows: before supplemental lighting (first 3 seconds), during supplemental lighting (last 2 seconds), and after supplemental lighting (last 5 seconds). These three waveforms are analyzed separately to determine whether the tissue makes a real blood flow response to the light stimulation.

[0087] During a monitoring session, the system triggered the 20th round of supplemental lighting at the 10-minute mark. The LED driver emitted a 16mA current to the near-infrared channel, generating a 2-second wide square wave supplemental lighting pulse, while simultaneously recording the corresponding waveform. Within the 3-second period before supplemental lighting, the system extracted the peak-to-valley amplitude difference of the PPG waveform, finding it to be 0.072V. The calculated average phase delay of the main pulse period was 0.35 seconds, and the waveform profile showed a smooth contour with no obvious shoulder peaks or second-order rising components. During the supplemental lighting period, the amplitude increased to 0.094V, and the phase advanced to 0.29 seconds. Simultaneously, the rapid rising segment on the left shoulder of the waveform became noticeably steeper, and the morphology recognition module identified two reflection-like peaks, which the system determined to be a high-response signal. Within 5 seconds after supplemental lighting, the amplitude slightly decreased to 0.085V, but remained higher than before supplemental lighting. The phase remained at 0.30 seconds, and the morphological structure retained a clear sharp-angle rising segment before the peak. Furthermore, the standard deviation of the waveform's micro-fluctuation period decreased to 0.08 seconds, indicating that capillary circulation was activated after supplemental lighting.

[0088] The system then compares the differences in amplitude, phase, and morphology of these three waveforms, and calculates the response score using an internally constructed difference mapping function. For example, the amplitude response score is (0.094-0.072) / 0.072 = 0.305, which is an improvement of 30.5%; the phase shortening response ratio is (0.35-0.29) / 0.35 = 0.171, which is a shortening of 17.1%; the morphological changes are compared using encoded vectors from a trained neural network model, and the similarity score increases from 0.41 to 0.74. The overall response score is synthesized to 0.78 (score range 0-1). The system sets the physiological response threshold to 0.65, therefore this supplementary lighting is judged as a significant response.

[0089] Based on this assessment, the system uses the PPG waveforms during and after supplemental lighting as the blood oxygen estimation signal sources for this cycle, while retaining the signal before supplemental lighting as a baseline to establish contextual differences. The system calculates the standard AC / DC ratio for the waveforms in both time periods as 0.043 for red light and 0.027 for near-infrared light, with a conventional R-value of 1.59, resulting in an initial estimate of 91.7%. Subsequently, the response score of 0.78 is used as a weighting factor, and the conventional R-value is fine-tuned through curve fitting and input into the response correction model, yielding a corrected blood oxygen saturation of 93.1%, consistent with Mr. Wang's clinical blood oxygen measurement at rest.

[0090] Conversely, in the 17th round of supplemental lighting events previously observed by the system, the waveform after supplemental lighting was not significantly different from that before supplemental lighting, with the amplitude increasing by only 2.1% and the phase remaining almost unchanged. The morphological feature score slightly increased from 0.43 to 0.46, and the total response score was 0.19, far below the threshold. Based on this, the system judged this supplemental lighting event as an invalid response, and the entire waveform signal was downweighted and retained only for data feature extraction and training samples, without being used for real-time SpO2 estimation.

[0091] By employing a process of segmented extraction, feature analysis, scoring, and dynamic optimization input, waveform segments truly relevant to physiological perfusion changes are efficiently selected. This process dynamically guides LED power control and estimation decisions, completely eliminating the error accumulation caused by traditional methods that rely on averaging across the entire segment or single-factor judgment based on signal strength. This ensures stable, accurate, and energy-controlled continuous blood oxygen monitoring even in patients like Mr. Wang with chronic hypoperfusion, providing a highly reliable solution for home-based chronic disease management and remote vital sign identification.

[0092] The amplitude and phase characteristics of the pulse waveform are directly used to determine whether the tissue has experienced enhanced reflection and changes in response speed to the supplemental lighting, thereby inferring the true physiological response of local blood flow to light stimulation, deciding whether to use this data segment for blood oxygen estimation, and dynamically adjusting the light compensation strategy accordingly. During the data acquisition phase, the device's LED emits modulated pulses of 660nm and 880nm wavelengths, which are collected by a near-contact photodetector to capture fingertip reflection signals. The sampling frequency for each wavelength is 250Hz. Each supplemental lighting event lasts for 2 seconds, with 3 seconds of data collected before and after the supplemental lighting, forming three complete time window data blocks. The raw PPG waveform undergoes data preprocessing, including bandpass filtering to remove baseline drift (filter range 0.5Hz-5Hz), wavelet denoising (db4 three-level decomposition), and normalization to ensure that peak and valley characteristics are comparable across segments. Amplitude feature extraction involves the system identifying local extreme points within each complete cardiac cycle using a period detection algorithm, measuring the amplitude difference between each pair of adjacent peaks and troughs, and normalizing the DC baseline value within that cycle to obtain a standardized amplitude value. In Mr. Wang's 21st supplemental lighting cycle, the near-infrared channel amplitude was 0.064V, 0.066V, and 0.062V in the three cycles before supplemental lighting, respectively. During the two cycles of supplemental lighting, this amplitude increased to 0.081V and 0.087V, respectively, with an amplitude increase between 25% and 40%. The system uses this result as significant evidence of enhanced light reflection and marks it as a high-amplitude response. Phase feature extraction involves timing each pulse initiation point, calculating the RR interval for each cycle, and determining the relative position change of the phase initiation point. The system uses the first derivative positive-negative-zero crossover method to detect the waveform slope reversal point, thereby determining the cardiac initiation point. Subtracting this from the initiation point of the previous cycle yields the phase delay or advance value. For example, before Mr. Wang received supplemental light, the average start-up time delay was 370ms. However, during supplemental light, due to the increased local capillary blood flow velocity, the waveform start-up time after reflection enhancement was advanced to 312ms. The system measured a phase advance value of 58ms, with a phase change ratio of 15.7%. This indicates an accelerated tissue response speed, a direct manifestation of local perfusion enhancement triggered by supplemental light. The above two features were encoded as input vectors and fed into the response recognition module, which is a lightweight multilayer perceptron (MLP). The network structure includes an input layer (two nodes: amplitude change ratio and phase advance value), two hidden layers (8 nodes per layer, with the activation function being Swish), and an output layer consisting of Sigmoid nodes that output a response score (ranging from 0 to 1). The training data for the model comes from more than 18,000 supplemental light cycle samples collected from real wearers. Doctors labeled the samples as responding or not responding based on changes in waveform morphology and the degree of perfusion improvement. During training, weighted binary cross-entropy was used as the loss function, Adam was used as the optimizer, and 100 training rounds were conducted. The accuracy on the validation set reached 92.7%.The model is ultimately deployed on the terminal chip, with a response time of no more than 15ms, which can meet the real-time judgment requirements of edge devices.

[0093] In Mr. Wang's supplemental lighting session, the model output response score was 0.84, far exceeding the set response threshold of 0.6. The system immediately marked this waveform segment as high-trust data and included it in the blood oxygen estimation. Simultaneously, the system input the amplitude boost ratio and phase advance value as feedback signals to the light compensation scheduling module, maintaining the current LED power and supplemental lighting rhythm unchanged. Conversely, in the 18th round of supplemental lighting, the amplitude change was less than 10%, the phase remained unchanged or even slightly delayed, and the score was only 0.22. The system downweighted this data segment and excluded it from the SpO2 calculation process.

[0094] During Mr. Wang's 22nd consecutive cycle of hypoperfusion oxygenation monitoring, the system initiated a standard LED dynamic light compensation operation. The device output 880nm near-infrared pulsed light with a 12mA drive current for 2 seconds. This supplementary lighting was divided into three time periods: a pre-supplementary window (3 seconds), a mid-supplementary window (2 seconds), and a post-supplementary window (5 seconds). Three core morphological features were extracted from the photoplethysmogram acquired in each window: waveform contour characteristics, rising edge slope, and micro-fluctuation structure. The system then constructed a composite waveform function based on these morphological factors and performed integral difference function calculations to determine whether this supplementary lighting triggered actual microcirculatory physiological changes.

[0095] Within the pre-illumination window, the system detected that Mr. Wang's near-infrared waveform exhibited a slightly flat-topped shape, with a waveform symmetry of 0.61 (calculated based on the normalized left-right area ratio). No obvious shoulder peak was formed, the steep rise lasted approximately 84 ms, the rising slope was 0.011 V / ms, and the micro-fluctuation period was irregular with a standard deviation of 0.15 s. Based on this, the system constructed the waveform morphology function χ. b The (τ) value was 0.48 (unit interval normalized). During the supplemental lighting period, significant changes in the waveform profile were observed: symmetry increased to 0.78, the left shoulder formation time was advanced by 18 ms, the steep rise was shortened to 61 ms, the slope increased to 0.017 V / ms, and periodic activation of the micro-fluctuation structure was observed. The autocorrelation value increased from 0.42 to 0.69, and the standard deviation decreased to 0.07 s. The corresponding morphological function χ... s (τ) reaches 0.77. After supplemental lighting, the waveform in the window slightly decreases, but the outline remains clear, the symmetry remains at 0.74, the slope slightly decreases to 0.015V / ms, and the micro-fluctuation structure still maintains good periodicity. Overall, χ² is... r (τ) = 0.69.

[0096] At this point, the system performs the following integral difference function calculation:

[0097]

[0098] The integration interval was set from t1 = 0s (before supplemental lighting) to t3 = 10s (after supplemental lighting), with sampling every 0.5 seconds, generating 20 integration points. The physiological reliability weighting function Ω(τ) was generated by the rhythm exponent (based on heart rate stability and micro-fluctuation synchronicity), with a typical value set between 0.8 and 1.2 for each integration point. After calculation, the total difference value ΔΨ = 0.183 was obtained after integration of the waveform response, while the system's preset response threshold ε was 0.12 (the threshold range is generally set between 0.10 and 0.15, and the actual value can be dynamically adjusted to match individual waveform noise characteristics).

[0099] Since ΔΨ = 0.183 > ε = 0.12 in this supplemental lighting action, the system determines that this action is an effective physiological response supplemental lighting. The system immediately marks the waveform of this time period as a high-confidence segment, enters the main path of blood oxygen estimation, and keeps the current LED power and pulse rhythm unchanged to continue the effective response strategy. Conversely, if ΔΨ is lower than 0.08, for example, in Mr. Wang's 19th round of supplemental lighting, the system finds that the waveform in the supplemental lighting only has a slight increase in slope, no significant change in symmetry, and the micro-fluctuation structure is still in a disordered state, and the resulting integral difference is only 0.06. Based on this, the system determines that it is an ineffective response supplemental lighting, and then in the next round, the LED current is reduced to 10mA and the duty cycle is adjusted from 40% to 25% to avoid the adverse effects of excessive supplemental lighting on tissue heat load or energy consumption.

[0100] Furthermore, each morphological sub-function in the integral difference function is generated through training. The system uses a large amount of real user sample data (more than 25,000), with response and non-response waveforms manually labeled. The distribution functions of each morphological feature are trained in the three-dimensional feature space to construct a weighted fusion function χ(τ). The model is then fitted and solved in real time on the edge device using a polynomial approximation, enabling a complete morphological response calculation to be completed once per second, even on low-power chips, truly achieving edge intelligence + physiological realism.

[0101] During Mr. Wang's 23rd round of continuous blood oxygen monitoring, the system successfully stimulated a microcirculatory response in his fingertips through dynamic LED supplemental lighting. To further improve the accuracy of blood oxygen estimation under low perfusion conditions, a response reliability scoring mechanism based on the differences in three waveform responses was designed. This mechanism weights the waveform segments collected in each round in the SpO2 estimation, distinguishing between strong response segments, weak response segments, and no response segments. This maximizes the use of data with true physiological reference value and filters out data segments that may be affected by noise disturbances, optical reflection failures, or physiological rest. In use, each LED light compensation action is explicitly calibrated by the system, dividing it into three windows: before supplemental lighting (e.g., t1 to t2), during supplemental lighting (t2 to t3), and after supplemental lighting (t3 to t4). Within each window, the system extracts multi-dimensional features from the PPG reflection waveform, including three main axes: amplitude, phase, and morphology, totaling nine secondary features, such as peak-to-valley amplitude difference, slope change rate, symmetry index, and micro-fluctuation periodic stability. The system compares the differences of these three data segments dimension by dimension to form a response difference vector Δ, where each dimension represents the degree of significance of a certain type of physiological parameter before, during, and after supplemental lighting. Subsequently, the system uses a lightweight multi-input scoring network to construct a response confidence scoring function, inputting the Δ vector into the model to output a normalized weight value R between 0 and 1. s (ResponseScore). The model structure is a three-layer perceptron network: the input layer contains 9 nodes, the two hidden layers contain 16 and 8 neurons respectively, the activation function adopts a hybrid structure of Tanh and Swish, and the output layer has 1 Sigmoid node. The training data comes from more than 37,000 dynamic light compensation sample segments. The supervision samples are formed by expert labeling (whether there is a real tissue response). The system is trained with weighted cross-entropy as the loss function, and the real SpO2 reference value is introduced as an auxiliary label to improve the model's ability to identify error risk segments. After training, the model is quantized and compressed and deployed on MCU, with a computation latency of less than 30ms. In Mr. Wang's supplementary lighting data, the average amplitude of the window before supplementary lighting was 0.066V, the slope was 0.011V / ms, and the waveform symmetry was 0.58. During supplementary lighting, these values ​​increased to 0.083V, 0.016V / ms, and 0.75 respectively. After supplementary lighting, they remained at a similar high level. The system calculates the response confidence score R through the feature difference vector Δ. s=0.87, the system considers this supplementary lighting response to be strong and the waveform reliability to be high. The waveforms from the first and latter parts of the supplementary lighting are input into the main channel for blood oxygen estimation, and a weighting coefficient of 0.87 is assigned to them. In contrast, in the 21st round of supplementary lighting, the amplitude only increased by 7%, the slope remained almost unchanged, the symmetry improvement was less than 0.1, the Δ vector difference was minimal, and the final response score was only 0.22. The system marked this waveform as low-confidence data, and its weighting coefficient was reduced to 0.22 or it was directly removed when participating in SpO2 estimation. This response scoring mechanism ultimately applies to the fusion calculation model of blood oxygen estimation; that is, the SpO2 estimation result output by the traditional ratio method will be adjusted according to the current response score R. s Linear or nonlinear weighting can be applied, or the system can be used as a sample importance factor in the regression weight allocation of machine learning estimation models. Testing after Mr. Wang wore the system for 30 minutes showed that the mean squared error of the estimation after introducing response scores was reduced by 38% compared to the original ratio method, with the error decreasing by more than 50% during low-perfusion (PI < 0.3) periods. Furthermore, regarding device power consumption, because the supplementary lighting rhythm is continued only during high-response periods, and power is reduced or the supplementary lighting frequency is decreased during low-response periods, the overall energy saving rate is improved by more than 25%, significantly extending battery life in wearable long-term working environments.

[0102] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting low perfusion oxygen saturation based on dynamic light compensation, characterized in that... Includes the following steps: An optical characterization of erythrocyte density in tissues is dynamically determined by constructing a combined characteristic relationship between multi-wavelength reflection intensity and fluctuation amplitude; the current perfusion level is inferred by the morphological characteristics of the reflection waveform, including pulse edge slope, peak-to-valley ratio, and micro-fluctuation cycle. The perfusion optical index is established to determine whether the current tissue is in a low perfusion state; when the perfusion state index is lower than the preset threshold, the dynamic light compensation control module is triggered to adjust the incident light power. The time series modeling method was used to analyze the dynamic changes in light reflection over a period of time. It also dynamically adjusts the LED driving strategy to form a physiological rhythm-like supplemental lighting behavior through three modes: gradual increase, pulse stimulation, and periodic maintenance. Within the time windows before, during, and after the supplemental lighting, the differences in amplitude, phase, and morphology of the capacitance-recorded waveforms of the three light segments are calculated respectively. Dynamic morphological similarity discrimination is introduced. By calculating the distance between the stimulation window and the morphological features of the windows before and after, the quantification of whether each supplemental lighting truly causes a tissue response is achieved, and the distinction between real physiological fluctuations and noise disturbances is made. For low-response supplemental lighting, the supplemental lighting intensity is reduced in response. Conversely, high-response supplemental lighting is used to extract key deformation features for signal reconstruction; The pulse component and baseline component of each wavelength signal are separated. Combined with the weights set by the perfusion optical index, the ratio conversion of the pulse information of different wavelengths is performed. The pre-calibrated conversion model is called to output the blood oxygen saturation value. At the same time, the result quality index based on waveform morphology similarity score is given as the final detection result.

2. The method for detecting low perfusion oxygen saturation based on dynamic light compensation according to claim 1, characterized in that... The method for establishing perfusion optical parameters includes: S1. Collect multi-wavelength light reflection signals from the target detection area and obtain pulse photoplethysmography waveform; S2. Extracting morphological features from the reflected waveform, including: a) The slope of the rising edge of the pulse. b) The amplitude ratio of peak to trough pulse values c) The repeatability and similarity of micro-fluctuation cycles; S3. During the execution of light compensation, continuously monitor the changes in waveform morphology, judge the effect of light compensation, and adjust the light output accordingly; calculate the blood oxygen saturation estimate based on the reflected signal after light compensation optimization.

3. The method for detecting low perfusion oxygen saturation based on dynamic light compensation according to claim 2, characterized in that... The extraction of the pulse edge slope includes analyzing the time and amplitude change rate of the rising edge of the photoplethysmography waveform to determine the blood propulsion speed and tissue photoresponse speed.

4. The method for detecting low perfusion oxygen saturation based on dynamic light compensation according to claim 3, characterized in that... The amplitude ratio of the peak value to the trough value is used to represent the magnitude of tissue volume change within the cardiac cycle; the smaller the amplitude, the weaker the perfusion. The micro-fluctuation cycle is a secondary fluctuation outside the main pulse waveform, used to determine capillary reflux or vascular elasticity in the tissue.

5. The method for detecting low perfusion oxygen saturation based on dynamic light compensation according to claim 4, characterized in that... The perfusion optical index is a comprehensive score calculated based on multiple waveform features, used to reflect the real-time tissue perfusion quality; the dynamic light compensation control module includes a light source driver and a feedback controller, used to dynamically adjust the power and pulse rhythm of the LED light source according to the perfusion optical index.

6. The method for detecting low perfusion oxygen saturation based on dynamic light compensation according to claim 5, characterized in that... The blood oxygen saturation calculation process introduces response weights based on the changes in waveform morphology before and after light compensation, which are used to correct the estimation bias of the traditional ratio method.

7. The method for detecting low perfusion oxygen saturation based on dynamic light compensation according to claim 1, characterized in that... The method for calculating the differences in amplitude, phase, and morphology of the capacitance-recorded waveforms of the three light segments includes: Dynamic light compensation signals are sent to the detected tissue at a preset rhythm; each light compensation action is divided into three time windows: before light compensation, during light compensation, and after light compensation; multi-dimensional features of amplitude, phase, and morphology are extracted from the photoplethysmography waveforms collected in each time window. The characteristics of the three waveforms are compared to determine the differences in the physiological response of tissues to supplemental lighting. The waveforms with physiological response characteristics within the supplemental lighting cycle are used for blood oxygen saturation estimation, while the remaining waveforms are downweighted or discarded.

8. The method for detecting low perfusion oxygen saturation based on dynamic light compensation according to claim 7, characterized in that... The amplitude characteristics include the amplitude variation between the peak and trough of the pulse waveform, used to determine whether supplemental lighting causes increased tissue reflection; the phase characteristics include the advance or delay of the pulse wave start time relative to the previous cycle reference point, used to assess the tissue's response speed to supplemental lighting.

9. The method for detecting low perfusion oxygen saturation based on dynamic light compensation according to claim 8, characterized in that... The morphological features include changes in the contour of the pulse waveform, the slope of the rising edge, and the appearance and disappearance of micro-wave structures. By constructing an integral function, the differences in the morphology of the three waveforms are quantified to determine whether the supplemental lighting caused physiological changes in tissue microcirculation. ； in: : Defined waveform response morphology difference index, used to measure the significance of physiological response; : These represent the start time of the window before illumination and the end time of the window after illumination, respectively; : Any point in time within the integration interval; The waveform composite morphology function within the window during supplemental lighting includes the fusion of contour factor, slope factor, and micro-fluctuation factor. , : Composite waveform morphology values ​​at corresponding time points before and after supplemental lighting; Physiological credibility weighting function; : Represents the absolute morphological shift between the three segments, indicating the net change in the waveform caused by supplementary lighting; when Exceeding the system's preset response threshold If the condition is met, the supplementary lighting behavior is determined to be effective physiological response supplementary lighting; otherwise, it is considered ineffective supplementary lighting or optical over-supplementation, and the LED intensity in the next supplementary lighting cycle will be reduced or the supplementary lighting duration will be shortened.

10. The method for detecting low perfusion oxygen saturation based on dynamic light compensation according to claim 9, characterized in that... The response reliability score is constructed by the response differences of the three waveforms, and different waveforms are weighted in the blood oxygen saturation calculation. The waveform segments collected in each round are weighted in the SpO2 estimation to distinguish between strong response segments, weak response segments, and no response segments.

Images