A Method and System for Detecting Biphasic Events in Cerebrospinal Fluid Based on Functional Magnetic Resonance Imaging

Abstract

This invention discloses a method and system for detecting biphasic events in cerebrospinal fluid (CSF) based on functional magnetic resonance imaging (fMRI), belonging to the field of medical image processing technology. The detection method specifically includes: acquiring and preprocessing FMRI data of the brain to extract time series of blood oxygenation level-dependent signals and time series of CSF signals from the whole brain gray matter; aligning the two types of signals to the same physical time axis based on inter-slice acquisition offset information; detecting local signal peaks in the blood oxygenation level-dependent signals, and using the signal characteristic moments within a preset time window before each local signal peak as candidate event zero points; locating the valley values ​​within a first preset time window after the candidate event zero points and the rebound peaks within a second preset time window later than the first preset time window in the CSF signals; and screening CSF biphasic events based on the located valley values ​​and rebound peaks. This invention detects CSF biphasic events at the unit of a single neural event to distinguish non-event-related changes and improve the accuracy of event detection.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a method and system for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging. Background Technology

[0002] The glymphatic system is an important brain clearance pathway discovered in recent years, responsible for clearing metabolic waste from the central nervous system. Global oxygenation level-dependent signal coupling with cerebrospinal fluid (gBOLD-CSF coupling) based on resting-state functional magnetic resonance imaging (fMRI) can reflect the driving effect of neurovascular activity on cerebrospinal fluid flow, and is currently an important non-invasive method for quantifying the dynamic function of the glymphatic system in vivo.

[0003] During neural activity, the expansion of cerebral blood volume causes compensatory changes in intracranial volume balance, resulting in an initial decrease in cerebrospinal fluid signal. When the neural activity ends and blood volume recovers, a compensatory rebound in cerebrospinal fluid signal occurs. In other words, the cerebrospinal fluid response to a single neural event will exhibit a biphasic waveform of initial decrease followed by rebound.

[0004] Existing gBOLD-CSF coupling analysis typically employs whole-segment correlation methods, calculating the Pearson correlation coefficient between the gBOLD signal and the cerebrospinal fluid signal over an entire time series. A common characteristic of these methods is that they combine all time frames, including both event and non-event periods, into a single statistical analysis. The output correlation coefficient is essentially a statistical average of the entire time series. Because this statistical average mixes event and non-event periods, when an individual's response time deviates from the group average due to differences in cerebrospinal fluid circulation velocity, the correlation coefficient under a fixed delay will decrease due to the time misalignment. In this case, it's impossible to distinguish whether the low value is caused by individual bias in the response time or by the absence of effective biphasic coupling in the data itself. In other words, existing gBOLD-CSF coupling analysis methods struggle to effectively distinguish between genuine biphasic pulsations and non-event-related changes such as single-phase signal drift and random noise fluctuations. The output results fail to accurately reflect the coupling state, resulting in low accuracy in quantification. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing gBOLD-CSF coupling analysis methods, which struggle to effectively distinguish between real biphasic pulsating events and non-event-related changes such as single-phase signal drift and random noise fluctuations. This invention provides a method and system for detecting biphasic events in cerebrospinal fluid (CSF) based on functional magnetic resonance imaging (fMRI). By using inter-slice acquisition offset information to time-align the blood oxygen level-dependent signal with the CSF signal, the interference of timing differences between different slices on the determination of event timing relationships is reduced. Furthermore, the zero point of candidate events is anchored in the aligned blood oxygen level-dependent signal using local pre-peak characteristic moments, achieving adaptive timing localization. This avoids timing misalignment caused by individual differences in CSF circulation velocity under the fixed delay assumption. Subsequently, the troughs and rebound peaks after the zero point of candidate events are sequentially located in the CSF signal. Combined with the pre-event baseline, biphasic morphology is jointly screened, effectively distinguishing real biphasic pulsating events from non-event-related changes such as single-phase signal drift and random noise fluctuations, improving the accuracy of CSF biphasic event detection, and thus ensuring the accuracy of subsequent coupling quantification results.

[0006] The objective of this invention is achieved through the following technical solution: Methods for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging include: We acquired and preprocessed functional magnetic resonance imaging (fMRI) data of the brain, and extracted the blood oxygenation level-dependent signal time series and cerebrospinal fluid signal time series of the whole brain gray matter from the preprocessed fMRI data. Based on the inter-slice acquisition offset information of functional magnetic resonance imaging data, the blood oxygen level dependent signal time series and the cerebrospinal fluid signal time series are aligned to the same physical time axis; Local signal peaks in the aligned blood oxygen level dependent signal time series are detected, and the signal characteristic moments within a preset time window before each local signal peak are taken as the corresponding candidate event zero points. In the aligned cerebrospinal fluid signal time series, the corresponding valley value within the first preset time window and the corresponding rebound peak within the second preset time window are located after the zero point of each candidate event, wherein the second preset time window is later than the first preset time window; Based on the corresponding trough and rebound peak of the location, biphasic events of cerebrospinal fluid were screened out.

[0007] Furthermore, the extraction of blood oxygenation level-dependent signal time series and cerebrospinal fluid signal time series from the preprocessed functional magnetic resonance imaging data includes: The preset whole-brain gray matter template is registered to the image coordinate system of the functional magnetic resonance imaging data, and the registered whole-brain gray matter template is superimposed on the preprocessed functional magnetic resonance images of each time frame. The lowest intracranial cerebrospinal fluid level is located on the average image of the preprocessed functional magnetic resonance imaging data. Within the lowest intracranial cerebrospinal fluid level, the region with an intracranial cerebrospinal fluid signal intensity higher than a preset threshold is selected as the region of interest. The region of interest is then superimposed onto the functional magnetic resonance images of each time frame after preprocessing. The average value of all voxel signals in the functional magnetic resonance images of each time frame after preprocessing is calculated within the coverage area of ​​the whole brain gray matter template and within the coverage area of ​​the region of interest. The corresponding average value calculated within the coverage area of ​​the whole brain gray matter template is used as the whole brain gray matter blood oxygenation level dependent signal value of each time frame, and the corresponding average value calculated within the coverage area of ​​the region of interest is used as the cerebrospinal fluid signal value of each time frame. The blood oxygenation level-dependent signal values ​​of whole-brain gray matter and the cerebrospinal fluid signal values ​​of each time frame were sorted in chronological order to obtain the time series of blood oxygenation level-dependent signal and cerebrospinal fluid signal of whole-brain gray matter.

[0008] Furthermore, aligning the blood oxygenation level-dependent signal time series and the cerebrospinal fluid signal time series to the same physical time axis based on the inter-slice acquisition offset information of functional magnetic resonance imaging data includes: Based on the inter-slice acquisition offset information of functional magnetic resonance imaging data, the inter-slice acquisition offset of the acquisition layers corresponding to blood oxygen level dependent signals and cerebrospinal fluid signals is determined respectively. The acquisition time of each time frame of functional magnetic resonance imaging is superimposed with the inter-layer acquisition offset of the acquisition layer corresponding to the blood oxygen level dependent signal to determine the actual physical sampling time of the blood oxygen level dependent signal in each time frame. The acquisition time of each time frame of functional magnetic resonance imaging is superimposed with the inter-slice acquisition offset of the corresponding acquisition layer of cerebrospinal fluid signal to determine the actual physical sampling time of cerebrospinal fluid signal in each time frame. According to the preset time grid, the blood oxygen level dependent signal time series and cerebrospinal fluid signal time series are interpolated according to the corresponding actual physical sampling time to align to the same physical time axis.

[0009] Furthermore, the detection of local signal peaks in the aligned blood oxygen level-dependent signal time series, with the signal characteristic moments within a preset time window preceding each local signal peak as the corresponding candidate event zero points, includes: In the aligned blood oxygen level dependent signal time series, local maxima with signal amplitudes greater than a preset amplitude threshold are selected as candidate peaks, and candidate peaks with time intervals between adjacent candidate peaks greater than a preset interval threshold are identified as local signal peaks. Within a preset four-window period before each local signal peak, the positive slope value of the blood oxygen level dependent signal is calculated point by point in time. The time point with the largest positive slope value is taken as the signal characteristic moment, and the signal characteristic moment is taken as the zero point of the corresponding candidate event.

[0010] Furthermore, the step of locating the corresponding valley value within the first preset time window and the corresponding rebound peak within the second preset time window after locating the zero point of each candidate event in the aligned cerebrospinal fluid signal time series includes: Calculate the median of the cerebrospinal fluid signal amplitude at all time points within the preset time window before the zero point of the corresponding candidate event, and determine the pre-event baseline of each candidate event based on the calculated median. Correct the corresponding cerebrospinal fluid signal based on the baseline before the corresponding event; Starting from the zero point of the corresponding candidate event, the amplitude of the corrected cerebrospinal fluid signal is compared point by point within the first preset time window after the zero point of the candidate event, and the time point with the smallest cerebrospinal fluid signal amplitude is taken as the valley value of the corresponding candidate event. Within a second preset time window after the zero point of the candidate event, the amplitude of the corrected cerebrospinal fluid signal is compared point by point, and the time point with the largest cerebrospinal fluid signal amplitude is taken as the rebound peak of the corresponding candidate event.

[0011] Furthermore, the cerebrospinal fluid biphasic events are screened based on the corresponding troughs and rebound peaks located in the location, including: Compare the times of the corresponding trough and rebound peak, and determine whether the amplitudes of the corresponding trough and rebound peak are positive respectively; If the amplitude of the corresponding trough is not positive, the amplitude of the corresponding rebound peak is positive, and the time of the corresponding rebound peak is later than the time of the corresponding trough, the candidate event is judged to meet the event qualification conditions. Candidate events that meet the event eligibility criteria are classified as biphasic events in cerebrospinal fluid.

[0012] Furthermore, the acquisition and preprocessing of functional magnetic resonance imaging (fMRI) data of the brain includes: Acquire functional magnetic resonance imaging (fMRI) data of the brain, remove the initial frame of the fMRI data, and perform rigid body motion correction on the fMRI data after removing the initial frame. Trend regression and motion parameter regression were performed on the functional magnetic resonance imaging (fMRI) data after rigid body motion correction to obtain preprocessed fMRI data.

[0013] Furthermore, after screening for biphasic cerebrospinal fluid events, the following steps were also performed: Based on the corresponding trough, rebound peak, pre-event baseline, and blood oxygen level dependent signal of the biphasic cerebrospinal fluid event, corresponding quantitative indicators are calculated. The quantitative indicators include the falling area, rising area, total exchange area, peak-trough swing, time series indicators, and exchange efficiency of the cerebrospinal fluid signal corresponding to the biphasic cerebrospinal fluid event. Based on the corresponding trough and rebound peak of the biphasic cerebrospinal fluid event, a corresponding reliability label is generated by combining the frame-by-frame displacement and the whole-brain signal change of each time frame. The frame-by-frame displacement is determined according to the rigid body motion correction parameters generated in the rigid body motion correction, and the whole-brain signal change is determined according to the signal intensity change of the functional magnetic resonance imaging data.

[0014] Furthermore, the generation of corresponding reliability labels based on the corresponding troughs and rebound peaks of biphasic cerebrospinal fluid events, combined with frame-by-frame shifts and whole-brain signal changes at each time frame, includes: The frame-by-frame displacement and whole-brain signal change of each time frame are compared with the corresponding preset displacement threshold and preset change threshold, respectively. Traverse all time frames, mark the time frames whose frame-by-frame displacement is less than or equal to a preset displacement threshold and whose whole-brain signal change is less than or equal to a preset change threshold as reliable time frames, and obtain the positions of the trough and rebound peak of the cerebrospinal fluid biphasic event within their respective preset time windows. A reliability label is generated when all time frames within the corresponding time window of the biphasic cerebrospinal fluid event are reliable, and the corresponding trough and rebound peak are not located at the window boundary within their respective preset time windows.

[0015] A cerebrospinal fluid biphasic event detection system based on functional magnetic resonance imaging (fMRI), used to perform any of the above-mentioned methods for detecting cerebrospinal fluid biphasic events based on fMRI, includes: The data processing module is used to acquire and preprocess functional magnetic resonance imaging data of the brain. The signal extraction module is used to extract blood oxygenation level-dependent signal time series and cerebrospinal fluid signal time series from preprocessed functional magnetic resonance imaging data. The time alignment module is used to align the blood oxygen level dependent signal time series and the cerebrospinal fluid signal time series to the same physical time axis based on the inter-slice acquisition offset information of functional magnetic resonance imaging data. The event detection module is used to detect local signal peaks in the aligned blood oxygen level dependent signal time series, and the signal characteristic moment within a preset time window before each local signal peak is taken as the corresponding candidate event zero point. The biphasic localization module is used to locate the corresponding valley value within the first preset time window and the corresponding rebound peak within the second preset time window after the zero point of each candidate event in the aligned cerebrospinal fluid signal time series. The event filtering module is used to filter out biphasic cerebrospinal fluid events based on the corresponding valley and rebound peak of the location.

[0016] The beneficial effects of this invention are: After extracting the blood oxygenation level-dependent signal time series and cerebrospinal fluid (CSF) signal time series from whole-brain gray matter using functional magnetic resonance imaging (fMRI) data, the blood oxygenation level-dependent signal and CSF signal are aligned to the same physical time axis based on inter-slice acquisition offset information to reduce the interference of acquisition time series differences between different slices on the determination of event temporal relationships. Furthermore, by detecting local signal peaks from the blood oxygenation level-dependent signal time series of whole-brain gray matter and using the signal characteristic moments within their preceding time windows as candidate event zero points, adaptive temporal anchoring of neurovascular events is achieved, avoiding temporal misalignment caused by individual differences in CSF circulation velocity under the fixed delay assumption. Moreover, the valleys and rebound peaks after the candidate event zero points are sequentially located in the CSF signal, and combined with the pre-event baseline for biphasic morphological joint screening, real biphasic CSF pulsation events are effectively distinguished from non-event-related changes such as single-phase signal drift and random noise fluctuations, improving the accuracy and reliability of CSF biphasic event detection, and thus improving the accuracy of subsequent functional quantification results based on CSF coupled events. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a process of the present invention; Figure 2 This is a schematic diagram of a typical event waveform that simultaneously satisfies motion quality control and timing reliability, as shown in an embodiment of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Example: Methods for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging, such as Figure 1 As shown, it includes: We acquired and preprocessed functional magnetic resonance imaging (fMRI) data of the brain, and extracted the blood oxygenation level-dependent signal time series and cerebrospinal fluid signal time series of the whole brain gray matter from the preprocessed fMRI data. Based on the inter-slice acquisition offset information of functional magnetic resonance imaging data, the blood oxygen level dependent signal time series and the cerebrospinal fluid signal time series are aligned to the same physical time axis; Local signal peaks in the aligned blood oxygen level dependent signal time series are detected, and the signal characteristic moments within a preset time window before each local signal peak are taken as the corresponding candidate event zero points. In the aligned cerebrospinal fluid signal time series, the corresponding valley value within the first preset time window and the corresponding rebound peak within the second preset time window are located after the zero point of each candidate event, wherein the second preset time window is later than the first preset time window; Based on the corresponding trough and rebound peak of the location, biphasic events of cerebrospinal fluid were screened out.

[0020] Considering that the raw functional magnetic resonance imaging (fMRI) data acquired contains various interference signals unrelated to neural activity, the results obtained from direct analysis without processing cannot accurately reflect the true physiological state. Therefore, after acquiring the fMRI data of the brain, preprocessing is performed to improve signal quality.

[0021] The preprocessing of functional magnetic resonance imaging (fMRI) data of the brain includes: Acquire functional magnetic resonance imaging (fMRI) data of the brain, remove the initial frame of the fMRI data, and perform rigid body motion correction on the fMRI data after removing the initial frame. Trend regression and motion parameter regression were performed on the functional magnetic resonance imaging (fMRI) data after rigid body motion correction to obtain preprocessed fMRI data.

[0022] The functional magnetic resonance imaging (fMRI) data refers to four-dimensional fMRI data covering the entire brain gray matter and the lowest layer of cerebrospinal fluid in the cranium, including three-dimensional images of the whole brain at multiple time frames, which are obtained by scanning the subject's brain using fMRI sequences.

[0023] Before scanning with functional magnetic resonance imaging (fMRI) sequences, the repetition time of the imaging sequence is set to 1.0 second or less to ensure sufficient sampling sensitivity for rapid dynamic changes in cerebrospinal fluid (CSF) signals. Simultaneously, the scanning field of view is set to cover the entire brain gray matter region and the lowest layer of CSF within the cranium, to simultaneously acquire whole-brain gray matter oxygenation level-dependent signals reflecting neural activity and CSF signals reflecting intracranial volume changes. Furthermore, during the scanning process, the repetition time of each volume acquisition, inter-slice acquisition offset information for each acquisition layer, and spatial geometric information are automatically recorded.

[0024] The inter-slice acquisition offset information refers to the inherent offset between the acquisition times of different slices during functional magnetic resonance imaging (fMRI). Specifically, fMRI uses a slice-by-slice scanning method to acquire whole-brain images. A complete three-dimensional volumetric image consists of multiple two-dimensional slices. These slices are not acquired at the same time, but sequentially in a certain order. Therefore, different slices have a fixed time offset relative to the starting time of the same volume. For example, the offset of the first acquired slice is close to zero, and the offset of the last acquired slice is close to one repetition time minus the acquisition interval. This offset is an inherent property of the scanning sequence, determined by the principles of imaging physics, and does not change with the subject or the number of scans.

[0025] The spatial geometric information is a set of metadata in functional magnetic resonance imaging data that describes the spatial location and geometric properties of the image. It mainly includes information such as the dimension of the image matrix, the actual physical size of voxels in each direction, the position of the origin of the image coordinate system, the tilt angle of the slice, and the slice gap.

[0026] At the start of a functional magnetic resonance imaging (fMRI) scan, the signals in the first few frames often exhibit unstable fluctuations because the magnetic field is not yet fully stable and the subject's tissues have not yet reached steady-state magnetization. If these initial frames are retained in the data, spurious signal changes unrelated to neural activity will be introduced. Therefore, the initial frames of the acquired fMRI data are removed. The number of initial frames can be determined based on the scan sequence and the time required for the magnetic field to reach steady state, for example, removing the first 5 to 10 frames, which can be set according to specific needs.

[0027] During the scanning process, subjects inevitably experience slight head movements, causing the brain's position in the image to shift over time, thus introducing interference signals caused by head movements into the functional magnetic resonance imaging (fMRI) data. Therefore, rigid body motion correction is performed on the fMRI data after removing the initial frame, aligning all time-frame images to the same spatial position and reducing image misalignment errors caused by head movements. Specifically, the head can be considered a rigid body without deformation, and its spatial position changes can be decomposed into six degrees of freedom rigid body motion parameters in three translational and three rotational directions. During rigid body motion correction, the current first frame image or the average image of all time frames is used as the reference image. An image registration algorithm estimates the six rigid body motion parameters of each frame relative to the reference image, determining a set of optimal parameters that minimizes the difference between the transformed image and the reference image. Then, based on these parameters, a three-dimensional spatial transformation is performed on each frame image, aligning all time-frame images to the same spatial position, resulting in rigid body motion-corrected fMRI data. The image registration algorithm can employ a least-squares-based method.

[0028] In addition to neural activity-related signals, functional magnetic resonance imaging (fMRI) signals also contain various non-interest noise components. Low-frequency signal drift primarily originates from the slow changes in the scanner's magnetic field over time and the subject's circadian rhythm, manifesting as a linear or non-linear trend over time. This low-frequency signal drift is corrected using trend regression. Specifically, a polynomial or spline function is fitted to the time series of each voxel to estimate and remove the corresponding trend component, causing the signal to fluctuate around the zero mean. Head motion-related noise, after rigid body motion correction, is further filtered out using motion parameter regression. For each voxel, a linear regression model is established with the voxel's time series as the dependent variable and 24 regression variables as independent variables. The regression coefficients are estimated using the least squares method, the fitted value is calculated, and the fitted value is subtracted from the original data, retaining the residual as the regressed signal. Furthermore, trend regression can be combined with motion parameter regression, with the trend term and the 24 motion parameters forming an extended confounding matrix, completing the regression in one step.

[0029] After trend regression and motion parameter regression processing, preprocessed functional magnetic resonance imaging (fMRI) data were obtained. Non-neural activity-related noise in the data was effectively suppressed, and the extracted blood oxygen level-dependent signals and cerebrospinal fluid signals could more realistically reflect neural activity and the cerebrospinal fluid response process it drives.

[0030] Since functional magnetic resonance imaging (fMRI) data is four-dimensional data, it contains three spatial dimensions and one temporal dimension, which is the signal intensity information of each voxel in each time frame. If we analyze each voxel individually, the amount of data is huge and the signals of each voxel are inconsistent, resulting in a large amount of computation and low efficiency.

[0031] Therefore, further time series of blood oxygenation level-dependent signals of whole-brain gray matter, which can reflect the trend of neural activity throughout the whole brain, and time series of cerebrospinal fluid signals, which can reflect the cerebrospinal fluid flow response induced by neural activity, are extracted from the preprocessed functional magnetic resonance imaging data. This is to convert four-dimensional functional magnetic resonance imaging data into one-dimensional time series data, ensuring processing efficiency while ensuring the integrity of data information and avoiding deviation between analysis results and reality.

[0032] The extraction of blood oxygenation level-dependent signal time series and cerebrospinal fluid signal time series from preprocessed functional magnetic resonance imaging data includes: The preset whole-brain gray matter template is registered to the image coordinate system of the functional magnetic resonance imaging data, and the registered whole-brain gray matter template is superimposed on the preprocessed functional magnetic resonance images of each time frame. The lowest intracranial cerebrospinal fluid level is located on the average image of the preprocessed functional magnetic resonance imaging data. Within the lowest intracranial cerebrospinal fluid level, the region with an intracranial cerebrospinal fluid signal intensity higher than a preset threshold is selected as the region of interest. The region of interest is then superimposed onto the functional magnetic resonance images of each time frame after preprocessing. The average value of all voxel signals in the functional magnetic resonance images of each time frame after preprocessing is calculated within the coverage area of ​​the whole brain gray matter template and within the coverage area of ​​the region of interest. The corresponding average value calculated within the coverage area of ​​the whole brain gray matter template is used as the whole brain gray matter blood oxygenation level dependent signal value of each time frame, and the corresponding average value calculated within the coverage area of ​​the region of interest is used as the cerebrospinal fluid signal value of each time frame. The blood oxygenation level-dependent signal values ​​of whole-brain gray matter and the cerebrospinal fluid signal values ​​of each time frame were sorted in chronological order to obtain the time series of blood oxygenation level-dependent signal and cerebrospinal fluid signal of whole-brain gray matter.

[0033] The whole-brain gray matter template can be a gray matter probability map from a standard brain atlas. Before registering this gray matter probability map from the standard space to the image coordinate system of the functional magnetic resonance imaging (fMRI) data, the spatial range of the whole-brain gray matter in the fMRI data is first determined. Specifically, high-resolution T1-weighted structural images of the subject are first acquired. These T1-weighted structural images are magnetic resonance images based on tissue T1 relaxation time differences, which can clearly show the anatomical boundaries of brain tissues such as gray matter and white matter. Their spatial resolution is higher than that of fMRI data, and they can be used as an anatomical reference for spatial localization. After acquiring the T1-weighted structural images, the whole-brain gray matter probability map from the standard brain atlas is then used to register it from the standard space to the image coordinate system of the fMRI data. During registration, the whole-brain gray matter probability map in standard space is first registered to the individual T1 space to obtain the transformation parameters from standard space to T1 space. Then, the individual T1-weighted structural image is registered with the average image of the functional magnetic resonance imaging (fMRI) data to obtain the transformation parameters from T1 space to functional image space. The two transformation parameters are cascaded to form a complete transformation from standard space to functional image space, thereby mapping the gray matter probability map to the image coordinate system of the fMRI data. After registration, the spatial location covered by the gray matter probability map is spatially mapped to the voxel signal at the corresponding location in the fMRI images of each time frame, thus superimposing the gray matter probability map onto the preprocessed fMRI images of each time frame.

[0034] Simultaneously, the lowest intracranial cerebrospinal fluid (CSF) layer is first determined on the average image of the preprocessed functional magnetic resonance imaging (fMRI) data. This lowest CSF layer is located at the skull base, and its signal intensity is significantly higher than that of adjacent brain tissue. Therefore, the lowest CSF layer can be located by identifying the region at the bottom with the highest CSF signal intensity in the average image of the fMRI data. Within this lowest CSF layer, a region with a CSF signal intensity higher than a preset threshold is selected as the region of interest (ROI). The defined ROI should be a high-signal area of ​​intracranial CSF, excluding areas such as the eyeball, nasopharynx, external air, prominent large blood vessels, and tissue boundaries. The determined ROI is then superimposed onto the preprocessed fMRI images of each time frame, establishing a spatial correspondence between the coverage area of ​​the ROI and the corresponding voxels in each time frame image.

[0035] For the gray matter probability map, a weighted average of the voxel signals is calculated based on the probability values ​​corresponding to the superimposed values. This involves multiplying each voxel signal value by its corresponding probability weight, summing the results, and then dividing by the sum of the weights. Alternatively, the probability map can be binarized, such as converting it into a binary mask using a preset probability threshold. The average value of all voxels within the mask's coverage area is then calculated. The calculated average values ​​for each time frame are used as the whole-brain gray matter oxygenation level-dependent signal values ​​for each time frame. These are then arranged in chronological order to form a time series of whole-brain gray matter oxygenation level-dependent signals. For regions of interest (ROIs), the average signal value of all voxels within its coverage area is directly calculated. The calculated average values ​​for each time frame are then used as the cerebrospinal fluid (CSF) signal values ​​for each time frame. These are then arranged in chronological order to form a CSF signal time series.

[0036] After obtaining the time series of blood oxygen level dependent signals and cerebrospinal fluid signals, since the two types of signals are located in different acquisition layers, the corresponding acquisition time offsets are also different. If event detection is performed directly, it will introduce systematic bias and lead to errors in the analysis results. Therefore, the two types of signals are further time-aligned to ensure the accuracy of the time relationship determination in event detection.

[0037] The step of aligning the blood oxygenation level-dependent signal time series and the cerebrospinal fluid signal time series to the same physical time axis based on the inter-slice acquisition offset information of functional magnetic resonance imaging data includes: Based on the inter-slice acquisition offset information of functional magnetic resonance imaging data, the inter-slice acquisition offset of the acquisition layers corresponding to blood oxygen level dependent signals and cerebrospinal fluid signals is determined respectively. The acquisition time of each time frame of functional magnetic resonance imaging is superimposed with the inter-layer acquisition offset of the acquisition layer corresponding to the blood oxygen level dependent signal to determine the actual physical sampling time of the blood oxygen level dependent signal in each time frame. The acquisition time of each time frame of functional magnetic resonance imaging is superimposed with the inter-slice acquisition offset of the corresponding acquisition layer of cerebrospinal fluid signal to determine the actual physical sampling time of cerebrospinal fluid signal in each time frame. According to the preset time grid, the blood oxygen level dependent signal time series and cerebrospinal fluid signal time series are interpolated according to the corresponding actual physical sampling time to align to the same physical time axis.

[0038] Based on the inter-slice acquisition offset information acquired during functional magnetic resonance imaging (fMRI) sequence scanning, the inter-slice acquisition offset of the corresponding acquisition layers for oxygenation-dependent signals and cerebrospinal fluid (CSF) signals were determined. The oxygenation-dependent signals originated from the whole-brain gray matter, which covers multiple acquisition layers, each with a different acquisition offset. The inter-slice acquisition offset of each acquisition layer in the whole-brain gray matter was first obtained, and then the average offset of the gray matter layer or the offset of a pre-defined representative layer was selected as the inter-slice acquisition offset for the oxygenation-dependent signals. The CSF signals originated from the lowest intracranial CSF layer, and their inter-slice acquisition offset was the same as that of the lowest intracranial CSF layer.

[0039] Then, the acquisition time of each time frame of the functional magnetic resonance imaging (fMRI) image is superimposed with the inter-slice acquisition offset of the corresponding acquisition layer of the blood oxygenation level-dependent signal to determine the actual physical sampling time of the blood oxygenation level-dependent signal in each time frame. Specifically, let the first... The nominal acquisition time of each time frame functional magnetic resonance image is The inter-clinical acquisition offsets of blood oxygenation level-dependent signals and cerebrospinal fluid signals were respectively and Then the physical sampling time for the two types of samples is defined as: ; ; in, For blood oxygen level-dependent signals in the first The actual sampling time of a functional magnetic resonance image at each time frame; For repetition time; This represents the inter-layer acquisition offset of the blood oxygen level-dependent signal; For cerebrospinal fluid signals The actual sampling time of each volume; This represents the inter-layer sampling offset of the cerebrospinal fluid signal.

[0040] Since the time-series signal values ​​of the two types of signals correspond to their respective actual physical sampling times, and these sampling times may differ in their positions on the time axis, they cannot be directly compared or manipulated on the same time axis. Therefore, based on a preset time grid, the blood oxygen level dependent signal time series and the cerebrospinal fluid signal time series are interpolated according to their corresponding actual physical sampling times to align them to the same physical time axis. The preset time grid is a pre-determined unified physical time coordinate system, such as a sequence of time points divided by a preset fixed time interval. This fixed time interval can be a repeating time or a finer interval, and can be adjusted according to actual needs. Taking the blood oxygen level dependent signal as an example, given the actual physical sampling times and corresponding signal values ​​of each time frame, interpolation methods, such as linear interpolation or spline interpolation, are used to calculate the signal values ​​at each time point of the preset time grid, resulting in the interpolated blood oxygen level dependent signal time series. Similarly, using the actual physical sampling times and corresponding signal values ​​of each time frame of the cerebrospinal fluid signal, interpolation methods are used to calculate the signal values ​​at each time point of the preset time grid, resulting in the interpolated cerebrospinal fluid signal time series. After interpolation, the two types of signals can have the same event coordinates on the preset time grid, that is, aligned to the same physical time axis, thereby avoiding timing deviations introduced by different acquisition levels.

[0041] In oxygen-dependent signal time series, neuronal activity causes local hemodynamic changes, resulting in a waveform that rises from the baseline to a peak and then falls back, forming a local signal peak. This local signal peak is a direct signal representation of neural activity and can serve as a basic marker of candidate events. However, because the local signal peak is the most easily identifiable feature point in the time series, the labeling of any candidate event must start from this peak. Since the local signal peak itself is located at the end of the rising segment, using only the peak time as the time marker results in a lack of temporal position information within the segment, making it impossible to distinguish different rising patterns of the same peak. Furthermore, the rate of change of signal at different time points within the rising segment is not uniformly distributed; the time position of the maximum rate of change may differ in different rising segments. For example, for the same local signal peak, the maximum rate of change may be near the beginning of the rising segment in some segments, near the end in others, or in the middle. If only the peak time is used as the time marker for the event, when aligning the time axes of different candidate events using the peak time as the reference point, events with the same peak time but different rising segments will be assigned the same time reference position, making it impossible to distinguish between events.

[0042] Therefore, by selecting the signal characteristic moment as the candidate event zero point within a preset time window before the local signal peak, the morphological differences within the rising segment can be distinguished, thus achieving effective event differentiation.

[0043] The detection of local signal peaks in the time series of blood oxygen level-dependent signals after alignment, with the signal characteristic moments within a preset time window preceding each local signal peak as the corresponding candidate event zero points, includes: In the aligned blood oxygen level dependent signal time series, local maxima with signal amplitudes greater than a preset amplitude threshold are selected as candidate peaks, and candidate peaks with time intervals between adjacent candidate peaks greater than a preset interval threshold are identified as local signal peaks. Within a preset four-window period before each local signal peak, the positive slope value of the blood oxygen level dependent signal is calculated point by point in time. The time point with the largest positive slope value is taken as the signal characteristic moment, and the signal characteristic moment is taken as the zero point of the corresponding candidate event.

[0044] When detecting local signal peaks in an aligned blood oxygen level-dependent signal time series, the time series is scanned point by point, and local maxima with signal amplitudes greater than a preset amplitude threshold are selected as candidate peaks. The preset amplitude threshold can be determined based on the time series standard deviation of the signal or set as a preset percentile of the signal amplitude distribution.

[0045] Among the identified candidate peaks, the time intervals between adjacent candidate peaks are compared sequentially. Candidate peaks with a time interval greater than a preset interval threshold are identified as local signal peaks. For candidate peaks with a time interval less than or equal to the preset interval threshold, only the one with the larger amplitude is retained as a local signal peak, and candidate peaks with smaller amplitudes are excluded.

[0046] For each detected local signal peak, the start time of the local signal peak is determined by tracing back the length of a preset time window, using the peak time as the endpoint. This defines the time range from the start time to the local signal peak time. Within this preset time window, the first derivative of the blood oxygen level-dependent signal is calculated point-by-point. The first derivative is calculated by dividing the signal difference between adjacent time points by the corresponding time interval. Within the preset time window, the first derivative values ​​are compared point-by-point. The time point with the largest positive first derivative value is identified as the signal characteristic moment, and this characteristic moment is used as the candidate event zero point for the corresponding candidate event.

[0047] The filtering expression for the signal characteristic moments is: ; in, For the first The signal characteristic moment of each candidate event, i.e., the event zero point; For the first Local signal peak moments of each candidate event; In this embodiment, the length of the pre-peak time window is preset. ; For the first The corresponding blood oxygen level dependent signal time curves for each candidate event; To trace back from the local signal peak time The time interval between the second and the peak moment, i.e. the pre-peak preset time window; This is the first derivative of the blood oxygen level-dependent signal with respect to time, i.e., the slope of the signal at each moment.

[0048] After determining the signal characteristic time of each local signal peak based on the filtering expression of the signal characteristic time, the candidate event zero point corresponding to each candidate event is obtained by taking the signal characteristic time as the corresponding candidate event zero point.

[0049] Cerebrospinal fluid (CSF) signals exhibit a specific temporal pattern of change following a neurological event. In the early stages after the event, intracranial vasodilation increases cerebral blood volume, disrupting intracranial volume balance and causing CSF to be expelled. This results in a decrease in the signal of the lowest layer of CSF, manifested as a local minimum (trough) in the waveform, occurring within the early time window after the event's zero point. As neural activity ceases and blood vessels return to rest, the expelled CSF flows back, causing a compensatory rebound in the signal of the lowest layer of CSF, manifested as a local maximum (bounce) in the waveform, occurring within the later time window after the trough. Thus, the trough and bounce together form the complete waveform pattern of a biphasic CSF event.

[0050] Based on this, the valleys and rebound peaks of each candidate event are first located in the cerebrospinal fluid signal sequence based on the zero point of the candidate event, so as to provide basic data for subsequent judgment of biphasic cerebrospinal fluid events.

[0051] The step of locating the corresponding valley value within the first preset time window and the corresponding rebound peak within the second preset time window after locating the zero point of each candidate event in the aligned cerebrospinal fluid signal time series includes: Calculate the median of the cerebrospinal fluid signal amplitude at all time points within the preset time window before the zero point of the corresponding candidate event, and determine the pre-event baseline of each candidate event based on the calculated median. Correct the corresponding cerebrospinal fluid signal based on the baseline before the corresponding event; Starting from the zero point of the corresponding candidate event, the amplitude of the corrected cerebrospinal fluid signal is compared point by point within the first preset time window after the zero point of the candidate event, and the time point with the smallest cerebrospinal fluid signal amplitude is taken as the valley value of the corresponding candidate event. Within a second preset time window after the zero point of the candidate event, the amplitude of the corrected cerebrospinal fluid signal is compared point by point, and the time point with the largest cerebrospinal fluid signal amplitude is taken as the rebound peak of the corresponding candidate event.

[0052] Considering the influence of baseline drift, physiological fluctuations, and average changes in cerebrospinal fluid (CSF) signal levels during signal acquisition, different candidate events will have different CSF signal amplitude levels before their respective zero points. This means that the pre-event baseline values ​​differ for each event. The trough and rebound peaks reflect the deviation of the signal from the pre-event baseline; the trough is the lowest deviation during the descent phase, and the rebound peak is the highest deviation during the ascent phase. If the original, located signal amplitude is directly used for biphasic event determination, the baseline differences between different events will be mixed into the values ​​of the trough and rebound peaks. Two events may be classified as having the same result, leading to misclassification of non-biphasic events as biphasic events or omission of biphasic events. Therefore, baseline correction of the CSF signal is performed before extracting the trough and rebound peaks.

[0053] For each candidate event, taking the event zero point as a reference, the amplitude of the cerebrospinal fluid signal is acquired point by point within a preset baseline time window before the event zero point. The amplitudes at each time point are sorted, and the value in the middle position after sorting, i.e., the median, is taken as the pre-event baseline of the cerebrospinal fluid signal for that candidate event. Its expression is: ; in, For the first Pre-event baseline of cerebrospinal fluid signaling for candidate events; The preset baseline window for the event has a value range of -5 seconds to -2 seconds. The time of the relative event; For the first The absolute moment of the zero point of each candidate event on a unified physical time axis; This represents the absolute time of each time point within the baseline window on a unified physical time axis; For the first Candidate events at time points The amplitude of the original cerebrospinal fluid signal at the location, This indicates the median operation.

[0054] After obtaining the pre-event baseline of the cerebrospinal fluid (CSF) signal, the pre-event baseline is subtracted from the raw CSF signal amplitude at each time point to obtain the baseline-corrected CSF signal curve, which is expressed as follows: ; in, For the first Baseline-corrected cerebrospinal fluid signals for each candidate event.

[0055] The corrected signal amplitude reflects the deviation of the signal from its own baseline at that moment. A corrected amplitude of 0 indicates that the signal is equal to the baseline value at that moment, a corrected amplitude greater than 0 indicates that the signal is higher than the baseline value, and a corrected amplitude less than 0 indicates that the signal is lower than the baseline value.

[0056] After baseline correction, the troughs and rebound peaks are located on the baseline-corrected cerebrospinal fluid (CSF) signal curve. Starting from the zero point of the corresponding candidate event, the amplitude of the CSF signal is acquired point-by-point within the first preset time window after the zero point. The amplitudes at each time point are compared, and the time point with the smallest amplitude is taken as the trough of the candidate event. The time point containing the trough is the trough moment, and the corresponding signal amplitude is the trough amplitude. Its location expression is: ; in, For the first The trough moment of each candidate event; This is the valley value search window, i.e., the first preset time window; The CSF event curve after baseline correction; This refers to the relative event time.

[0057] To locate the rebound peak, the amplitude of the cerebrospinal fluid signal is acquired point-by-point within a second preset time window after the zero point of the candidate event. The amplitudes at each time point are compared, and the time point with the largest amplitude is taken as the rebound peak of the candidate event. The time point where the rebound peak occurs is the rebound peak moment, and the signal amplitude corresponding to that moment is the rebound peak amplitude. Its location expression is as follows: ; in, For the first The rebound peak times of each candidate event, expressed in relative time; This is the rebound peak search window, i.e., the second preset time window; The constraint is that the rebound peak must occur at least one repetition time later than the trough.

[0058] In locating the trough and rebound peak, the second preset time window is later than the first preset time window to ensure that the location of the rebound peak occurs within a time period after the location of the trough.

[0059] After location is completed, cerebrospinal fluid biphasic event identification and judgment are performed in combination with preset event qualification conditions.

[0060] The method of screening for biphasic cerebrospinal fluid events based on location-based corresponding troughs and rebound peaks includes: Compare the times of the corresponding trough and rebound peak, and determine whether the amplitudes of the corresponding trough and rebound peak are positive respectively; If the amplitude of the corresponding trough is not positive, the amplitude of the corresponding rebound peak is positive, and the time of the corresponding rebound peak is later than the time of the corresponding trough, the candidate event is judged to meet the event qualification conditions. Candidate events that meet the event eligibility criteria are classified as biphasic events in cerebrospinal fluid.

[0061] For each candidate event, the chronological order of the trough and rebound peak times is compared, and the amplitudes of the trough and rebound peaks are determined to satisfy the directional conditions. Since the trough and rebound peaks are located on the baseline-corrected cerebrospinal fluid signal curve, the amplitude of the corrected signal directly reflects the deviation of the signal from its own baseline level at that moment. A corrected amplitude less than 0 indicates that the signal is below the baseline, a corrected amplitude greater than 0 indicates that the signal is above the baseline, and a corrected amplitude equal to 0 indicates that the signal is equal to the baseline.

[0062] Based on this, when making a judgment, the trough amplitude is compared with 0. If the trough amplitude is less than 0, it indicates that the cerebrospinal fluid signal drops below the baseline level in the early stage after the event, that is, there is a decline phase. If the trough amplitude is greater than or equal to 0, it indicates that there is no real decline phase in the candidate event, that is, the signal does not drop below the baseline, and the candidate event does not meet the condition of decline first in biphasic events.

[0063] Next, compare the rebound peak amplitude with 0. If the rebound peak amplitude is greater than 0, it indicates that the cerebrospinal fluid signal rebounds to a level higher than the baseline after the decline, that is, there is a rebound phase. If the rebound peak amplitude is less than or equal to 0, it indicates that there is no real rebound phase for the candidate event, that is, the signal is not higher than the baseline, and the candidate event does not meet the condition of a rebound in a biphasic event.

[0064] Simultaneously, the order of the trough and rebound peak times is compared to determine whether the rebound peak time is later than the trough time. Since the physiological process of biphasic cerebrospinal fluid events is characterized by a decline followed by a rebound, the lowest point of the decline phase should occur earlier than the highest point of the rise phase. If the rebound peak time is earlier than or equal to the trough time, the order of the signals does not conform to the physiological time sequence of biphasic events.

[0065] The preset event qualification criteria are: when the trough amplitude of a candidate event is less than 0, the rebound peak amplitude is greater than 0, and the rebound peak time is later than the trough time, the candidate event is determined to be a biphasic cerebrospinal fluid event if all the event qualification criteria are met. If a candidate event fails to meet any one of the conditions, the candidate event is determined to be a non-biphasic event.

[0066] After identifying biphasic cerebrospinal fluid events, the following steps were also performed: Based on the corresponding trough, rebound peak, pre-event baseline, and blood oxygen level dependent signal of the biphasic cerebrospinal fluid event, corresponding quantitative indicators are calculated. The quantitative indicators include the falling area, rising area, total exchange area, peak-trough swing, time series indicators, and exchange efficiency of the cerebrospinal fluid signal corresponding to the biphasic cerebrospinal fluid event. Based on the corresponding trough and rebound peak of the biphasic cerebrospinal fluid event, a corresponding reliability label is generated by combining the frame-by-frame displacement and the whole-brain signal change of each time frame. The frame-by-frame displacement is determined according to the rigid body motion correction parameters generated in the rigid body motion correction, and the whole-brain signal change is determined according to the signal intensity change of the functional magnetic resonance imaging data.

[0067] In this embodiment, in addition to correcting the cerebrospinal fluid signal using the pre-event baseline, the blood oxygen level-dependent signal is also corrected using the pre-event baseline, and the corrected blood oxygen level-dependent signal is used as the basis for calculating the quantitative index.

[0068] Specifically, for each biphasic cerebrospinal fluid (CSF) event, the event zero point is used as a reference. Within a preset baseline time window before the event zero point, the amplitude of the oxygenation level-dependent signal is acquired point by point. The amplitudes at each time point are sorted, and the value in the middle position after sorting, i.e., the median, is taken as the pre-event baseline of the oxygenation level-dependent signal for that CSF biphasic event. Its expression is: ; in, For the first Blood oxygenation levels in a biphasic cerebrospinal fluid event depend on the baseline before the signaling event. For the first A biphasic cerebrospinal fluid event at a specific time point The original blood oxygen level at the site depends on the signal amplitude; The time of the relative event; For the first The absolute moment of the event zero point of a biphasic cerebrospinal fluid event on a unified physical time axis; This represents the absolute time of each time point within the baseline window on a unified physical time axis.

[0069] After obtaining the baseline before the oxygenation level-dependent signal event, the baseline before the oxygenation level-dependent signal event is subtracted from the original oxygenation level-dependent signal amplitude at each time point to obtain the baseline-corrected oxygenation level-dependent signal curve, which is expressed as follows: ; in, For the first A baseline-corrected oxygenation level-dependent signal for biphasic cerebrospinal fluid events.

[0070] Based on this, for each candidate event identified as a biphasic cerebrospinal fluid event, its quantitative index is calculated according to the event's trough amplitude, rebound peak amplitude, trough time, rebound peak time, event zero point, pre-event baseline, and the corresponding positive half-wave area of ​​the oxygenation level dependent signal.

[0071] The area of ​​decline reflects the overall magnitude of the drop in cerebrospinal fluid signal from the baseline after the event. It is calculated by integrating the portion of the cerebrospinal fluid signal curve that is less than the pre-event baseline (i.e., the negative deviation) over the time interval from the event's zero point to the rebound peak. The calculation expression is as follows: ; in, For the first The area of ​​decline in cerebrospinal fluid biphasic events; For the first One biphasic event in cerebrospinal fluid; For the first time after baseline correction Cerebrospinal fluid signals of a biphasic cerebrospinal fluid event; It is a positive part function. , For any real number, when Time output Otherwise, output 0.

[0072] The area of ​​rise reflects the overall magnitude of the cerebrospinal fluid signal recovery from its trough. It is calculated by integrating the portion of the cerebrospinal fluid signal curve that exceeds the pre-event baseline (i.e., the positive deviation) over the time interval from the trough to the preset analysis endpoint. The calculation expression is as follows: ; in, For the first The area of ​​elevation in a biphasic cerebrospinal fluid event. For the first The trough of a biphasic cerebrospinal fluid event The preset analysis endpoint, i.e. the pre-set integration cutoff time, is used to limit the end position of the calculation of the rising area. It can be adjusted according to actual needs. In this embodiment, the value can be 10 seconds or 12 seconds after the zero point of the event, so that the rising area ends within a period of time after the rebound peak ends, avoiding the inclusion of subsequent signal fluctuations unrelated to the event into the integration. It is a positive part function; For the first time after baseline correction Cerebrospinal fluid signals in a biphasic cerebrospinal fluid event.

[0073] The total exchange area is the sum of the falling area and the rising area, used to describe the overall exchange load of cerebrospinal fluid signals deviating from the baseline during a complete biphasic cerebrospinal fluid event. Its expression is: ; in, For the first The total exchange area of ​​a biphasic cerebrospinal fluid event.

[0074] Peak-valley amplitude describes the dynamic range of cerebrospinal fluid signal variation from trough to rebound peak in a biphasic cerebrospinal fluid event. It is obtained by calculating the difference between the rebound peak amplitude and the trough amplitude, and its expression is: ; in, For the first The peak-to-trough amplitude of a biphasic cerebrospinal fluid event For the first The rebound peak amplitude of a biphasic cerebrospinal fluid event; For the first The trough amplitude of a biphasic cerebrospinal fluid event.

[0075] The time-series indicators include the time difference from event zero to trough, the time difference from event zero to rebound peak, and the time difference from trough to rebound peak in a biphasic cerebrospinal fluid event, used to characterize the response sequence and recovery speed. Their expressions are as follows: ; ; ; in, For the first The time difference between the zero point and the trough of a biphasic cerebrospinal fluid event; For the first The time difference between event zero and rebound peak in a biphasic cerebrospinal fluid event; For the first The time difference between the trough and the rebound peak in a biphasic cerebrospinal fluid event; For the first The trough moment of a biphasic cerebrospinal fluid event; For the first The rebound peak moment of a biphasic cerebrospinal fluid event.

[0076] The exchange efficiency is used to characterize the exchange response intensity of cerebrospinal fluid signals driven by a unit oxygen level-dependent signal. It is the ratio of the total exchange area to the area of ​​the corresponding positive half-wave of the oxygen level-dependent signal, and is expressed as: ; ; in, For the first The blood oxygen level of a biphasic cerebrospinal fluid event depends on the positive half-wave area of ​​the signal. The preset dose integration window endpoint is a pre-set integration cutoff time point used to limit the calculation end position of the positive half-wave area of ​​the blood oxygen level dependent signal. It can be set to 6 seconds or 8 seconds after the zero point of the event, so that the integration interval of the driving dose covers the main part of the blood oxygen level dependent signal from the beginning to the fall, while avoiding the inclusion of subsequent irrelevant signal fluctuations in the calculation. It is a positive part function; For the first A baseline-corrected oxygenation level-dependent signal for a biphasic cerebrospinal fluid event; For the first The exchange efficiency of biphasic cerebrospinal fluid events; For the first The total exchange area of ​​a biphasic cerebrospinal fluid event.

[0077] By calculating quantitative indicators, the waveform morphology of biphasic cerebrospinal fluid events can be quantified into numerical form. Different indicators correspond to different aspects of the waveform's numerical characteristics, avoiding the ambiguity caused by the mutual cancellation of decreasing and rebound signals in a single correlation coefficient. Statistical comparisons can be directly performed between subjects or groups based on these values, thereby improving the efficiency of quantitative analysis of biphasic cerebrospinal fluid events.

[0078] In addition to quantitative indicators, corresponding reliability labels are generated so that the differences in the credibility of each cerebrospinal fluid biphasic event in terms of temporal localization and motion interference can be expressed in a labeled form. After the qualified events are screened by reliability labels, the numerical differences between events can reflect the real physiological differences, rather than the noise differences introduced by head movement or localization errors, thus ensuring the accuracy of quantitative analysis of cerebrospinal fluid biphasic events.

[0079] The method for generating corresponding reliability labels based on the corresponding troughs and rebound peaks of biphasic cerebrospinal fluid events, combined with frame-by-frame shifts and whole-brain signal changes at each time frame, includes: The frame-by-frame displacement and whole-brain signal change of each time frame are compared with the corresponding preset displacement threshold and preset change threshold, respectively. Traverse all time frames, mark the time frames whose frame-by-frame displacement is less than or equal to a preset displacement threshold and whose whole-brain signal change is less than or equal to a preset change threshold as reliable time frames, and obtain the positions of the trough and rebound peak of the cerebrospinal fluid biphasic event within their respective preset time windows. A reliability label is generated when all time frames within the corresponding time window of the biphasic cerebrospinal fluid event are reliable, and the corresponding trough and rebound peak are not located at the window boundary within their respective preset time windows.

[0080] When generating reliability labels, the frame-by-frame displacement and whole-brain signal change of each cerebrospinal fluid biphasic event within a preset time window before and after the event are obtained.

[0081] The frame-by-frame displacement is determined based on the rigid body motion correction parameters generated during rigid body motion correction. After rigid body motion correction, motion parameters in three translational and three rotational directions can be obtained for each time frame. The displacement of each time frame relative to the previous frame is calculated based on these parameters. The calculated frame-by-frame displacement is used to quantify the degree of head movement of the subject during the scanning process.

[0082] The change in whole-brain signal is determined based on the signal intensity changes in functional magnetic resonance imaging data. Specifically, for each time frame, the change in signal intensity of each voxel in the whole brain range of that frame relative to the previous frame is calculated, and the root mean square of the changes in all voxels is taken as the change in whole-brain signal for that time frame. This value can reflect the overall intensity of the change in whole-brain blood oxygenation level dependent signal from one frame to the next.

[0083] After acquiring the frame-by-frame displacement and whole-brain signal change within a preset time window before and after each biphasic cerebrospinal fluid event, the frame-by-frame displacement of each time frame is compared with a preset displacement threshold, and the whole-brain signal change of each time frame is compared with a preset change threshold. The preset displacement threshold and preset change threshold are fixed values ​​set in advance to determine whether the signal of each time frame is affected by motion interference. For example, the preset displacement threshold can be set to 0.30 mm, and the preset change threshold can be set to 3.0 mm. The specific values ​​can be predetermined according to the scanning protocol of the functional magnetic resonance imaging sequence.

[0084] Iterate through all time frames within the time window corresponding to the biphasic cerebrospinal fluid event, and mark time frames whose frame-by-frame displacement is less than or equal to a preset displacement threshold and whose whole-brain signal change is less than or equal to a preset change threshold as reliable time frames. If all time frames within the time window corresponding to the biphasic cerebrospinal fluid event are reliable time frames, then the event passes motion quality control and is marked as motion reliable; otherwise, it is marked as motion unreliable.

[0085] Simultaneously, the positions of the troughs and rebound peaks of each biphasic cerebrospinal fluid event within their respective preset time windows are obtained, and it is determined whether the troughs and rebound peaks are located at the boundaries of their respective preset time windows. When the trough or rebound peak is located at the boundary of the search window, the true location of the extreme point may be outside the search window range, reducing the reliability of the time series indicator. If neither the trough nor the rebound peak is located at the boundary of its respective search window, the biphasic cerebrospinal fluid event is marked as time-reliable; otherwise, it is marked as time-unreliable.

[0086] An event is marked as reliable if it satisfies both motion reliability and timing reliability; otherwise, it is marked as unreliable or left unmarked. After generating a reliability label, this label is output along with the corresponding event's quantitative metrics, indicating the reliability of the event's quantitative metrics in terms of timing positioning and motion disturbances.

[0087] Taking a resting-state functional magnetic resonance imaging (fMRI) image as an example, the biphasic event detection method for cerebrospinal fluid (CSF) in this embodiment is verified. The scanning parameters are: repetition time TR = 0.607 seconds, a total of 482 time frames are collected; the region of interest of the lowest CSF layer is located at z=0 level, containing 56 voxels; the interlayer acquisition offset of the CSF acquisition layer is 0.440 seconds.

[0088] Based on the cerebrospinal fluid biphasic event detection method of this embodiment, a total of 13 candidate events were detected. After judgment, 5 of the candidate events met the conditions of negative trough value, positive rebound peak, and rebound peak later than trough value, and were screened as cerebrospinal fluid biphasic events, with a biphasic qualification rate of 38.5%. Moreover, all 5 detected cerebrospinal fluid biphasic events passed the motion quality control, and the trough value and rebound peak of one event were not located at the boundary of their respective search windows, meeting the temporal reliability condition.

[0089] Five qualified biphasic events were quantitatively analyzed, and the median values ​​of each indicator are as follows: the trough amplitude of the cerebrospinal fluid (CSF) signal was -0.480%, the rebound peak amplitude was 3.373%, the peak-to-trough swing was 4.248%, the total exchange area was 23.143%·s, and the exchange efficiency was 4.350% / z. Therefore, this embodiment can effectively identify CSF biphasic events from functional magnetic resonance imaging (fMRI) data and quantify the descent amplitude, rebound amplitude, exchange load, and exchange efficiency of qualified CSF biphasic events.

[0090] in, Figure 2 This illustrates a typical event that simultaneously satisfies both motion quality control and timing reliability. After the maximum upward slope of the blood oxygen level-dependent signal, the cerebrospinal fluid signal first shows a trough, followed by a rebound peak, exhibiting a complete biphasic waveform of first decreasing and then rebounding.

[0091] Another aspect of this embodiment provides a cerebrospinal fluid biphasic event detection system based on functional magnetic resonance imaging, including: The data processing module is used to acquire and preprocess functional magnetic resonance imaging data of the brain. The signal extraction module is used to extract blood oxygenation level-dependent signal time series and cerebrospinal fluid signal time series from preprocessed functional magnetic resonance imaging data. The time alignment module is used to align the blood oxygen level dependent signal time series and the cerebrospinal fluid signal time series to the same physical time axis based on the inter-slice acquisition offset information of functional magnetic resonance imaging data. The event detection module is used to detect local signal peaks in the aligned blood oxygen level dependent signal time series, and the signal characteristic moment within a preset time window before each local signal peak is taken as the corresponding candidate event zero point. The biphasic localization module is used to locate the corresponding valley value within the first preset time window and the corresponding rebound peak within the second preset time window after the zero point of each candidate event in the aligned cerebrospinal fluid signal time series. The event filtering module is used to filter out biphasic cerebrospinal fluid events based on the corresponding valley and rebound peak of the location.

[0092] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Other variations and modifications are possible without departing from the technical solutions described in the claims.

Claims

1. A method for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging, characterized in that, include: We acquired and preprocessed functional magnetic resonance imaging (fMRI) data of the brain, and extracted the blood oxygenation level-dependent signal time series and cerebrospinal fluid signal time series of the whole brain gray matter from the preprocessed fMRI data. Based on the inter-slice acquisition offset information of functional magnetic resonance imaging data, the blood oxygen level dependent signal time series and the cerebrospinal fluid signal time series are aligned to the same physical time axis; Local signal peaks in the aligned blood oxygen level dependent signal time series are detected, and the signal characteristic moments within a preset time window before each local signal peak are taken as the corresponding candidate event zero points. In the aligned cerebrospinal fluid signal time series, the corresponding valley value within the first preset time window and the corresponding rebound peak within the second preset time window are located after the zero point of each candidate event, wherein the second preset time window is later than the first preset time window; Based on the corresponding trough and rebound peak of the location, biphasic events of cerebrospinal fluid were screened out.

2. The method for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging according to claim 1, characterized in that, The extraction of blood oxygenation level-dependent signal time series and cerebrospinal fluid signal time series from preprocessed functional magnetic resonance imaging data includes: The preset whole-brain gray matter template is registered to the image coordinate system of the functional magnetic resonance imaging data, and the registered whole-brain gray matter template is superimposed on the preprocessed functional magnetic resonance images of each time frame. The lowest intracranial cerebrospinal fluid level is located on the average image of the preprocessed functional magnetic resonance imaging data. Within the lowest intracranial cerebrospinal fluid level, the region with an intracranial cerebrospinal fluid signal intensity higher than a preset threshold is selected as the region of interest. The region of interest is then superimposed onto the functional magnetic resonance images of each time frame after preprocessing. The average value of all voxel signals in the functional magnetic resonance images of each time frame after preprocessing is calculated within the coverage area of ​​the whole brain gray matter template and within the coverage area of ​​the region of interest. The corresponding average value calculated within the coverage area of ​​the whole brain gray matter template is used as the whole brain gray matter blood oxygenation level dependent signal value of each time frame, and the corresponding average value calculated within the coverage area of ​​the region of interest is used as the cerebrospinal fluid signal value of each time frame. The blood oxygenation level-dependent signal values ​​of whole-brain gray matter and the cerebrospinal fluid signal values ​​of each time frame were sorted in chronological order to obtain the time series of blood oxygenation level-dependent signal and cerebrospinal fluid signal of whole-brain gray matter.

3. The method for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging according to claim 1, characterized in that, The step of aligning the blood oxygenation level-dependent signal time series and the cerebrospinal fluid signal time series to the same physical time axis based on the inter-slice acquisition offset information of functional magnetic resonance imaging data includes: Based on the inter-slice acquisition offset information of functional magnetic resonance imaging data, the inter-slice acquisition offset of the acquisition layers corresponding to blood oxygen level dependent signals and cerebrospinal fluid signals is determined respectively. The acquisition time of each time frame of functional magnetic resonance imaging is superimposed with the inter-layer acquisition offset of the acquisition layer corresponding to the blood oxygen level dependent signal to determine the actual physical sampling time of the blood oxygen level dependent signal in each time frame. The acquisition time of each time frame of functional magnetic resonance imaging is superimposed with the inter-slice acquisition offset of the corresponding acquisition layer of cerebrospinal fluid signal to determine the actual physical sampling time of cerebrospinal fluid signal in each time frame. According to the preset time grid, the blood oxygen level dependent signal time series and cerebrospinal fluid signal time series are interpolated according to the corresponding actual physical sampling time to align to the same physical time axis.

4. The method for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging according to claim 1, characterized in that, The detection of local signal peaks in the aligned blood oxygen level-dependent signal time series, with the signal characteristic moments within a preset time window preceding each local signal peak as the corresponding candidate event zero points, includes: In the aligned blood oxygen level dependent signal time series, local maxima with signal amplitudes greater than a preset amplitude threshold are selected as candidate peaks, and candidate peaks with time intervals between adjacent candidate peaks greater than a preset interval threshold are identified as local signal peaks. Within a preset four-window period before each local signal peak, the positive slope value of the blood oxygen level dependent signal is calculated point by point in time. The time point with the largest positive slope value is taken as the signal characteristic moment, and the signal characteristic moment is taken as the zero point of the corresponding candidate event.

5. The method for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging according to claim 1, characterized in that, The step of locating the corresponding valley value within the first preset time window and the corresponding rebound peak within the second preset time window after locating the zero point of each candidate event in the aligned cerebrospinal fluid signal time series includes: Calculate the median of the cerebrospinal fluid signal amplitude at all time points within the preset time window before the zero point of the corresponding candidate event, and determine the pre-event baseline of each candidate event based on the calculated median. Correct the corresponding cerebrospinal fluid signal based on the baseline before the corresponding event; Starting from the zero point of the corresponding candidate event, the amplitude of the corrected cerebrospinal fluid signal is compared point by point within the first preset time window after the zero point of the candidate event, and the time point with the smallest cerebrospinal fluid signal amplitude is taken as the valley value of the corresponding candidate event. Within a second preset time window after the zero point of the candidate event, the amplitude of the corrected cerebrospinal fluid signal is compared point by point, and the time point with the largest cerebrospinal fluid signal amplitude is taken as the rebound peak of the corresponding candidate event.

6. The method for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging according to claim 1, characterized in that, The location-based corresponding trough and rebound peaks are used to screen for biphasic cerebrospinal fluid events, including: Compare the times of the corresponding trough and rebound peak, and determine whether the amplitudes of the corresponding trough and rebound peak are positive respectively; If the amplitude of the corresponding trough is not positive, the amplitude of the corresponding rebound peak is positive, and the time of the corresponding rebound peak is later than the time of the corresponding trough, the candidate event is judged to meet the event qualification conditions. Candidate events that meet the event eligibility criteria are classified as biphasic events in cerebrospinal fluid.

7. The method for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging according to claim 1, characterized in that, The acquisition and preprocessing of functional magnetic resonance imaging (fMRI) data of the brain includes: Acquire functional magnetic resonance imaging (fMRI) data of the brain, remove the initial frame of the fMRI data, and perform rigid body motion correction on the fMRI data after removing the initial frame. Trend regression and motion parameter regression were performed on the functional magnetic resonance imaging (fMRI) data after rigid body motion correction to obtain preprocessed fMRI data.

8. The method for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging according to claim 7, characterized in that, After identifying biphasic cerebrospinal fluid events, the following steps were also performed: Based on the corresponding trough, rebound peak, pre-event baseline, and blood oxygen level dependent signal of the biphasic cerebrospinal fluid event, corresponding quantitative indicators are calculated. The quantitative indicators include the falling area, rising area, total exchange area, peak-trough swing, time series indicators, and exchange efficiency of the cerebrospinal fluid signal corresponding to the biphasic cerebrospinal fluid event. Based on the corresponding trough and rebound peak of the biphasic cerebrospinal fluid event, a corresponding reliability label is generated by combining the frame-by-frame displacement and the whole-brain signal change of each time frame. The frame-by-frame displacement is determined according to the rigid body motion correction parameters generated in the rigid body motion correction, and the whole-brain signal change is determined according to the signal intensity change of the functional magnetic resonance imaging data.

9. The method for detecting biphasic events in cerebrospinal fluid based on functional magnetic resonance imaging according to claim 8, characterized in that, The method for generating corresponding reliability labels based on the corresponding troughs and rebound peaks of biphasic cerebrospinal fluid events, combined with frame-by-frame shifts and whole-brain signal changes at each time frame, includes: The frame-by-frame displacement and whole-brain signal change of each time frame are compared with the corresponding preset displacement threshold and preset change threshold, respectively. Traverse all time frames, mark the time frames whose frame-by-frame displacement is less than or equal to a preset displacement threshold and whose whole-brain signal change is less than or equal to a preset change threshold as reliable time frames, and obtain the positions of the trough and rebound peak of the cerebrospinal fluid biphasic event within their respective preset time windows. A reliability label is generated when all time frames within the corresponding time window of the biphasic cerebrospinal fluid event are reliable, and the corresponding trough and rebound peak are not located at the window boundary within their respective preset time windows.

10. A cerebrospinal fluid biphasic event detection system based on functional magnetic resonance imaging (fMRI), used to execute the cerebrospinal fluid biphasic event detection method based on fMRI according to any one of claims 1 to 9, characterized in that, include: The data processing module is used to acquire and preprocess functional magnetic resonance imaging data of the brain. The signal extraction module is used to extract blood oxygenation level-dependent signal time series and cerebrospinal fluid signal time series from preprocessed functional magnetic resonance imaging data. The time alignment module is used to align the blood oxygen level dependent signal time series and the cerebrospinal fluid signal time series to the same physical time axis based on the inter-slice acquisition offset information of functional magnetic resonance imaging data. The event detection module is used to detect local signal peaks in the aligned blood oxygen level dependent signal time series, and the signal characteristic moment within a preset time window before each local signal peak is taken as the corresponding candidate event zero point. The biphasic localization module is used to locate the corresponding valley value within the first preset time window and the corresponding rebound peak within the second preset time window after the zero point of each candidate event in the aligned cerebrospinal fluid signal time series. The event filtering module is used to filter out biphasic cerebrospinal fluid events based on the corresponding valley and rebound peak of the location.

Images