Light stimulation effectiveness evaluation method and system based on multi-modal feature fusion

Abstract

This invention discloses a method and system for evaluating the effectiveness of photostimulation based on multimodal feature fusion, belonging to the field of big data analysis technology. The method and system include the following steps: S1, acquiring raw observation data and performing time axis relocation, dynamic denoising, normalization, and sliding time window segmentation; S2, evaluating the state of the stimulation process and selecting effective windows; S3, extracting neural activity response features and hemodynamic response features, and performing physiological response reliability assessment; S4, performing consistency fusion to evaluate the effectiveness of photostimulation. This invention effectively improves the stability and reliability of effectiveness evaluation during head-mounted photostimulation, solving the problems of difficulty in distinguishing between abnormal stimulation pathways and true brain tissue responses, as well as insufficient consistency of results across batches of experiments.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, specifically to a method and system for evaluating the effectiveness of light stimulation based on multimodal feature fusion. Background Technology

[0002] With the development of head-mounted photostimulation devices, multimodal wearable data acquisition devices, and the demand for brain function assessment, existing technologies are gradually using electroencephalogram (EEG) signals, near-infrared signals, or single-type physiological parameters for photostimulation process analysis. Current systems typically judge the effectiveness of photostimulation based on amplitude or trend changes of single-channel or single-type physiological signals, lacking unified modeling of factors such as fit, cumulative effective exposure, artifact levels, superficial interference levels, and cross-modal response temporal consistency. Therefore, they cannot provide a stable, verifiable, and reproducible basis for determining whether the photostimulation process has achieved its intended effect.

[0003] For example, invention patent CN111012325B discloses a precise light control system and method, including: a laser spot illumination subsystem providing laser spot illumination light; a spatial lithography exciter system providing spatial lithography laser; an imaging convergence subsystem receiving the laser spot illumination light and the spatial lithography laser, coupling the two, and providing an imaging field of view for laser speckle imaging; and an image processing subsystem including: a feedback adjustment module that outputs an adjustment signal to the spatial lithography exciter system based on the laser spot imaging result, causing it to output the adjusted spatial lithography laser; and a field-of-view matching module that spatially matches the adjusted spatial lithography laser field with the imaging field of view before the next imaging. By acquiring images in real time and providing feedback through the image processing subsystem to adjust the spatial lithography laser, the adjusted spatial lithography laser forms a new lithography laser, and new biological imaging is obtained based on the lithographic excitation, ultimately realizing a closed-loop light control and tracking system for real-time adjustment of the spatial lithography laser.

[0004] In existing technologies, the effectiveness of stimulation and the occurrence of response are usually determined based on the amplitude and trend changes of single-channel or single-type physiological signals. There is a lack of cross-channel observable criteria that can uniformly model the fit, dose attainment, artifact level and differences in contribution between superficial and deep layers. This makes it difficult to distinguish between signal changes caused by link anomalies and the actual response of brain tissue, leading to misjudgment and missed judgment, increased delay in confirming effective response, and reduced consistency and reproducibility of cross-batch experimental results. Therefore, it is impossible to provide a stable and verifiable basis for determining whether the photostimulation process has achieved an effective effect.

[0005] Therefore, in order to address the above problems, there is an urgent need for a method and system for evaluating the effectiveness of light stimulation based on multimodal feature fusion. Summary of the Invention

[0006] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method and system for evaluating the effectiveness of light stimulation based on multimodal feature fusion, which solves the problems of difficulty in distinguishing between abnormal stimulation links and the true response of brain tissue, as well as the lack of consistency in results across batches of experiments.

[0007] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a method and system for evaluating the effectiveness of light stimulation based on multimodal feature fusion, comprising: S1, acquiring raw observation data and performing time axis relocation, dynamic denoising, normalization, and sliding time window segmentation on the raw observation data to construct a fused sensing quality perception frame; S2, evaluating the stimulation process state based on the fused sensing quality perception frame, filtering effective windows for the process, and constructing a process state feature set; S3, extracting neural activity response features and hemodynamic response features based on the stimulation process state feature set, performing physiological response reliability evaluation, and filtering response reliability windows; S4, performing consistency fusion based on the stimulation process state feature set and the physiological response feature set to evaluate the effectiveness of light stimulation, identify the current light stimulation state, and generate a light stimulation effectiveness label.

[0008] Further, the specific process of acquiring raw observation data and performing time-axis repositioning, dynamic denoising, normalization, and sliding time window segmentation on the raw observation data to construct a fusion sensing quality perception frame is as follows: The entire process of performing light stimulation on the forehead region using a head-mounted light stimulation device is dynamically acquired to obtain raw observation data. The raw observation data includes light stimulation power, stimulation duration, scattered light fluctuation coefficient, cumulative target irradiation, scattered light centroid angle, baseline centroid angle, scattered phase offset, acceleration modulus change rate, angular velocity modulus change rate, EEG timing signal, and near-infrared raw light intensity signal. Unified time-division multiplexing control is performed on the near-infrared spectral measurement link and the light stimulation link. Within the near-infrared measurement sub-slot, the stimulation emission unit is turned off, and the near-infrared raw light intensity signal is acquired. Within the output sub-slot of the light stimulation, the near-infrared emitting unit is turned off, and the near-infrared detector is blanked or shielded to suppress crosstalk and saturation effects of the stimulation light on the measurement link. The original observation data is marked with a sampling time identifier by a unified clock reference, and time axis repositioning is performed based on the timestamp information of different sampling links. For data with inconsistent sampling rates, an interpolation resampling method is used to construct a unified time series to correct time misalignment problems caused by sampling frequency differences, buffer delays, or local frame drops in the acquisition link. The original observation data is dynamically denoised by a combination of bandpass filtering and notch filtering. The original observation data is normalized by min-max normalization based on the statistical range of the baseline segment before stimulation, and a sliding time window is used to construct a fusion sensing quality perception frame.

[0009] Furthermore, the specific process for evaluating the state of the stimulation process based on the fusion sensing quality perception frame is as follows: Based on the fusion sensing quality perception frame, the theoretical irradiation is obtained by multiplying the light stimulation power by the stimulation duration, and then multiplied by the scattered light fluctuation coefficient to obtain the current cumulative effective irradiation. The scattering angle offset is obtained by the difference between the centroid angle of the scattered light and the centroid angle of the baseline. The motion artifact risk value is obtained by normalizing the rate of change of acceleration modulus and the rate of change of angular velocity modulus and then weighting them. The absolute value of the deviation between the cumulative effective irradiation of the current time window and the target cumulative irradiation is taken and divided by the sum of the target cumulative irradiation and the zero-prevention term to obtain the dose deviation term. The scattered light angle offset is multiplied by the angle weighting coefficient to obtain the angle offset term. The scattered light phase offset is multiplied by the phase weighting coefficient to obtain the phase offset term. The motion artifact risk value is multiplied by the motion artifact weighting coefficient to obtain the motion disturbance term. The dose deviation term, angle offset term, phase offset term, and motion disturbance term are added together, and then the negative value of the exponent is taken to obtain the stimulation process state evaluation value.

[0010] Furthermore, the specific process of screening effective windows and constructing the process state feature set is as follows: the stimulus process state evaluation value is compared with the stimulus process state threshold in real time. When the stimulus process state evaluation value is less than the stimulus process state threshold, it is determined that the current time window has an abnormal state, the time window is excluded from the validity evaluation calculation, and the current time window is marked as an abnormal process window. When the stimulus process state evaluation value is greater than or equal to the stimulus process state threshold, it is determined that the current time window meets the stimulus process state constraint, the current time window is marked as a valid process window, and the stimulus process state feature set is constructed.

[0011] Furthermore, the specific process for extracting neural activity response features and hemodynamic response features based on the stimulus process state feature set is as follows: Based on the stimulus process state feature set, band filtering, notch filtering, and artifact suppression are performed on the EEG time-series signal to calculate the absolute power, relative power, and band ratio of the target frequency band, generating neural activity response features; optical density change calculation, differential path correction, and modified Beer-Lambert law inversion are performed on the near-infrared raw light intensity signal to obtain the changes in oxyhemoglobin, deoxyhemoglobin, and total hemoglobin, generating hemodynamic response features; the influence of tissue blood flow components on brain region response signals is reduced through channel regression, baseline drift removal, and interference suppression methods; and the quality enhancement processing of hemodynamic response features and neural activity response features is performed through outlier detection, missing point interpolation, failed channel marking, and motion artifact correction methods. The corrected EEG response feature sequences and corrected hemodynamic response feature sequences are generated. The proportions of abnormal sampling, failed channels, and interference residuals are obtained using threshold discrimination statistics, channel validity detection, and root mean square evaluation of residual signals, respectively. Then, the residual noise value is calculated using a weighted normalization fusion method. Based on the corrected EEG response feature sequences, the target frequency band power change amplitude is extracted to obtain the EEG response amplitude value. The ratio of the standard deviation to the absolute mean of the corrected EEG response feature sequences within the current time window is calculated to obtain the EEG response fluctuation value. Based on the corrected hemodynamic response feature sequences, the target response amplitude in the oxyhemoglobin change is extracted to obtain the hemodynamic response amplitude value. The ratio of the standard deviation to the absolute mean of the corrected hemodynamic response feature sequences within the current time window is calculated to obtain the hemodynamic response fluctuation value.

[0012] Further, the specific process of conducting a physiological response reliability assessment and screening the response reliability window is as follows: Multiply the EEG response amplitude value by the EEG amplitude coefficient to obtain the EEG response contribution term; multiply the hemodynamic response amplitude value by the blood amplitude coefficient to obtain the hemodynamic response contribution term; add the EEG response contribution term and the hemodynamic response contribution term to obtain the physiological intensity term; multiply the EEG fluctuation value by the EEG fluctuation coefficient to obtain the EEG fluctuation influence term; multiply the hemodynamic fluctuation value by the blood fluctuation coefficient to obtain the hemodynamic fluctuation influence term; multiply the response residual noise value by the residual coefficient to obtain... The residual noise influence term is obtained; the constant 1 is added to the EEG fluctuation influence term, hemodynamic fluctuation influence term, and residual noise influence term to obtain the comprehensive constraint term; the physiological intensity term is then divided by the comprehensive constraint term to construct the physiological response confidence assessment value; the physiological response confidence assessment value is compared with the physiological response confidence threshold in real time: when the physiological response confidence assessment value is less than the physiological response confidence threshold, the current time window is marked as the response unconfidential window; when the physiological response confidence assessment value is greater than or equal to the physiological response confidence threshold, the current time window is marked as the response confidence window, and the physiological response feature set is constructed.

[0013] Furthermore, the specific process for evaluating the effectiveness of photostimulation based on the consistency fusion of the stimulus process state feature set and the physiological response feature set is as follows: Based on the stimulus process state feature set and the physiological response feature set, a unified mapping is performed on the process state evaluation results and the physiological response evaluation results. The process state evaluation results serve as a pre-constraint, and only those within the effective process window are included in the photostimulation effectiveness evaluation. A sliding time window analysis is performed on the EEG response feature sequence to identify the first moment when the target frequency band power change reaches the neural activity response threshold after N consecutive sampling points, and the time difference between this moment and the start time of photostimulation is used to determine the neural activity response delay. The process also identifies the first moment when the target hemodynamic response amplitude reaches the threshold after M consecutive sampling points. The hemodynamic response delay is obtained by determining the first moment of the hemodynamic response threshold and the time difference between that moment and the start of photostimulation. The joint coupling term is obtained by multiplying the stimulus process state assessment value and the physiological response reliability assessment value by a constant k. The joint normalization term is obtained by adding the stimulus process state assessment value and the physiological response reliability assessment value, plus a zero-prevention term. The joint evaluation term is obtained by dividing the joint coupling term by the joint normalization term. The cross-modal response delay difference is obtained by taking the absolute value of the difference between the neural activity response delay and the hemodynamic response delay. The delay penalty term is obtained by multiplying the cross-modal response delay difference by a delay penalty coefficient. The photostimulation effectiveness assessment value is obtained by subtracting the delay penalty term from the joint evaluation term.

[0014] Furthermore, the specific process for identifying the current state of light stimulation is as follows: the light stimulation effectiveness assessment value is compared with the light stimulation effectiveness threshold. When the light stimulation effectiveness assessment value is greater than or equal to the light stimulation effectiveness threshold, it is determined that the stimulation is valid and the physiological response result is reliable within the current time window, and the current time window is identified as a valid stimulation window. If the light stimulation effectiveness assessment value is less than the light stimulation effectiveness threshold, the abnormal cause analysis stage is entered, and threshold judgments are made on the stimulation process state assessment value, the physiological response reliability assessment value, and the response delay difference, respectively.

[0015] Furthermore, the specific process for generating photostimulation effectiveness labels is as follows: the identified effective stimulation states are structurally bound with the current time window number, sampling time information, stimulation process state assessment value, physiological response reliability assessment value, and photostimulation effectiveness assessment value to generate time window-level photostimulation effectiveness labels; at the same time, a re-examination flag is set for time windows identified as abnormal, and the corresponding abnormality source is marked; after all time window assessment tasks are completed, the photostimulation effectiveness labels of each time window are summarized to generate photostimulation effectiveness statistical results.

[0016] Furthermore, a second aspect of the present invention provides a light stimulation effectiveness evaluation system based on multimodal feature fusion, applied to a light stimulation effectiveness evaluation method based on multimodal feature fusion, comprising: a multimodal data acquisition and preprocessing module for acquiring raw observation data and performing time axis repositioning, dynamic denoising, normalization, and sliding time window segmentation on the raw observation data to construct a fused sensing quality perception frame; a stimulation process state feature construction module for evaluating the stimulation process state based on the fused sensing quality perception frame, filtering effective windows, and constructing a process state feature set; a response feature extraction and quality enhancement module for extracting neural activity response features and hemodynamic response features based on the stimulation process state feature set, performing physiological response reliability evaluation, and filtering response reliability windows; and a multimodal consistency fusion evaluation module for performing consistency fusion based on the stimulation process state feature set and the physiological response feature set to evaluate the light stimulation effectiveness, identify the current light stimulation state, and generate a light stimulation effectiveness label.

[0017] Beneficial effects The present invention has the following beneficial effects: (1) This invention, by constructing a fitting state characterization quantity, a dose deviation quantity, a scattering distribution stability characterization quantity, and a motion artifact risk characterization quantity, and further forming a stimulation process state evaluation value, can determine in advance whether the current photostimulation link meets the working conditions, and solves the problem in the existing scheme that it is difficult to identify process abnormalities such as loose wearing, insufficient local fitting, irradiation detachment, and failure to meet the cumulative effective irradiation amount, thereby avoiding misjudging the signal changes caused by link abnormalities as effective responses.

[0018] (2) This invention extracts the absolute power, relative power and frequency band ratio of the target frequency band from the EEG time-series signal, and performs optical density change calculation, differential path correction and modified Beer-Lambert law inversion on the near-infrared raw light intensity signal to obtain the changes in oxyhemoglobin, deoxyhemoglobin and total hemoglobin, thus realizing the synchronous characterization of changes in neural activity and hemodynamics, and solving the problem that existing systems rely only on single channels or single types of physiological signals and cannot fully reflect the response to light stimulation.

[0019] (3) This invention maps the stimulus process state assessment value and the physiological response reliability assessment value into a unified assessment space, and combines the neural activity response delay and the hemodynamic response delay to assess the effectiveness of light stimulation. This achieves a dual constraint judgment on the validity of process conditions and the reliability of response results, and solves the problem of the separation of link state and response results in the existing scheme, making it difficult to form a unified validity conclusion.

[0020] (4) This invention generates time window-level light stimulation effectiveness labels, abnormal distribution results and effectiveness change trend results by structurally binding the light stimulation effectiveness assessment value, abnormal source of process state, low confidence source of physiological response and cross-modal response delay difference. It can provide verifiable data basis for misjudgment analysis, re-examination location and cross-batch experiment consistency analysis, and solves the problems of difficulty in tracing, comparing and reproducing the assessment results in the existing scheme.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] Figure 1 This is a flowchart of the photostimulation effectiveness evaluation method based on multimodal feature fusion of the present invention; Figure 2 This is a diagram of the optical stimulation effectiveness evaluation system based on multimodal feature fusion according to the present invention. Figure 3 This is a diagram showing the results of the stimulus process state assessment and effective time window determination in this invention. Figure 4 This is a graph showing the reliable evaluation results of the physiological response of the present invention; Figure 5 This is a diagram showing the comprehensive evaluation results of the photostimulation effectiveness of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figures 1-5 This invention provides a technical solution: a method for evaluating the effectiveness of light stimulation based on multimodal feature fusion, comprising: S1, acquiring raw observation data and performing time axis relocation, dynamic denoising, normalization, and sliding time window segmentation on the raw observation data to construct a fused sensing quality perception frame; S2, evaluating the stimulation process state based on the fused sensing quality perception frame, filtering effective windows for the process, and constructing a process state feature set; S3, extracting neural activity response features and hemodynamic response features based on the stimulation process state feature set, and performing a physiological response reliability assessment to filter a response reliability window; S4, performing consistency fusion based on the stimulation process state feature set and the physiological response feature set to evaluate the effectiveness of light stimulation, identify the current light stimulation state, and generate a light stimulation effectiveness label.

[0025] Specifically, the process of acquiring raw observation data and performing time-axis repositioning, dynamic denoising, normalization, and sliding time window segmentation on the raw observation data to construct fused sensing quality perception frames is as follows: The entire process of photostimulation, including the output behavior, local near-infrared scattering, electroencephalogram (EEG) activity, near-infrared hemodynamic changes, and wear-related motion disturbances of a head-mounted photostimulation device in the forehead region, was dynamically acquired to obtain raw observation data. This raw data included photostimulation power, stimulation duration, scattered light fluctuation coefficient, cumulative target irradiation, scattered light centroid angle, baseline centroid angle, scattering phase shift, acceleration modulus change rate, angular velocity modulus change rate, EEG timing signal, and raw near-infrared light intensity signal. The cumulative target irradiation was preset by the photostimulation protocol. The scattered light fluctuation coefficient was calculated using the coefficient of variation of scattered light intensity within a sliding time window and was used to characterize the degree of fluctuation in optical coupling efficiency. The centroid angle of the scattered light is calculated by weighted averaging of the received light intensity and geometric position of each channel of the scattered light detection array, and is used to characterize the spatial centroid of the scattered light distribution; the baseline centroid angle is obtained by averaging the centroid angle of the scattered light within the baseline segment before stimulation, and serves as a spatial reference for the scattering distribution; the scattering phase offset is calculated by the phase difference between the scattered light signal and the stimulation modulation signal; the rate of change of acceleration modulus is calculated by the first difference of the three-axis modulus of the accelerometer, and the rate of change of angular velocity modulus is calculated by the first difference of the three-axis modulus of the gyroscope.

[0026] A near-infrared scattered light detection unit is integrated around the photostimulation source to acquire scattered light signals in real time during photostimulation. Unified time-division multiplexing control is implemented between the near-infrared spectral measurement link and the photostimulation link. During the near-infrared measurement sub-slot, the stimulation emission unit is turned off, and the raw near-infrared light intensity signal is acquired. During the photostimulation output sub-slot, the near-infrared emission unit is turned off, and the near-infrared detector is either blanked or shielded to suppress crosstalk and saturation effects of the stimulation light on the measurement link. The near-infrared scattered light detection unit and the photostimulation execution unit form a surrounding integrated structure, enabling the scattered light signal to reflect whether irradiation has detached, whether the local adhesion state is abnormal, and whether the cumulative effective irradiation amount has reached the target range.

[0027] A unified clock reference is used to add sampling time identifiers to the raw observation data, and time axis repositioning is performed based on the timestamp information of different sampling links. Hardware triggering or network time protocols are used as the unified clock source, and microsecond-level timestamps are added to data packets of each link. The time axis of the data is aligned according to the timestamp sequence. For data with inconsistent sampling rates, an interpolation resampling method is used to construct a unified time series to correct time misalignment problems caused by differences in sampling frequency, buffer delays, or local frame drops in the acquisition links. The interpolation resampling method uses linear interpolation or cubic spline interpolation, and the target sampling rate is uniformly set to the least common multiple of the original sampling rates of all links. The raw observation data is dynamically denoised by a combination of bandpass filtering and notch filtering. Based on the statistical range of the pre-stimulus baseline segment, the raw observation data is normalized by minimum-maximum normalization, and a sliding time window is used to construct a fusion sensing quality perception frame. The pre-stimulus baseline segment is a continuous signal segment before the stimulus begins. The minimum and maximum values ​​are taken from the statistical extreme values ​​within the baseline segment. The window length and step size of the sliding time window are obtained based on the stimulus paradigm and sampling rate.

[0028] This implementation plan provides a unified and reliable data foundation for fit status identification, dose deviation determination, artifact risk assessment, neural activity response analysis, and hemodynamic response analysis. Through unified time-division multiplexing control, it can effectively suppress crosstalk and saturation effects of stimulating near-infrared light on the near-infrared measurement link, significantly improve the separability of the original near-infrared light intensity signal, and thus enhance the reliability of hemodynamic inversion results.

[0029] Specifically, the process of evaluating the state of the stimulus process based on the fused sensing quality perception frame is as follows: Based on the fusion sensing quality perception frame, the theoretical irradiation is obtained by multiplying the light stimulation power by the stimulation duration, and then multiplying by the scattered light fluctuation coefficient to obtain the current cumulative effective irradiation. The scattered light attenuation coefficient is used to characterize the proportion of energy loss caused by changes in scattering, absorption and coupling efficiency in the transmission path. The actual effective irradiation applied to the tissue is obtained by multiplying the current cumulative effective irradiation by the scattered light attenuation coefficient. The scattering angle offset is obtained by the difference between the centroid angle of the scattered light and the centroid angle of the baseline. The motion artifact risk value is obtained by normalizing the rate of change of acceleration modulus and the rate of change of angular velocity modulus and then weighting them.

[0030] The dose deviation term is obtained by taking the absolute value of the deviation between the cumulative effective exposure dose and the target cumulative exposure dose in the current time window, dividing it by the sum of the target cumulative exposure dose and the zero-prevention term; the angle deviation term is obtained by multiplying the scattered light angle offset by the angle weighting coefficient; the phase deviation term is obtained by multiplying the scattered light phase offset by the phase weighting coefficient; the motion artifact risk value is obtained by multiplying the motion artifact weighting coefficient; the dose deviation term, angle deviation term, phase deviation term, and motion disturbance term are added together, and the negative value of the exponent is taken to obtain the stimulus process state assessment value. The exponential function can map any non-negative input to the (0,1] interval, ensuring that the assessment value has a clear upper bound and is dimensionless; when any abnormal term increases, the assessment value decays rapidly exponentially, improving the sensitivity to abnormal states and facilitating the use of a unified threshold for process state discrimination.

[0031] The specific formula for calculating the stimulus process state assessment value is as follows: ; In the formula, This represents the evaluation value of the stimulation process status, used to characterize whether the current photostimulation link meets the working conditions; This indicates the current cumulative effective radiation dose, used to reflect the cumulative dose of the current stimulus actually irradiated into the brain tissue; This represents the target cumulative exposure dose, which is used as a reference for determining whether the dose has met the target. This term represents the zero-prevention term, which is obtained by combining the measurement error of scattered light intensity and the sampling accuracy. Its value range is a constant greater than zero, used to prevent the denominator from being zero. The scattering angle offset weighting coefficient is determined by statistically analyzing the impact of different angle offsets on link stability. Its value ranges from 0.1 to 1.0 and is used to adjust the contribution of angle offset to process status assessment. It represents the scattering angle offset, used to reflect changes in the optical path caused by loose bonding or local offset; The scattering phase shift weighting coefficient is obtained by statistical analysis of experimental data to determine the impact of different phase shifts on link stability. Its value ranges from 0.1 to 1.0 and is used to adjust the contribution of phase shift to process state assessment. This represents the scattering phase offset, used to reflect link instability caused by scattered light interference or changes in the path between the light source and the detector; The motion artifact weighting coefficient is determined by comparing the changes in motion amplitude and stimulus state evaluation value using a regression method. Its value ranges from 0.1 to 1.0 and is used to adjust for the influence of wearer's head movement or environmental disturbance on the evaluation value. This represents the motion artifact risk value, used to quantify the impact of wearer head movements or posture changes on the stability of the photostimulation link.

[0032] This implementation scheme achieves joint discrimination of adhesion status, dose compliance status, link stability, and motion artifact risk; by using unified constraints of dose deviation, angle offset, phase offset, and motion disturbance terms, and compressing the evaluation result range through exponential decay, it can quickly lower the evaluation value when any abnormal factor increases, improve the sensitivity to the identification of irradiation detachment, insufficient local adhesion, unstable scattering distribution, and head movement interference, and facilitate the use of a unified threshold to determine the stability of the stimulation process.

[0033] Specifically, the process of filtering the effective window and constructing the process state feature set is as follows: The stimulus process state assessment value is compared with the stimulus process state threshold in real time. When the stimulus process state assessment value is less than the stimulus process state threshold, it is determined that the current time window has at least one abnormal state, such as irradiation detachment, insufficient local adhesion, insufficient dose, unstable scattering distribution, or excessive risk of motion artifacts. The time window is excluded from the effectiveness assessment calculation and marked as an abnormal window. When the stimulus process state assessment value is greater than or equal to the stimulus process state threshold, it is determined that the current time window meets the stimulus process state constraints and is marked as a valid window. The valid window records the stimulus process state assessment value and the values ​​of each offset term for the current time window, constructing a stimulus process state feature set. The stimulus process state feature set includes the time window number, the start and end time of the time window, the valid window marker, the stimulus process state assessment value, the dose deviation term, the angle offset term, the phase offset term, and the motion perturbation term.

[0034] like Figure 3 The diagram showing the stimulation process state assessment and effective time window determination results illustrates the following: The upper part displays the curve of the stimulation process state assessment value change. The blue curve represents the stimulation process state assessment value corresponding to each time window, and the red dashed line represents the process state determination threshold. It can be seen that the assessment value is higher than the threshold in some time windows and lower than the threshold in others, indicating that the photostimulation link state fluctuates over time. The middle part displays the curve of the current cumulative effective irradiation dose change. The green curve represents the current cumulative effective irradiation dose calculated within each time window, and the red dashed line represents the target irradiation dose, visually reflecting the deviation between the current irradiation dose and the target range. The lower part shows the time window validity marking results. Green dots represent valid windows, and red dots represent invalid windows, showing that the corresponding time window is only determined to be valid when the stimulation process state assessment value meets the threshold condition. This diagram intuitively illustrates that based on the joint changes in the stimulation process state assessment value and the cumulative effective irradiation dose, continuous discrimination can be made regarding whether the photostimulation link meets the working conditions, and time window-level validity markings can be output, providing a preliminary basis for the fusion determination of physiological response reliability assessment and photostimulation effectiveness.

[0035] In this implementation scheme, the comparison results between the stimulus process state assessment value and the stimulus process state threshold can be directly converted into time window-level process effective window markers and process abnormal window markers. This transforms the originally discrete process state changes into continuous, traceable, and comparable time-series judgment results, thereby improving the identification efficiency of abnormal states such as irradiation detachment, insufficient local adhesion, dose failure, unstable scattering distribution, and excessive risk of motion artifacts. It also avoids abnormal time windows from entering the subsequent physiological response reliability assessment and photostimulation effectiveness fusion calculation process, reduces the interference of link anomalies on the final effectiveness judgment results, and improves the stability of the assessment process, the accuracy of the judgment results, and the verifiability of the result output.

[0036] Specifically, the process of extracting neural activity response features and hemodynamic response features based on the stimulus process state feature set is as follows: Based on the stimulus process state feature set, band filtering, notch filtering, and artifact suppression are performed on the EEG time-series signal. The absolute power, relative power, and band ratio of the target band are calculated to generate neural activity response features. The absolute power is converted into a dimensionless quantity after being processed by the maximum value normalization method within the band, and is used for unified fusion with other modal features. The passband range of the band filtering is preset according to the target frequency of neural activity. The center frequency of the notch filtering is set to the power frequency. Artifact suppression adopts threshold discrimination or independent component analysis methods. The absolute power, relative power, and band ratio of the target band are all dimensionless feature quantities. The near-infrared raw light intensity signal is subjected to optical density change calculation, differential path correction, and modified Beer-Lambert law inversion to obtain the changes in oxyhemoglobin, deoxyhemoglobin, and total hemoglobin, generating hemodynamic response features. The differential path factor of the differential path correction is 6.0 based on tissue optical properties, and the path length is 3.0 based on the light source-detector distance.

[0037] The influence of tissue blood flow components on brain region response signals was reduced using channel regression, baseline drift removal, and interference suppression. The regression coefficients of channel regression were estimated within the pre-stimulation baseline using the least squares method. Baseline drift removal employed a high-pass filter with a cutoff frequency of 0.01, and interference suppression utilized adaptive filtering or principal component analysis. Hemodynamic and neural activity response features were enhanced using outlier detection, missing point interpolation, invalid channel labeling, and motion artifact correction methods to generate corrected EEG response feature sequences and corrected hemodynamic response features. For sequence outlier detection, a threshold discrimination based on the standard deviation multiple is used, with a standard deviation multiple of 3. Missing point interpolation uses linear interpolation or spline interpolation. Failed channel marking is determined based on the signal variance being lower than the variance threshold or the signal-to-noise ratio being lower than the signal-to-noise ratio threshold. Motion artifact correction uses sliding window smoothing or regression correction based on acceleration signals. The proportion of outliers, the proportion of failed channels, and the proportion of interference residuals are obtained through threshold discrimination statistics, channel validity detection, and root mean square evaluation of residual signals, respectively. Then, the response residual noise value is calculated through a weighted normalization fusion method.

[0038] Based on the corrected EEG response feature sequence, the target frequency band power variation amplitude is extracted to obtain the EEG response amplitude value. The target frequency band power variation amplitude is based on the average power within the pre-stimulation baseline segment, and the difference or ratio between the power and the baseline during the stimulation period is calculated. The ratio of the standard deviation to the absolute mean of the corrected EEG response feature sequence within the current time window is calculated to obtain the EEG response fluctuation value. The ratio of the standard deviation to the absolute mean is dimensionless. Based on the corrected hemodynamic response feature sequence, the target response amplitude in the oxyhemoglobin variation is extracted to obtain the hemodynamic response amplitude value. The target response amplitude is based on the average value within the pre-stimulation baseline segment, and the difference between the peak or mean and the baseline during the stimulation period is calculated. The ratio of the standard deviation to the absolute mean of the corrected hemodynamic response feature sequence within the current time window is calculated to obtain the hemodynamic response fluctuation value.

[0039] In this implementation scheme, by constructing response residual noise values, EEG response amplitude values, EEG response fluctuation values, hemodynamic response amplitude values, and hemodynamic response fluctuation values, the response intensity, response stability, and residual interference degree can be uniformly transformed into quantifiable characterization results. This provides a consistent, reliable, and verifiable data foundation for the reliable assessment of physiological responses and the fusion determination of the effectiveness of light stimulation, thereby improving the accuracy of response identification, the consistency of cross-modal analysis, and the credibility of the final evaluation results.

[0040] Specifically, the process of conducting a reliable assessment of the physiological response and selecting a reliable response window is as follows: Multiplying the EEG response amplitude value by the EEG amplitude coefficient yields the EEG response contribution term; multiplying the hemodynamic response amplitude value by the blood amplitude coefficient yields the hemodynamic response contribution term; adding the EEG and hemodynamic response contribution terms yields the physiological intensity term; multiplying the EEG fluctuation value by the EEG fluctuation coefficient yields the EEG fluctuation influence term; multiplying the hemodynamic fluctuation value by the blood fluctuation coefficient yields the hemodynamic fluctuation influence term; multiplying the residual noise value by the residual coefficient yields the residual noise influence term; adding the constant 1 to the EEG fluctuation influence term, hemodynamic fluctuation influence term, and residual noise influence term yields the comprehensive constraint term; finally, dividing the physiological intensity term by the comprehensive constraint term constructs the reliable physiological response assessment value. By separating the physiological intensity term from the signal quality constraint term, a ratio form with both numerator and denominator being dimensionless quantities is constructed, ensuring the assessment value is strictly dimensionless and avoiding arbitrariness introduced by dimensional differences.

[0041] The specific formula for calculating the reliable assessment value of physiological response is as follows: ; In the formula, This represents a reliable assessment value for physiological response, used to characterize the reliability of EEG and hemodynamic signals within the current time window; The amplitude value represents the brain electrical response and is used to reflect the intensity of neural activity response. The larger the amplitude, the more obvious the brain electrical response. This represents the amplitude of the hemodynamic response, used to characterize changes in cerebral blood flow or the intensity of cerebral oxygenation response; This represents the fluctuation value of the brain electrical response, used to reflect the stability of the brain electrical response; This represents the fluctuation value of the hemodynamic response, used to reflect the stability of the blood flow response; This represents the residual noise value of the response, used to quantify the impact of residual noise in the system on physiological signals; The amplitude coefficient of brain electrical response is determined by statistically analyzing the contribution of brain electrical response amplitude to the reliable assessment result of physiological response, and then normalizing it. The value ranges from 0.1 to 1.0 and is used to adjust the contribution of brain electrical response to the reliable assessment value of physiological response. The blood amplitude coefficient is determined by statistically analyzing the contribution of hemodynamic response amplitude to the reliable assessment result of physiological response, and then normalizing it. Its value ranges from 0.1 to 1.0, and it is used to adjust the contribution of hemodynamic response to the reliable assessment value of physiological response. The EEG fluctuation coefficient is the inhibitory strength of EEG fluctuations on the reliable assessment results of physiological responses through the fluctuation influence calibration method, and is obtained after normalization. The value range is 0.1-1.0, and it is used to adjust the influence of EEG fluctuations on the reliable assessment values ​​of physiological responses. The blood fluctuation coefficient represents the inhibition strength of hemodynamic fluctuations on the reliable assessment results of physiological responses, and is obtained after normalization. Its value ranges from 0.1 to 1.0, and it is used to adjust the influence of hemodynamic fluctuations on the reliable assessment values ​​of physiological responses. The residual coefficient represents the suppression strength of residual noise on the reliable assessment results of physiological response, obtained after normalization. Its value ranges from 0.1 to 1.0 and is used to adjust the influence of residual noise on the reliable assessment value of physiological response.

[0042] The physiological response reliability assessment value is compared with the physiological response reliability threshold in real time. The physiological response reliability threshold is analyzed based on the receiver's operational characteristic curve, and its value range is in the same domain as the physiological response reliability assessment value. A window-by-window comparison is performed using the same time window granularity as the stimulus process state threshold. When the physiological response reliability assessment value is less than the physiological response reliability threshold, the current time window is marked as an unreliable response window. The unreliable response window is appended with the physiological response reliability assessment value, EEG response amplitude value, hemodynamic response amplitude value, and residual noise value of the current time window for anomaly tracing analysis. When the physiological response reliability assessment value is greater than or equal to the physiological response reliability threshold, the current time window is marked as a reliable response window. The reliable response window is appended with the physiological response reliability assessment value, EEG response amplitude value, hemodynamic response amplitude value, and values ​​of each fluctuation term of the current time window, constructing a physiological response feature set. The physiological response feature set includes the time window number, time window start and end times, reliable response window label, physiological response reliability assessment value, EEG response amplitude value, hemodynamic response amplitude value, EEG response fluctuation value, hemodynamic response fluctuation value, and residual noise value.

[0043] Table 1 shows the physiological response reliability assessment data, used to determine the reliability of physiological signals at each time window by combining EEG and hemodynamic indicators. Window 0: EEG response amplitude value 0.697, hemodynamic response amplitude value 0.031, hemodynamic response fluctuation value 1, physiological response reliability assessment value 0.398, reliability is FALSE; Window 1: EEG response amplitude value 0.696, hemodynamic response amplitude value 0.081, hemodynamic response fluctuation value 0.16, physiological response reliability assessment value 0.477, reliability is TRUE; Window 2: EEG response amplitude value 0.701, hemodynamic response amplitude value 0.033, hemodynamic response fluctuation value 1, physiological response reliability assessment value 0.398, reliability is TRUE. The confidence assessment value for the response is 0.402, and the confidence level is FALSE; Window number 3: EEG response amplitude is 0.702, hemodynamic response amplitude is 0.143, hemodynamic response fluctuation is 0.12, and the confidence assessment value for the physiological response is 0.522, and the confidence level is TRUE; Window number 4: EEG response amplitude is 0.703, hemodynamic response amplitude is 0.13, hemodynamic response fluctuation is 0.42, and the confidence assessment value for the physiological response is 0.493, and the confidence level is TRUE; Window number 5: EEG response amplitude is 0.703, hemodynamic response amplitude is 0.123, hemodynamic response fluctuation is 0.34, and the confidence assessment value for the physiological response is 0.495, and the confidence level is TRUE.

[0044] Table 1. Data on the Reliability Assessment of Physiological Responses like Figure 4The graph showing the reliable assessment results of the physiological response is presented as a bar chart, displaying the reliable assessment values ​​of the physiological response for each time window. The vertical axis represents the reliable assessment value of the physiological response, and the horizontal axis represents time. The orange dashed line represents the reliability threshold of 0.45, and the yellow shaded area represents the light stimulation execution interval of 20-40 seconds. The graph shows that before light stimulation, the reliable assessment values ​​of the physiological response in some time windows are close to or slightly lower than the reliability threshold, indicating that there are still some fluctuations or residual interference in the background signal before stimulation. After entering the light stimulation execution interval, the reliable assessment values ​​of the physiological response in most time windows are significantly higher than the reliability threshold and remain at a high level in the middle of the stimulation. This indicates that after EEG response feature extraction, hemodynamic response feature extraction, outlier correction, channel regression, baseline drift removal, and motion artifact suppression, the changes in neural activity and hemodynamic changes within the current time window have high reliability. After stimulation ends, the assessment values ​​of some time windows drop back to near or below the threshold, reflecting a weakening of the physiological response intensity or a renewed increase in noise influence. This figure visually illustrates that the reliability of physiological responses within each time window can be quantitatively determined based on the combined results of EEG response amplitude, hemodynamic response amplitude, response fluctuation, and residual noise, and provides a reliable basis for the fusion assessment of the effectiveness of light stimulation.

[0045] In this implementation plan, the previously scattered response intensity, response stability, and residual interference are integrated into a unified and comparable reliability assessment result. At the same time, by using a construction method that separates the amplitude contribution term from the fluctuation and noise constraint terms, the ability to distinguish between weak response, high fluctuation, and high residual noise states can be improved while maintaining the dimensionless nature and scale uniformity of the assessment value. This avoids time windows with large fluctuations or strong interference from entering the photostimulation effectiveness fusion assessment process, enhances the reliability of the joint assessment of changes in neural activity and hemodynamics, improves the accuracy, continuity, and verifiability of response screening results, and provides a stable response quality basis for photostimulation effectiveness assessment.

[0046] Specifically, the process of evaluating the effectiveness of light stimulation based on the consistency fusion of the stimulus process state feature set and the physiological response feature set is as follows: Based on the stimulus process state feature set and physiological response feature set, a unified mapping is performed on the process state assessment results and physiological response assessment results. The process state assessment results serve as a pre-constraint, and only those within the effective process window are included in the photostimulation effectiveness assessment. A sliding time window analysis is performed on the EEG response feature sequence to identify the first moment when the target frequency band power change reaches the neural activity response threshold after N consecutive sampling points. The time difference between this moment and the start time of photostimulation is used to determine the neural activity response delay. Here, N is the number of consecutive points that meet the target neural activity response threshold, set according to the sampling rate and response duration. A typical value is the number of sampling points corresponding to the sampling rate of 0.1 seconds to 0.3 seconds, used to filter out transient noise interference and ensure the reliability of response judgment. The first moment when the target hemodynamic response amplitude reaches the hemodynamic response threshold after M consecutive sampling points is identified, and the time difference between this moment and the start time of photostimulation is used to determine the hemodynamic response delay. Where M is the number of consecutive points that meet the target hemodynamic response amplitude, which is set according to the response rise time of the near-infrared signal and the sampling rate. The typical value is the number of sampling points corresponding to the sampling rate of 0.2 seconds to 0.5 seconds, which is used to avoid false triggering caused by single-point noise fluctuations and to ensure accurate identification of the hemodynamic response start time.

[0047] The joint coupling term is obtained by multiplying the stimulus process state assessment value and the physiological response reliability assessment value by a constant k. The constant k is an adjustment coefficient, preferably 2, used to balance the dimensional matching relationship between the joint coupling term and the joint normalization term, ensuring that the assessment value is mapped to a reasonable range. The joint normalization term is obtained by adding the stimulus process state assessment value and the physiological response reliability assessment value, plus a zero-prevention term. The joint evaluation term is obtained by dividing the joint coupling term by the joint normalization term. The joint evaluation term ranges from 0 to 1 and is used to characterize the degree of consistency coupling between the process state and the physiological response. The cross-modal response delay difference is obtained by taking the absolute value of the difference between the neural activity response delay and the hemodynamic response delay. The cross-modal response delay difference is multiplied by the delay difference penalty coefficient to obtain the delay penalty term. The delay penalty term is subtracted from the joint evaluation term to obtain the photostimulation effectiveness assessment value, which is used to determine whether the photostimulation has achieved an effective state.

[0048] By constructing a joint evaluation term using the ratio of the joint coupling term and the joint normalization term, the limitations of a single-dimensional criterion are avoided. A cross-modal response delay difference is introduced as a penalty term, incorporating the temporal synchronization of neural activity and hemodynamic response into the evaluation system. When the delay difference between the two modal responses is too large, the effectiveness score is automatically reduced, solving the problem that a single modal response cannot confirm the authenticity of the response. The output is a scalar evaluation value with clear physical boundaries, providing a quantifiable basis for engineering judgment and realizing the mapping of photostimulation effectiveness from multi-source heterogeneous evidence to a unified verifiable result quantity.

[0049] The specific formula for calculating the effectiveness assessment value of light stimulation is as follows: ; In the formula, This represents the evaluation value of the effectiveness of light stimulation, used to output a comprehensive judgment result of the current light stimulation state; This represents the evaluation value of the stimulation process status, used to characterize whether the current photostimulation link meets the working conditions; This represents a reliable assessment value for physiological response, used to characterize the reliability of EEG and hemodynamic signals within the current time window; The zero-prevention term is a small positive number selected based on the formula's numerical stability requirements. It is a constant with a value greater than zero, used to prevent the denominator from being zero. It indicates the response delay of neural activity and is used to characterize how quickly the brain signals respond to the current light stimulus; This represents the hemodynamic response delay, used to characterize how quickly a hemodynamic signal responds to a current light stimulus; The time delay difference penalty coefficient is obtained by combining the influence of cross-modal response asynchrony on the effectiveness judgment result. The value ranges from 0.1 to 1.0 and is used to adjust the influence of the difference between the neural activity response delay and the hemodynamic response delay on the light stimulation effectiveness assessment value.

[0050] In this implementation plan, the previously scattered process state constraints, physiological response reliability, and cross-modal time consistency are integrated into a single, continuous, and quantifiable evaluation value for the effectiveness of photostimulation. This avoids misjudgments caused by single-modal or single-dimensional criteria, improves the accuracy, stability, and verifiability of determining whether the photostimulation process meets the conditions for effective action, and provides a unified quantitative basis for abnormal state identification, time window-level label generation, and statistical result output.

[0051] Specifically, the process of identifying the current state of light stimulation is as follows: The photostimulation effectiveness assessment value is compared with the photostimulation effectiveness threshold, which is obtained based on clinical validation results and has the same range as the photostimulation effectiveness assessment value. When the photostimulation effectiveness assessment value is greater than or equal to the photostimulation effectiveness threshold, the stimulation within the current time window is determined to be valid and the physiological response result is reliable. The current time window is identified as the effective stimulation window, and the effective stimulation window is appended with the photostimulation effectiveness assessment value, stimulation process state assessment value, and physiological response reliability assessment value of the current time window. If the photostimulation effectiveness assessment value is less than the photostimulation effectiveness threshold, the abnormal cause analysis stage is entered. Threshold judgments are made for the stimulation process state assessment value, the physiological response reliability assessment value, and the response delay difference. The threshold judgment for the stimulation process state assessment value is based on the stimulation process state threshold, the threshold judgment for the physiological response reliability assessment value is based on the physiological response reliability threshold, and the threshold judgment for the response delay difference is based on the delay consistency threshold.

[0052] In the abnormal cause analysis stage, when the stimulus process state assessment value is lower than the process state threshold, it is identified as an abnormal process state caused by irradiation detachment, abnormal adhesion, insufficient dose, unstable scattering distribution, or excessive motion artifacts; when the physiological response confidence assessment value is lower than the response confidence threshold, it is identified as an unreliable response state caused by insufficient response amplitude, excessive response fluctuation, residual shallow interference, or channel failure; when the difference between the neural activity response delay and the hemodynamic response delay is greater than the delay consistency threshold, it is identified as a cross-modal response inconsistency state.

[0053] In this implementation plan, by using the effective stimulation window to record the light stimulation effectiveness assessment value, stimulation process state assessment value, and physiological response reliability assessment value, the problem of only outputting a single effective or ineffective conclusion without being able to explain the cause is avoided. This enhances the interpretability, traceability, and verifiability of the light stimulation state identification results, and provides a unified data basis for subsequent statistical summarization, anomaly distribution analysis, and full-process effectiveness evaluation.

[0054] Specifically, the process of generating a light stimulation effectiveness label is as follows: The identified effective stimulus states are structurally bound to the current time window number, sampling time information, stimulus process state assessment value, physiological response reliability assessment value, and photostimulation effectiveness assessment value. This structural binding uses key-value pairs or data frames, with the time window number as the primary key, associated with the start and end times of the stored time window, effective stimulus flags, stimulus process state assessment values, physiological response reliability assessment values, and photostimulation effectiveness assessment values. This generates a time window-level photostimulation effectiveness label, which includes the unique identifier of the time window, the effectiveness judgment result, and various intermediate assessment values ​​for statistical analysis. Simultaneously, a re-examination flag is set for time windows identified as abnormal. The re-examination flag is set according to the abnormality source classification, including three categories: process abnormality re-examination, unreliable response re-examination, and cross-modal inconsistency re-examination. The corresponding abnormality source is also labeled, including process state abnormality, unreliable physiological response, and cross-modal response inconsistency. Each type of abnormality records the specific judgment criteria value. After all time window evaluation tasks are completed, the photostimulation effectiveness labels for each time window are summarized to generate statistical results on photostimulation effectiveness.

[0055] like Figure 5The graph showing the comprehensive evaluation results of photostimulation effectiveness displays the stimulation process status assessment value, physiological response reliability assessment value, and photostimulation effectiveness assessment value on the same time axis. The green line represents the stimulation process status assessment value, the blue line represents the physiological response reliability assessment value, and the bar chart represents the photostimulation effectiveness assessment value. The orange dashed line represents the photostimulation effectiveness threshold of 0.6, the red dashed line represents the response reliability threshold of 0.45, and the yellow shaded area represents the photostimulation execution range of 20-40. As shown in the figure, before the light stimulation is executed, the overall evaluation value of the stimulation process state is at a moderate level, the physiological response reliability evaluation value is close to or slightly below the reliability threshold, and the overall light stimulation effectiveness evaluation value is low. This indicates that only background fluctuations exist before stimulation, and an effective stimulation state satisfying the dual constraints has not yet been formed. After entering the light stimulation execution period, the evaluation value of the stimulation process state rises significantly and remains above the process state judgment threshold for most time windows. The physiological response reliability evaluation value also increases synchronously, driving the light stimulation effectiveness evaluation value to reach a relatively high value in the middle of the stimulation. This indicates that not only does the stimulation link meet the working conditions within the current time window, but the EEG and hemodynamic response results also have high reliability. After the stimulation ends, as the process state evaluation value and the physiological response reliability evaluation value fall back, the light stimulation effectiveness evaluation value decreases again. This figure intuitively illustrates that the stimulation process state constraints, physiological response reliability, and time window-level joint evaluation results can be uniformly mapped onto the same graphical representation, achieving continuous quantitative judgment of the effectiveness of the entire light stimulation process.

[0056] In this implementation plan, the effectiveness determination results throughout the entire photostimulation process are transformed into indexable, statistically verifiable, and traceable tagged data. By using classification settings such as effective stimulus markers, process anomaly re-examination, unreliable response re-examination, and cross-modal inconsistency re-examination, the ability to distinguish different sources of anomalies can be improved, avoiding the problem of outputting only a single effectiveness conclusion without being able to locate the cause of the anomaly. After all time window evaluation tasks are completed, by summarizing the photostimulation effectiveness tags of each time window, a statistical result of photostimulation effectiveness covering the entire process can be formed, improving the consistency, comparability, and verifiability of the results output, and providing a unified data foundation for anomaly distribution analysis, effective stimulus window statistics, and overall effectiveness evaluation.

[0057] Specifically, the second aspect of this invention provides a photostimulation effectiveness evaluation system based on multimodal feature fusion, applied to a photostimulation effectiveness evaluation method based on multimodal feature fusion, comprising: a multimodal data acquisition and preprocessing module for acquiring raw observation data and performing time axis repositioning, dynamic denoising combining bandpass filtering and notch filtering, minimum-maximum normalization based on the statistical range of the baseline segment before stimulation, and sliding time window segmentation on the raw observation data to construct a fused sensing quality perception frame; a stimulation process state feature construction module for evaluating the stimulation process state based on the fused sensing quality perception frame by weighted fusion of dose deviation, angle offset, phase offset, and motion perturbation terms, screening effective windows of the process, and constructing a process state feature set; and response feature extraction. The quality enhancement module is used to extract EEG neural activity response features based on the stimulus process state feature set through frequency band filtering and notch filtering, extract hemodynamic response features through optical density change calculation and correction Beer-Lambert inversion, and calculate the response residual noise value by weighted fusion of abnormal sampling ratio, failure channel ratio, and interference residual ratio to perform physiological response credibility assessment and screen response credibility windows. The multimodal consistency fusion assessment module is used to construct joint assessment terms by performing joint coupling terms and joint normalization terms based on the stimulus process state feature set and physiological response feature set, and calculate the consistency fusion of the photostimulation effectiveness assessment value by combining cross-modal response delay difference and delay penalty term to perform photostimulation effectiveness assessment, identify the current photostimulation state, and generate photostimulation effectiveness labels.

[0058] This implementation plan establishes a complete structured evaluation process, from process state constraints to the reliability of response results and the final effectiveness determination. It improves the photostimulation effectiveness evaluation system's ability to identify states such as irradiation detachment, abnormal adhesion, insufficient dosage, residual shallow interference, excessive motion artifacts, and inconsistent cross-modal responses. It also enhances the continuity, accuracy, interpretability, and verifiability of the evaluation results, providing a unified data foundation for the statistical analysis of photostimulation effectiveness throughout the entire process.

[0059] 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 process, method, article, or apparatus.

[0060] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for evaluating the effectiveness of optical stimulation based on multimodal feature fusion, characterized in that, Includes the following steps: S1. Acquire raw observation data and perform time axis repositioning, dynamic denoising, normalization and sliding time window segmentation on the raw observation data to construct a fusion sensing quality perception frame. S2, based on the fusion sensing quality perception frame, evaluate the state of the stimulus process, screen the effective window of the process, and construct the process state feature set; S3, based on the stimulus process state feature set, extracts neural activity response features and hemodynamic response features, performs physiological response reliability assessment, and filters response reliability windows; S4 performs consistency fusion based on the stimulus process state feature set and physiological response feature set to evaluate the effectiveness of light stimulation, identify the current light stimulation state, and generate a light stimulation effectiveness label.

2. The method for evaluating the effectiveness of optical stimulation based on multimodal feature fusion according to claim 1, characterized in that: The specific process of acquiring raw observation data, performing time-axis repositioning, dynamic denoising, normalization, and sliding time window segmentation on the raw observation data to construct a fused sensing quality perception frame is as follows: The entire process of applying light stimulation to the forehead region using a head-mounted light stimulation device was dynamically acquired to obtain raw observation data. The raw observation data included light stimulation power, stimulation duration, scattered light fluctuation coefficient, cumulative target irradiation, scattered light centroid angle, baseline centroid angle, scattered phase offset, acceleration modulus change rate, angular velocity modulus change rate, EEG timing signal, and near-infrared raw light intensity signal. Unified time-division multiplexing control is implemented for the near-infrared spectroscopy measurement link and the optical stimulation link; during the near-infrared measurement sub-time slot, the stimulation emission unit is turned off and the raw near-infrared light intensity signal is acquired; during the optical stimulation output sub-time slot, the near-infrared emission unit is turned off and blanking or shielding processing is performed on the near-infrared detector to suppress the crosstalk and saturation effect of the stimulation light on the measurement link. The original observation data is marked with a sampling time identifier by using a unified clock reference, and time axis relocation is performed based on the timestamp information of different sampling links. For data with inconsistent sampling rates, an interpolation resampling method is used to construct a unified time series to fix the time misalignment problem caused by sampling frequency differences, buffer delays or local frame drops in the acquisition link. The original observation data is dynamically denoised by combining bandpass filtering and notch filtering. The original observation data is normalized by min-max normalization based on the statistical range of the baseline segment before stimulation, and a sliding time window is used to construct a fusion sensing quality perception frame.

3. The method for evaluating the effectiveness of optical stimulation based on multimodal feature fusion according to claim 1, characterized in that: The specific process for evaluating the state of the stimulus process based on the fused sensing quality perception frame is as follows: Based on the fusion sensing quality perception frame, the theoretical irradiation is obtained by multiplying the light stimulation power and the stimulation duration, and then multiplied by the scattered light fluctuation coefficient to obtain the current cumulative effective irradiation. The scattering angle offset is obtained by the difference between the centroid angle of the scattered light and the centroid angle of the baseline. The motion artifact risk value is obtained by normalizing the rate of change of acceleration modulus and the rate of change of angular velocity modulus and then weighting them. The absolute value of the deviation between the cumulative effective exposure dose and the target cumulative exposure dose in the current time window is taken and divided by the sum of the target cumulative exposure dose and the zero-prevention term to obtain the dose deviation term; the angle offset of the scattered light is multiplied by the angle weighting coefficient to obtain the angle offset term; the phase offset of the scattered light is multiplied by the phase weighting coefficient to obtain the phase offset term; the motion artifact risk value is multiplied by the motion artifact weighting coefficient to obtain the motion disturbance term; the dose deviation term, angle offset term, phase offset term and motion disturbance term are added together and the negative value of the exponent is taken to obtain the stimulation process state assessment value.

4. The method for evaluating the effectiveness of optical stimulation based on multimodal feature fusion according to claim 1, characterized in that: The specific process for constructing the process state feature set within the effective window of the filtering process is as follows: The stimulus process state assessment value is compared with the stimulus process state threshold in real time: when the stimulus process state assessment value is less than the stimulus process state threshold, it is determined that the current time window has an abnormal state, the time window is excluded from the validity assessment calculation, and the current time window is marked as an abnormal process window; when the stimulus process state assessment value is greater than or equal to the stimulus process state threshold, it is determined that the current time window meets the stimulus process state constraint, the current time window is marked as a valid process window, and a stimulus process state feature set is constructed.

5. The method for evaluating the effectiveness of optical stimulation based on multimodal feature fusion according to claim 1, characterized in that: The specific process of extracting neural activity response features and hemodynamic response features based on the stimulus process state feature set is as follows: Based on the stimulus process state feature set, frequency band filtering, notch filtering and artifact suppression are performed on the EEG time sequence signal. The absolute power, relative power and frequency band ratio of the target frequency band are calculated to generate neural activity response features. The optical density change is calculated, differential path correction and modified Beer-Lambert law inversion are performed on the near-infrared raw light intensity signal to obtain the changes in oxyhemoglobin, deoxyhemoglobin and total hemoglobin, and generate hemodynamic response characteristics. The influence of tissue blood flow components on brain region response signals was reduced by channel regression, baseline drift removal, and interference suppression. Hemodynamic and neural activity response features were enhanced using outlier detection, missing point interpolation, failed channel labeling, and motion artifact correction methods to generate corrected EEG and hemodynamic response feature sequences. The proportions of outlier sampling, failed channels, and interference residuals were obtained using threshold discrimination statistics, channel validity detection, and root mean square residual evaluation, respectively. Finally, the residual noise value was calculated using a weighted normalization fusion method. Based on the corrected EEG response feature sequence, the power change amplitude of the target frequency band is extracted to obtain the EEG response amplitude value; The ratio of the standard deviation to the absolute value of the mean of the corrected EEG response feature sequence within the current time window is calculated to obtain the EEG response fluctuation value; based on the corrected hemodynamic response feature sequence, the target response amplitude in the change of oxyhemoglobin is extracted to obtain the hemodynamic response amplitude value. The hemodynamic response fluctuation value is obtained by calculating the ratio of the standard deviation to the absolute value of the mean of the corrected hemodynamic response characteristic sequence within the current time window.

6. The method for evaluating the effectiveness of optical stimulation based on multimodal feature fusion according to claim 1, characterized in that: The specific process for conducting a physiological response reliability assessment and selecting a response reliability window is as follows: Multiply the EEG response amplitude value by the EEG amplitude coefficient to obtain the EEG response contribution term; multiply the hemodynamic response amplitude value by the blood amplitude coefficient to obtain the hemodynamic response contribution term; add the EEG response contribution term and the hemodynamic response contribution term to obtain the physiological intensity term; Multiplying the EEG fluctuation value by the EEG fluctuation coefficient yields the EEG fluctuation effect term. Multiplying the hemodynamic fluctuation value by the hemodynamic fluctuation coefficient yields the hemodynamic fluctuation influence term. Multiply the residual noise value of the response by the residual coefficient to obtain the residual noise effect term; Add the constant 1 to the terms affecting brain electrical fluctuations, hemodynamic fluctuations, and residual noise to obtain the comprehensive constraint term; then divide the physiological intensity term by the comprehensive constraint term to construct a reliable assessment value of the physiological response. The physiological response confidence assessment value is compared with the physiological response confidence threshold in real time: when the physiological response confidence assessment value is less than the physiological response confidence threshold, the current time window is marked as an unreliable response window; When the physiological response confidence assessment value is greater than or equal to the physiological response confidence threshold, the current time window is marked as the response confidence window, and a physiological response feature set is constructed.

7. The method for evaluating the effectiveness of optical stimulation based on multimodal feature fusion according to claim 1, characterized in that: The specific process for evaluating the effectiveness of light stimulation based on the consistency fusion of the stimulus process state feature set and the physiological response feature set is as follows: Based on the stimulus process state feature set and physiological response feature set, a unified mapping is performed on the process state assessment results and physiological response assessment results. The process state assessment results serve as a prerequisite constraint, and only the process effective window is included in the light stimulation effectiveness assessment. By performing sliding time window analysis on the EEG response feature sequence, the first moment when the target frequency band power change reaches the neural activity response threshold after N consecutive sampling points is identified, and the time difference between the moment and the light stimulation start moment is determined to obtain the neural activity response delay. The first moment when the target hemodynamic response amplitude reaches the hemodynamic response threshold after M consecutive sampling points is identified, and the time difference between this moment and the light stimulation start moment is used to determine the hemodynamic response delay. Multiplying the stimulus process state assessment value and the physiological response confidence assessment value by a constant k yields the joint coupling term; adding the stimulus process state assessment value and the physiological response confidence assessment value, plus a zero-prevention term, yields the joint normalization term; dividing the joint coupling term by the joint normalization term yields the joint assessment term; taking the absolute value of the difference between the neural activity response delay and the hemodynamic response delay yields the cross-modal response delay difference. Multiply the cross-modal response delay difference by the delay difference penalty coefficient to obtain the delay penalty term; subtract the delay penalty term from the joint evaluation term to obtain the photostimulation effectiveness evaluation value.

8. The method for evaluating the effectiveness of optical stimulation based on multimodal feature fusion according to claim 1, characterized in that: The specific process for identifying the current state of light stimulation is as follows: The light stimulation effectiveness assessment value is compared with the light stimulation effectiveness threshold: when the light stimulation effectiveness assessment value is greater than or equal to the light stimulation effectiveness threshold, it is determined that the stimulation is valid and the physiological response result is reliable within the current time window, and the current time window is identified as an effective stimulation window. If the light stimulation effectiveness assessment value is less than the light stimulation effectiveness threshold, the abnormal cause analysis stage will be entered, and threshold judgments will be made on the stimulation process state assessment value, physiological response reliability assessment value, and response delay difference, respectively.

9. The method for evaluating the effectiveness of optical stimulation based on multimodal feature fusion according to claim 1, characterized in that: The specific process for generating the photostimulation effectiveness label is as follows: The identified effective stimulus states are structurally bound to the current time window number, sampling time information, stimulus process state assessment value, physiological response reliability assessment value, and photostimulation effectiveness assessment value to generate time window-level photostimulation effectiveness labels. At the same time, a re-examination flag is set for time windows identified as abnormal, and the corresponding abnormality source is marked. After all time window assessment tasks are completed, the photostimulation effectiveness labels of each time window are summarized to generate photostimulation effectiveness statistics.

10. A photostimulation effectiveness evaluation system based on multimodal feature fusion, employing the photostimulation effectiveness evaluation method based on multimodal feature fusion as described in any one of claims 1-9, characterized in that, include: The multimodal data acquisition and preprocessing module is used to acquire raw observation data and perform time axis repositioning, dynamic denoising, normalization and sliding time window segmentation on the raw observation data to construct fusion sensing quality perception frames. The stimulus process state feature construction module is used to evaluate the stimulus process state based on the fused sensing quality perception frame, filter the effective window of the process, and construct the process state feature set. The response feature extraction and quality enhancement module is used to extract neural activity response features and hemodynamic response features based on the stimulus process state feature set, and to perform physiological response reliability assessment and screen response reliability windows. The multimodal consistency fusion assessment module is used to perform consistency fusion based on the stimulus process state feature set and the physiological response feature set to assess the effectiveness of light stimulation, identify the current light stimulation state, and generate a light stimulation effectiveness label.

Images