A method for motion artifact correction of time-domain near-infrared raw signals

Abstract

The application discloses a motion artifact correction method for time-domain near-infrared original signals. The method first performs time-gated decomposition on a sequence of diffuse time flight curves collected continuously, extracts shallow sensitive signals representing surface disturbance and deep sensitive signals representing brain tissue change, then performs multi-dimensional artifact detection on the original curve and the shallow and deep signals to generate an artifact mask sequence, performs dynamic regression correction on the deep sensitive signals with the shallow sensitive signals as reference to remove the shallow coupling components, performs time alignment on the peak shift segments and neighborhood reconstruction on the abnormal segments according to the artifact mask sequence, and finally outputs the corrected deep brain signals and their confidence. The application utilizes the time-gated information unique to the time-domain near-infrared, can effectively separate the shallow interference from the deep signals, and jointly corrects various artifacts such as sudden jumps, peak shift and signal loss, thereby improving the reliability of deep brain signal recovery.

Description

Technical Field

[0001] This invention belongs to the field of time-domain near-infrared signal correction and brain signal recovery technology, specifically involving a method for detecting, separating and correcting motion artifacts in brain signals by utilizing the superficial and deep time-gated information in a time-domain near-infrared functional imaging system (TD-fNIRS). Background Technology

[0002] In actual measurements, the TD-fNIRS probe is susceptible to motion artifacts such as signal jumps, sudden changes in count rate, peak shifts, and waveform distortion when affected by factors such as slight slippage, changes in contact pressure, hair obstruction, and subject movements. These artifacts simultaneously affect the amplitude, temporal position, and morphological structure of the DTOF curve, further interfering with subsequent time-gated analysis, rapid fitting, and parameter inversion, thus reducing the accuracy and stability of brain signal extraction.

[0003] Most existing methods for motion artifact removal draw on experience from continuous-wave near-infrared (TIR) ​​techniques, such as bandpass filtering, spline interpolation, wavelet artifact removal, and accelerometer-assisted regression. These methods can suppress slow drift or large-amplitude transient artifacts to some extent, but because they do not utilize the unique shallow and deep time-gated information of TD-fNIRS signals, they often struggle to effectively distinguish between genuine changes in cerebral blood flow and surface disturbances caused by changes in probe contact. Summary of the Invention

[0004] To address the problems in existing TD-fNIRS signal processing, such as incomplete motion artifact identification, difficulty in effectively separating superficial perturbations from deep brain signals, difficulty in jointly processing various artifacts such as peak shift and local shading, and insufficient reliability of correction results, this invention provides a time-domain near-infrared motion artifact correction method. This method utilizes the unique time gating difference between superficial and deep layers in TD signals to separate superficial interference components from deep brain signals, and jointly detects and corrects various types of artifacts such as jumps, misalignments, and shading, thereby improving the reliability of deep brain signal recovery results.

[0005] According to one aspect of the present invention, a method for motion artifact correction of raw near-infrared signals in the time domain is provided, comprising: acquiring a sequence of raw diffuse time-of-flight curves continuously acquired by a time-domain near-infrared functional imaging system; performing time-gated decomposition on the raw diffuse time-of-flight curve sequence to extract shallow sensitive signals and deep sensitive signals, wherein the shallow sensitive signals characterize surface disturbances caused by changes in scalp and probe contact, and the deep sensitive signals characterize changes related to brain tissue; performing multidimensional artifact detection on the raw diffuse time-of-flight curve sequence, the shallow sensitive signals, and the deep sensitive signals respectively, identifying time segments contaminated by motion artifacts in each signal, and merging them to generate an artifact mask sequence; and then... Using the layer-sensitive signal as an interference reference, dynamic regression correction is performed on the deep layer-sensitive signal to obtain a regression-corrected deep layer signal. Based on the artifact mask sequence, peak shift correction is performed on the regression-corrected deep layer signal, including temporal realignment of the disturbed time segments to obtain a time-aligned deep layer signal. Based on the artifact mask sequence, data continuity restoration processing is performed on the time-aligned deep layer signal, including neighborhood reconstruction of the abnormal time segments to obtain a corrected deep brain signal. The correction confidence of the corrected deep brain signal is determined, and the corrected deep brain signal and correction confidence are output.

[0006] As a further technical solution, the original diffuse time-of-flight curve sequence is subjected to time-gated decomposition, including: setting a shallow time window and a deep time window, wherein the shallow time window corresponds to the flight time of photons arriving earlier and the deep time window corresponds to the flight time of photons arriving later; and summing the original diffuse time-of-flight curve sequence within the shallow time window and the deep time window respectively to obtain the shallow sensitive signal and the deep sensitive signal.

[0007] As a further technical solution, the multidimensional artifact detection includes: jump detection, calculating the time difference of the deep sensitive signal, and determining that a jump artifact exists in the corresponding time segment when the time difference exceeds a first threshold set based on the absolute deviation of the median; count rate anomaly detection, calculating the total photon count at each moment in the original diffuse time-of-flight curve sequence, and determining that a count rate anomaly artifact exists in the corresponding time segment when the total photon count exceeds a second threshold range set based on the mean and standard deviation; and peak position drift detection, calculating the first moment of the diffuse time-of-flight curve at each moment in the original diffuse time-of-flight curve sequence, and determining that a peak position drift artifact exists in the corresponding time segment when the deviation of the first moment relative to the baseline value exceeds a third threshold.

[0008] As a further technical solution, the process of merging and generating the artifact mask sequence includes: performing a logical union operation on the detection results of the jump detection, count rate anomaly detection, and peak position drift detection to obtain the artifact mask sequence, which is used to calibrate whether the artifacts at each time point are contaminated by motion artifacts.

[0009] As a further technical solution, using the shallow sensitive signal as an interference reference term, the dynamic regression correction of the deep sensitive signal includes: estimating the coupling coefficient between the shallow and deep sensitive signals using a sliding window method within the time segment where the artifact mask sequence is labeled as clean or slightly contaminated; applying the coupling coefficient to the time segment contaminated by artifacts to dynamically weight the deep sensitive signal to suppress the interference of the shallow sensitive signal, thereby obtaining the regression-corrected deep signal.

[0010] As a further technical solution, estimating the coupling coefficient between the shallow sensitive signal and the deep sensitive signal using a sliding window method includes: within the current sliding window, calculating the ratio of the covariance of the deep sensitive signal and the shallow sensitive signal to the variance of the shallow sensitive signal, as the coupling coefficient, which is dynamically updated as the sliding window moves.

[0011] As a further technical solution, performing peak shift correction on the regression-corrected deep signal based on the artifact mask sequence includes: determining a time segment labeled as peak shift based on the artifact mask sequence; selecting diffuse time-of-flight curves within clean segments before and after the time segment as reference curves; determining the optimal time offset between the diffuse time-of-flight curves of each frame within the time segment and the reference curve through cross-correlation or dynamic time warping; performing time-domain translation on the diffuse time-of-flight curves of each frame within the time segment based on the optimal time offset, so that the time structure of the time segment is restored to the position of the reference curve; and recalculating the deep sensitive signal based on the time-domain translated diffuse time-of-flight curves to obtain the time-aligned deep signal.

[0012] As a further technical solution, the data continuity restoration processing of the time-aligned deep signal according to the artifact mask sequence includes: determining the abnormal time segments marked as continuous as the interval to be repaired according to the artifact mask sequence; extracting reliable data points before and after the interval to be repaired as interpolation nodes; and reconstructing the time-aligned deep signal in the interval to be repaired using spline interpolation or reliable region splicing methods to obtain the continuous and complete corrected deep brain signal.

[0013] As a further technical solution, determining the correction confidence of the corrected deep brain signal includes: calculating the count signal-to-noise ratio of the corrected deep brain signal; obtaining the goodness of fit of the dynamic regression correction; evaluating the waveform stability of the corrected deep brain signal; and combining the artifact mask sequence, count signal-to-noise ratio, goodness of fit, and waveform stability to calculate the confidence value at each time point or time segment as the correction confidence.

[0014] According to one aspect of the present invention, a motion artifact correction device for time-domain near-infrared raw signals is provided, comprising: one or more processors; a storage device for storing one or more programs; and when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the motion artifact correction method for time-domain near-infrared raw signals.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: First, this invention utilizes the unique time-gated separation mechanism of TD-fNIRS between superficial and deep layers, which can more effectively distinguish deep brain signals from surface motion artifacts.

[0016] Second, this invention combines jump detection, dynamic regression correction, time alignment and missing reconstruction, which can simultaneously handle different types of motion artifacts such as amplitude anomalies, temporal misalignment and local shading.

[0017] Third, the present invention outputs a correction confidence level, which can provide a basis for subsequent inversion, result weighting, and data screening, thereby improving the interpretability and reliability of the correction results.

[0018] Fourth, this invention does not require additional auxiliary hardware and can be directly used for motion artifact correction processing in existing TD-fNIRS systems. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a method for motion artifact correction of raw time-domain near-infrared signals provided in an embodiment of the present invention. Detailed Implementation

[0021] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the single-channel time-domain near-infrared layered inversion method of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, 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. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0023] like Figure 1 As shown, this invention addresses the motion artifact problem caused by factors such as probe slippage, contact pressure changes, hair occlusion, and subject movements during continuous acquisition in time-domain near-infrared functional imaging systems (TD-fNIRS), and proposes a motion artifact correction method for raw time-domain near-infrared signals.

[0024] This invention utilizes the unique shallow and deep time-gated information of the TD-fNIRS system. By performing time-gated decomposition on continuously acquired diffuse time-of-flight (DTOF) sequences, it extracts shallow sensitive signals that primarily characterize surface perturbations and deep sensitive signals that primarily characterize changes related to brain tissue. Based on this, multidimensional artifact detection (including one or more of sudden jump detection, count rate anomaly detection, and peak shift detection) is performed on the original DTOF curves and the aforementioned shallow and deep sensitive signals to generate an artifact mask sequence for precise location of contaminated areas. The process involves several steps: First, a time segment is defined. Then, using the shallow sensitive signal as an interference reference, dynamic regression correction or weighted correction is performed on the deep sensitive signal to remove shallow coupling components. Next, time alignment processing is performed on the peak-shifting artifact segments, and neighborhood reconstruction or reliable region splicing repair is performed on abnormal segments caused by short-term shading, probe detachment, or local count loss. Finally, the confidence level of the corrected deep brain signal is evaluated, and the corrected deep brain signal and its corresponding correction confidence level are output, providing a weighting basis for subsequent inversion modules, quality control modules, or data screening modules.

[0025] Therefore, this invention can effectively separate superficial interference components from deep brain signals and perform joint detection and correction of various types of artifacts such as sudden jumps, misalignments, and light shading, thereby effectively improving the reliability of deep brain signal recovery results.

[0026] Step 1, Time-gated decomposition

[0027] like Figure 1 As shown, the continuously acquired DTOF sequences were first decomposed using time gating to extract superficial and deep sensitive signals. The superficial sensitive signals primarily characterize surface disturbances caused by changes in scalp and probe contact, while the deep sensitive signals primarily characterize changes related to brain tissue.

[0028] Specifically, the raw DTOF sequence continuously acquired by the system can be represented as a three-dimensional data form H(t,τ), where t is the macroscopic acquisition time and τ is the photon flight time. Based on the difference in photon flight time, this invention decomposes the raw DTOF sequence into shallow-layer sensitive signals and deep-layer sensitive signals.

[0029] Setting a shallow time window and deep time window Among them, shallow time window For photons arriving early (i.e., those with short flight times), these photons primarily carry information about superficial tissues such as the scalp and skull; deeper time windows... For photons arriving late (i.e., those with long flight times), these photons primarily carry information from deep brain tissue. The extraction methods for superficial and deep sensitive signals are as follows:

[0030]

[0031]

[0032] in, This will serve as an interference reference signal for subsequent steps. The target signal to be processed.

[0033] Step 2, Multidimensional Artifact Detection

[0034] Multidimensional artifact detection is performed on the original DTOF curve, as well as the shallow and deep sensitive signals, to identify time segments that may be contaminated by motion artifacts. The multidimensional artifact detection includes at least one or more of the following: jump detection, count rate anomaly detection, and peak shift detection.

[0035] To ensure the accuracy of subsequent corrections, contaminated data segments need to be identified. This invention employs a multi-dimensional artifact detection strategy, detecting the original DTOF sequence, shallow sensitive signals, and deep sensitive signals separately, and merging the detection results to generate an artifact mask sequence M(t). If M(t)=1, it indicates the presence of artifacts at that moment; if M(t)=0, it indicates a clean signal at that moment.

[0036] (a) Sudden jump detection

[0037] This can be based on the characteristics of the time derivative distribution, utilizing the idea of ​​Time Derivative Distribution Repair (TDDR) to calculate the signal difference. Sudden jump artifacts manifest as drastic amplitude jumps in the signal within a short period. This invention calculates the time difference of deeply sensitive signals as follows:

[0038]

[0039] If the difference value exceeds the threshold set based on the median absolute deviation (MAD), a sudden jump artifact is determined to exist at that moment.

[0040]

[0041] in, This represents the absolute deviation of the median. This is the threshold coefficient.

[0042] (ii) Detection of abnormal count rates

[0043] This invention employs a statistical distribution-based count rate anomaly detection method to detect sudden drops or increases in the total photon count. Sudden drops or increases in the total photon count occur when there is relative displacement between the probe and the scalp or when there is localized shading. The invention first calculates the total photon count at each time point.

[0044]

[0045] Subsequently, a normal range was set based on the mean and standard deviation. When the total photon count exceeds this range, it is determined to be an abnormal count rate.

[0046]

[0047] (III) Peak position drift detection

[0048] This can be based on changes in the DTOF distribution moment or peak position. When the optical path undergoes a physical change due to probe displacement, the peak position of the DTOF distribution will shift. This invention detects peak position drift based on changes in the first-order moment (centroid) of the DTOF distribution:

[0049]

[0050] When the deviation of the first moment from the baseline value at a certain moment exceeds a set threshold, it is determined that a peak drift artifact exists at that moment.

[0051]

[0052] Perform a logical union operation on the above detection results to generate the final artifact mask sequence:

[0053]

[0054] Step 3, Dynamic Regression Correction Based on Shallow Reference

[0055] Using the shallow sensitive signal as an interference reference, dynamic regression correction or weighted correction is performed on the deep sensitive signal to remove shallow coupling components caused by probe contact changes, surface tissue disturbance, or common-mode interference. Since the shallow and deep sensitive signals have a high correlation in motion artifacts, this invention uses the shallow sensitive signal as an interference reference and performs dynamic regression correction on the deep sensitive signal.

[0056] Utilizing shallow signals With deep signals High correlation with motion artifacts was observed, and noise was removed through regression. A dynamically weighted regression model was used, assuming that deep signals contain genuine brain signals. This consists of noise from shallow coupling. Linear regression within a sliding window is used to estimate the coupling coefficient. The formula is as follows:

[0057]

[0058] Wherein, coupling coefficient It is obtained by minimizing the error within the window, as shown in the following formula:

[0059]

[0060] Preferably, the dynamic regression correction estimates the coupling relationship between shallow and deep sensitive signals only in clean or lightly contaminated segments, and uses this coupling relationship for noise removal in artifact-contaminated segments. In other words, this regression correction step only estimates the coupling coefficient in clean or lightly contaminated segments where M(t)=0 to avoid estimation distortion in severely artifact-contaminated segments. This coupling coefficient is then applied to the artifact-contaminated time segments for correction, resulting in the regression-corrected deep signal. Severely artifact-contaminated segments are left for subsequent repair.

[0061] Step 4, Time Alignment of Peak Shift

[0062] Time alignment processing is performed on pseudo-film segments where peak position drift is detected, restoring the rising edge, peak position, or local time structure of the disturbed segment to the reference position, thereby reducing timing misalignment caused by optical path changes or probe displacement. For pseudo-film segments where peak position drift has been detected (i.e., The pseudo-film segments need to be aligned and corrected on the timeline. Specifically, the diffuse time-flight curve within the clean time segment before and after the peak-shifted segment is selected as the reference curve, and the optimal time offset between the current frame and the reference frame is found through cross-correlation or dynamic time warping methods. In this embodiment, the cross-correlation alignment method is used to find the optimal time delay. This maximizes the cross-correlation value between the current frame and the reference frame.

[0063]

[0064] in, Indicates mutual correlation.

[0065] Based on the obtained optimal time offset, the diffuse time-of-flight curves of each frame within the peak-shifted segment are shifted in the time domain to restore the rising edge, peak position, or local time structure of the perturbed segment to the reference position. The corrected DTOF distribution is as follows:

[0066]

[0067] Subsequently, based on the diffuse time flight curve after time-domain translation... The deep-sensitive signal is recalculated to obtain a time-aligned deep signal. .

[0068] Step 5: Neighborhood reconstruction and repair of abnormal fragments

[0069] For abnormal segments caused by short-term light shading, probe detachment, or local count loss, neighborhood reconstruction, trusted region splicing, or other timing recovery methods are used to repair them in order to restore signal continuity.

[0070] Specifically, for severe detachment, that is To address the data gaps caused by artifacts, interpolation is performed using temporal continuity. The continuously anomalous time segments identified in the artifact mask sequence are determined as the intervals to be repaired. The reliable data points before and after the interval are extracted as interpolation nodes. Spline interpolation is then used to reconstruct the time-aligned deep signal within the interval to be repaired, resulting in a continuous and complete signal. In this embodiment, cubic spline interpolation is used for reconstruction.

[0071]

[0072] in These are the nearest points outside the artifact interval. Interpolation nodes are reliable data points on both sides of the interval to be repaired, that is, data points located outside the interval to be repaired that are not marked as abnormal by the artifact mask sequence, or data points whose signal quality meets the preset threshold after correction.

[0073] Through the aforementioned neighborhood reconstruction, the temporal continuity of the signal can be effectively restored, yielding the corrected deep brain signal. Those skilled in the art should understand that, in addition to spline interpolation, other temporal recovery methods such as trusted region splicing are also applicable to this invention.

[0074] Step 6, Confidence Assessment and Output

[0075] The confidence level of the calibrated deep brain signal is assessed, and the calibrated deep brain signal and its corresponding calibration confidence level are output. The calibration confidence level is used to characterize the credibility of the current calibration result and is provided as a weighting basis for subsequent inversion modules, quality control modules, or data filtering modules.

[0076] Preferably, the correction confidence level can comprehensively consider indicators such as the counting signal-to-noise ratio of the current signal, the goodness of fit of the regression model, and the stability of the corrected waveform, so as to improve the reliability of subsequent use. Specifically, the correction confidence level... It is obtained by weighted summation of the following three parts:

[0077]

[0078] Where M(t) is the artifact mask sequence, and (1-M(t)) reflects whether an artifact is detected at that moment; This represents the photon count signal-to-noise ratio at the current moment. This is the maximum signal-to-noise ratio reference value; the ratio of the two reflects the signal quality. The goodness of fit of the regression model in step 3 reflects the degree to which the shallow reference signal interprets noise; , , These are the weighting coefficients for each item.

[0079] Ultimately, this invention simultaneously outputs the corrected deep brain signal and its corresponding correction confidence level, providing a basis for subsequent signal inversion, result weighting, and data screening, thereby improving the interpretability of the correction results and the reliability of subsequent use.

[0080] The present invention also provides a motion artifact correction device for time-domain near-infrared raw signals, comprising one or more processors and a storage device. The storage device stores one or more programs, which, when executed by the one or more processors, cause the processors to implement the motion artifact correction method for time-domain near-infrared raw signals described in the above method embodiments.

[0081] In one specific embodiment, the device further includes a communication interface for data interaction with a time-domain near-infrared functional imaging system, receiving continuously acquired raw diffuse time-of-flight curve sequences from the system, and outputting corrected deep brain signals and their corresponding corrected confidence levels to external modules. The communication interface can be a wired interface or a wireless interface.

[0082] In one specific embodiment, the device further includes a display module for displaying the corrected deep brain signal and its corresponding correction confidence level in real time.

[0083] For example, when performing the correction method, the processor acquires the sequence of raw diffuse time-of-flight curves continuously acquired by the time-domain near-infrared functional imaging system through the communication interface.

[0084] The processor performs time-gated decomposition on the original diffuse time-of-flight curve sequence to extract superficial and deep sensitive signals. The superficial sensitive signals characterize surface disturbances caused by changes in scalp and probe contact, while the deep sensitive signals characterize related changes in brain tissue.

[0085] The processor performs multidimensional artifact detection on the original diffuse time-of-flight curve sequence, the shallow sensitive signal, and the deep sensitive signal, respectively, identifies time segments contaminated by motion artifacts in each signal, and merges them to generate an artifact mask sequence.

[0086] The processor uses the shallow sensitive signal as an interference reference and performs dynamic regression correction on the deep sensitive signal to obtain the regression-corrected deep signal.

[0087] The processor performs peak shift correction on the regression-corrected deep signal according to the artifact mask sequence. The peak shift correction includes temporal realignment of the disturbed time segment to obtain a time-aligned deep signal.

[0088] The processor performs data continuity restoration processing on the time-aligned deep signal according to the artifact mask sequence. The data continuity restoration processing includes neighborhood reconstruction of abnormal time segments to obtain corrected deep brain signals.

[0089] The processor determines the correction confidence level of the corrected deep brain signal and outputs the corrected deep brain signal and the correction confidence level through the communication interface.

[0090] Those skilled in the art will understand that the steps in the above method embodiments can be implemented by computer program instructions related hardware (such as processors, controllers, etc.). The processor can be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or any combination thereof.

[0091] The storage device can be any medium capable of storing program code, including but not limited to: USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk. The processor executes the steps in the above method embodiments by running the program stored in the storage device.

[0092] Preferably, when the processor performs the dynamic regression correction, it estimates the coupling coefficient between the shallow sensitive signal and the deep sensitive signal only within the time segment where the artifact mask sequence is labeled as clean or slightly contaminated, and applies the coupling coefficient to the time segment contaminated by the artifact for correction.

[0093] Preferably, when the processor performs the peak shift correction, it determines the optimal time offset between the diffuse time flight curve of each frame and the reference curve by cross-correlation or dynamic time warping, and performs time-domain translation based on the optimal time offset.

[0094] Preferably, when the processor performs the data continuity recovery process, it uses spline interpolation or trusted region splicing to reconstruct the abnormal segments.

[0095] Preferably, when the processor determines the correction confidence level, it calculates the confidence level value at each time point by comprehensively considering the current signal-to-noise ratio, the goodness of fit of the regression model, and the stability index of the corrected waveform.

[0096] In a specific application scenario, the device is integrated into the data acquisition workstation of a time-domain near-infrared functional imaging system to perform streaming processing on the real-time acquired DTOF sequences and continuously output the corrected deep brain signals and corresponding confidence values ​​within a preset time window.

[0097] In another specific application scenario, the device serves as an offline data processing platform, performing batch correction processing on stored historical DTOF data and outputting the corrected complete time series and time-by-time confidence curves.

[0098] For example, when the calibration confidence level is lower than a preset threshold, the device sends a prompt message through the display module or communication interface to prompt the operator to check the probe contact status or eliminate environmental interference.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for motion artifact correction of raw near-infrared signals in the time domain, characterized in that, include: Acquire the raw diffuse time-of-flight curve sequence continuously acquired by the time-domain near-infrared functional imaging system; The original diffuse time-of-flight curve sequence is subjected to time-gated decomposition to extract superficial and deep sensitive signals. The superficial sensitive signals characterize surface disturbances caused by changes in scalp and probe contact, while the deep sensitive signals characterize brain tissue-related changes. Multidimensional artifact detection is performed on the original diffuse time-flight curve sequence, shallow sensitive signal and deep sensitive signal respectively to identify time segments contaminated by motion artifacts in each signal and merge them to generate an artifact mask sequence. Using the shallow sensitive signal as an interference reference, dynamic regression correction is performed on the deep sensitive signal to obtain the regression-corrected deep signal. Based on the artifact mask sequence, peak shift correction is performed on the regression-corrected deep signal. The peak shift correction includes temporal realignment of the disturbed time segment to obtain the time-aligned deep signal. Based on the artifact mask sequence, data continuity restoration processing is performed on the time-aligned deep signal. The data continuity restoration processing includes neighborhood reconstruction of abnormal time segments to obtain the corrected deep brain signal. Determine the corrected confidence level of the corrected deep brain signal, and output the corrected deep brain signal and the corrected confidence level.

2. The method for motion artifact correction of raw near-infrared signals in the time domain according to claim 1, characterized in that, The original diffuse time-flight curve sequence is subjected to time-gated decomposition, including: A shallow time window and a deep time window are set, wherein the shallow time window corresponds to the flight time of photons arriving earlier, and the deep time window corresponds to the flight time of photons arriving later; The original diffuse time-flight curve sequence is summed within shallow and deep time windows respectively to obtain shallow and deep sensitive signals.

3. The method for motion artifact correction of raw near-infrared signals in the time domain according to claim 1, characterized in that, The multidimensional artifact detection includes: Sudden jump detection involves calculating the time difference of the deep sensitive signal. When the time difference exceeds a first threshold set based on the absolute deviation of the median, it is determined that a sudden jump artifact exists in the corresponding time segment. Count rate anomaly detection involves calculating the total photon count at each moment in the original diffuse time-of-flight curve sequence. When the total photon count exceeds a second threshold range set based on the mean and standard deviation, it is determined that there is a count rate anomaly artifact in the corresponding time segment. Peak position drift detection involves calculating the first moment of the diffuse time-flight curve at each moment in the original diffuse time-flight curve sequence. When the deviation of the first moment from the baseline value exceeds a third threshold, it is determined that there is a peak position drift artifact in the corresponding time segment.

4. The method for motion artifact correction of raw near-infrared signals in the time domain according to claim 3, characterized in that, The merged artifact mask sequence includes: The detection results of the jump detection, count rate anomaly detection, and peak position drift detection are subjected to a logical union operation to obtain the artifact mask sequence, which is used to calibrate whether the device is contaminated by motion artifacts at each time point.

5. The method for motion artifact correction of raw near-infrared signals in the time domain according to claim 1, characterized in that, Using the shallow-layer sensitive signal as an interference reference, dynamic regression correction is performed on the deep-layer sensitive signal, including: Within the time segment where the artifact mask sequence is labeled as clean or slightly contaminated, the coupling coefficient between the shallow and deep sensitive signals is estimated using a sliding window method. The coupling coefficient is applied to the time segment contaminated by artifacts to dynamically weight the deep sensitive signal to suppress the interference of the shallow sensitive signal, thereby obtaining the regression-corrected deep signal.

6. The method for motion artifact correction of raw near-infrared signals in the time domain according to claim 5, characterized in that, Estimating the coupling coefficient between the shallow sensitive signal and the deep sensitive signal using a sliding window method includes: Within the current sliding window, the ratio of the covariance of the deep sensitive signal and the shallow sensitive signal to the variance of the shallow sensitive signal is calculated and used as the coupling coefficient. The coupling coefficient is dynamically updated as the sliding window moves.

7. The method for motion artifact correction of raw near-infrared signals in the time domain according to claim 1, characterized in that, Based on the artifact mask sequence, peak shift correction is performed on the regression-corrected deep signal, including: Based on the artifact mask sequence, determine the time segment labeled as peak position drift; The diffuse time-of-flight curves within the clean segments before and after the time segment are selected as reference curves; For each frame of the diffuse time-flight curve within the time segment, the optimal time offset between it and the reference curve is determined by cross-correlation or dynamic time warping. Based on the optimal time offset, the diffuse time-flight curves of each frame within the time segment are shifted in the time domain so that the time structure of the time segment is restored to the position of the reference curve. The deep sensitive signal is recalculated based on the diffuse time-flight curve after time-domain shift, and the time-aligned deep signal is obtained.

8. The method for motion artifact correction of raw near-infrared signals in the time domain according to claim 1, characterized in that, Based on the artifact mask sequence, performing data continuity restoration processing on the time-aligned deep signal includes: Based on the artifact mask sequence, the abnormal time segments that are marked as continuous are identified as the intervals to be repaired; Extract reliable data points before and after the interval to be repaired as interpolation nodes; The time-aligned deep brain signal within the interval to be repaired is reconstructed using spline interpolation or credible region splicing methods to obtain a continuous and complete corrected deep brain signal.

9. A method for motion artifact correction of raw near-infrared signals in the time domain according to claim 1, characterized in that, Determining the corrected confidence level of the corrected deep brain signal includes: Calculate the signal-to-noise ratio of the corrected deep brain signals; Obtain the goodness of fit of the dynamic regression correction; Evaluate the waveform stability of the corrected deep brain signal; By combining the artifact mask sequence, counting signal-to-noise ratio, goodness of fit, and waveform stability, the confidence value at each time point or time segment is calculated as the correction confidence value.

10. A device for correcting motion artifacts in time-domain near-infrared raw signals, characterized in that, include: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a motion artifact correction method for time-domain near-infrared raw signals as described in any one of claims 1 to 9.

Images