Method for extracting mri image features for icu intracranial pressure dynamic assessment

Abstract

This invention relates to the field of intracranial pressure monitoring technology, specifically to an MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU. The method involves acquiring a flow velocity time series; segmenting the flow velocity time series into a systolic outflow sequence and a diastolic return sequence; constructing a reference baseline sequence based on the systolic outflow sequence; processing the diastolic return sequence with the reference baseline sequence to obtain a reference data sequence and an observation data sequence; constructing a coordinate graph with the reference data sequence as the horizontal axis and the observation data sequence as the vertical axis; using the flow velocity values ​​corresponding to the same sampling point in both sequences as coordinate points to form a flow velocity change trajectory; determining the amplitude deviation and waveform hysteresis based on the flow velocity change trajectory; calculating the return deficit index based on the amplitude deviation; calculating the return hysteresis index based on the waveform hysteresis; and fusing the return deficit index and the return hysteresis index to obtain a score that accurately reflects the cerebrospinal fluid dynamic state.

Description

Technical Field

[0001] This invention relates to the field of intracranial pressure monitoring technology, and more specifically to an MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU. Background Technology

[0002] Elevated intracranial pressure (ICP) is a common critical complication in ICU patients (such as those with traumatic brain injury, cerebral hemorrhage, or intracranial infection). If it is not assessed dynamically and promptly, it can easily lead to serious consequences such as brain herniation and insufficient cerebral perfusion, even endangering the patient's life. Moreover, especially in neurosurgical intensive care, dynamic monitoring of intracranial pressure is crucial for decisions regarding weaning patients undergoing external ventilator drainage.

[0003] Currently, clinical assessment of cerebrospinal fluid circulation capacity relies primarily on static intracranial pressure (PCP) values ​​obtained during clipping tests. However, PCP values ​​only reflect the current pressure balance and are insufficient to reveal the reserve of intracranial compensatory space. In some patients, although the PCP value remains within the normal range during clipping, their compensatory space may be exhausted, potentially leading to recurrence of hydrocephalus after extubation due to impaired drainage.

[0004] To obtain richer information on cerebrospinal fluid (CSF) flow velocity, current technologies often employ phase-contrast magnetic resonance imaging (MRI) to measure CSF flow velocity within the cerebral aqueduct. However, this method typically measures peak or average flow rate directly and compares it to population statistical averages. Due to differences in heart rate and aqueduct anatomy among patients, directly using population averages as a reference lacks specificity and makes it difficult to accurately assess the reflux status of individual patients. Furthermore, current techniques cannot effectively distinguish between two pathological mechanisms causing reflux obstruction when analyzing flow velocity waveforms: one is limited flow amplitude due to physical stenosis of the aqueduct, and the other is waveform phase lag caused by cerebral edema or decreased compliance. Since these two mechanisms require different clinical management strategies, current methods cannot decouple and identify them, limiting the role of imaging assessment in assisting weaning decisions and resulting in inaccurate dynamic information related to intracranial pressure. Summary of the Invention

[0005] To address the technical problems of existing MRI cerebrospinal fluid flow velocity analysis lacking specificity, failing to accurately assess reflux obstruction mechanisms, and having feature extraction results that are insufficient to precisely assist in weaning decisions for ICU patients undergoing external ventricular drainage, this invention provides an MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU. The specific technical solution adopted is as follows: This invention proposes an MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU, the method comprising: Magnetic resonance flow velocity data of the cerebral aqueduct region of the patient were acquired and a flow velocity time series was generated. The flow reversal time point in the flow velocity time series was identified, and the flow velocity time series was divided into systolic outflow sequence and diastolic reflux sequence with the flow reversal time point as the dividing point. A reference baseline sequence was constructed based on the systolic effluent sequence, and the diastolic reflux sequence was aligned and normalized with the reference baseline sequence to obtain a comparable reference data sequence and observation data sequence. Construct a coordinate graph with the reference data sequence as the horizontal axis and the observed data sequence as the vertical axis, and extract a reference line from the coordinate graph; use the velocity values ​​corresponding to the same sampling point in the reference data sequence and the observed data sequence as coordinate points, and connect them sequentially to form a velocity change trajectory; determine the amplitude deviation based on the distance of each point on the velocity change trajectory to the reference line; and determine the waveform hysteresis based on the velocity change trajectory. The reflux deficit index is calculated based on the amplitude deviation; the reflux hysteresis index is calculated based on the waveform hysteresis; the reflux deficit index and the reflux hysteresis index are fused to obtain a dynamic risk score, and the score is used as the result of MRI image feature extraction for dynamic assessment of intracranial pressure.

[0006] Furthermore, the flow velocity time series generation process includes: Phase-contrast magnetic resonance imaging was used to locate the scanning plane on a cross section perpendicular to the long axis of the cerebral aqueduct, and to collect flow velocity data of the cerebral aqueduct region of the patient during multiple cardiac cycles. ECG gating technology is used to synchronously acquire ECG signals, with the peak of the R wave in the ECG signal as the start time of the cardiac cycle; The flow velocity data from multiple cardiac cycles are aligned according to the cardiac cycle and then arithmetically averaged to obtain a flow velocity time series reflecting the changes in cerebrospinal fluid flow velocity within a cardiac cycle.

[0007] Furthermore, identifying the moment of flow direction reversal in the flow velocity time series includes: The moment corresponding to the maximum value of the forward velocity in the velocity time series is determined as the peak moment of the forward velocity. After the peak of the positive flow velocity, search for the first time point when the flow velocity value turns from positive to negative, and use this as the moment of flow reversal.

[0008] Furthermore, the reference baseline sequence is obtained by rearranging the data points in the systolic outflow sequence in the reverse chronological order; The process of aligning and normalizing the diastolic reflux sequence with the reference baseline sequence to obtain comparable reference data sequences and observation data sequences includes: The absolute value of each velocity value in the diastolic refluxing sequence is taken to obtain the positive diastolic sequence; The reference sequence and the positive diastolic sequence are phase-resampled to a preset number of sampling points respectively; The flow velocity values ​​at each sampling point in the reference sequence are squared to obtain the squared values; all squared values ​​are summed to obtain the sum; the square root of the sum is taken to obtain the signal energy value; the signal energy value is added to the preset minimum value to obtain the normalized reference value. The reference data sequence is obtained by dividing the flow velocity values ​​at each sampling point in the reference baseline sequence by the normalized baseline value; the observation data sequence is obtained by dividing the flow velocity values ​​at each sampling point in the positive diastolic sequence by the normalized baseline value.

[0009] Furthermore, the reference line is a straight line in the coordinate graph where the horizontal axis value is equal to the vertical axis value; The amplitude deviation determination process includes: For each coordinate point on the trajectory of the flow velocity change, obtain the x-coordinate and y-coordinate values ​​of the coordinate point; Calculate the absolute difference between the x-coordinate and y-coordinate values, and use this as the coordinate difference. Divide the coordinate difference by the square root of two to get the vertical distance from the coordinate point to the reference line. The vertical distance is used as the amplitude deviation of the sampling point corresponding to the coordinate point.

[0010] Furthermore, the waveform hysteresis determination process includes: Connect the last coordinate point of the velocity change trajectory with the origin to form the first closed line segment; connect the origin with the first coordinate point of the velocity change trajectory to form the second closed line segment. The closed figure is formed by the trajectory of the change in flow velocity, the first closed line segment, and the second closed line segment. The polygon area calculation method is used to calculate the area enclosed by the closed figure, which is used as the waveform hysteresis. In the case where the flow velocity change trajectory generates self-intersection and divides the closed figure into multiple sub-regions, the area enclosed by each sub-region is calculated separately, and the absolute values ​​are accumulated. The accumulated area is used as the waveform hysteresis.

[0011] Furthermore, the calculation of the reflux deficit index based on the amplitude deviation includes: For each sampling point, the flow velocity value of that sampling point in the reference data sequence is recorded as the reference flow velocity value; the flow velocity value of that sampling point in the observation data sequence is recorded as the observation flow velocity value. If the reference velocity value is greater than the observed velocity value, the difference between the reference velocity value and the observed velocity value is calculated as the single-point deficit; if the reference velocity value is not greater than the observed velocity value, the single-point deficit is set to zero. Multiply the preset first constant by the amplitude deviation to obtain the product value; add one to the product value to obtain the weighting coefficient; multiply the single-point deficit by the weighting coefficient to obtain the weighted deficit. The total reference flow rate value is obtained by summing up all the reference flow rate values ​​in the reference data sequence. The weighted deficit of all sampling points is summed to obtain the cumulative deficit; the sum of the total reference flow rate and the preset minimum value is calculated as the total flow rate; the cumulative deficit is divided by the total flow rate to obtain the reflux deficit index.

[0012] Furthermore, the calculation of the return hysteresis index based on the waveform hysteresis includes: The centroid calculation method is used to calculate the centroid positions of the reference data sequence and the observed data sequence, respectively. The centroid position difference is obtained by subtracting the centroid position of the reference data sequence from the centroid position of the observed data sequence. If the centroid position difference is positive, it is used as the centroid lag. If the centroid position difference is not greater than zero, the centroid lag is set to zero. Multiply the preset second constant by the waveform hysteresis to obtain the product result; then add one to the product result to obtain the correction coefficient; The product of the center of gravity lag and the correction coefficient is calculated as the reflux hysteresis index.

[0013] Furthermore, the fusion of the reflux deficit index and the reflux hysteresis index yields a dynamic risk score, including: Based on preset weighting coefficients, a weighted summation is performed on the return deficit indicator and the return lag indicator to obtain a weighted sum value. Compare the weighted sum with the preset upper limit; if the weighted sum is less than or equal to the preset upper limit, retain the weighted sum; if the weighted sum is greater than the preset upper limit, truncate the weighted sum to the preset upper limit. The weighted sum is multiplied by a preset coefficient to obtain the dynamic risk score.

[0014] An MRI image feature extraction system for dynamic assessment of intracranial pressure in the ICU, the system comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU described above.

[0015] The present invention has the following beneficial effects: The present invention uses the patient's own systolic outflow waveform to construct a reference baseline sequence, rather than relying on the statistical average of the population. By aligning and normalizing the diastolic reflux sequence with its own systolic reference sequence, the assessment benchmark bias caused by differences in aqueduct diameter, heart rate, and cardiac output among patients is effectively eliminated, improving the specificity and accuracy of the assessment results. By constructing a coordinate graph with the reference data sequence as the horizontal axis and the observed data sequence as the vertical axis, the time-domain waveform problem is transformed into a geometric problem. Based on this, independent amplitude deviation and waveform hysteresis are extracted, which point to different pathological mechanisms: amplitude deviation reflects the patency of the cerebrospinal fluid reflux channel, and waveform hysteresis reflects the compliance status of brain tissue, so as to effectively distinguish between physical stenosis of the aqueduct and decreased brain tissue compliance. By fusing the reflux deficit index and the reflux hysteresis index to obtain a dynamic risk score, multi-dimensional flow velocity characteristics are integrated to comprehensively reflect the degree of intracranial pressure-related cerebrospinal fluid dynamic abnormalities. Compared with existing single-parameter assessment methods, this significantly improves the correlation between MRI image features and intracranial pressure changes, making the extracted feature results more accurate for dynamic assessment of intracranial pressure. Attached Figure Description

[0016] Figure 1 The flowchart illustrates an MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU, as provided by this invention. Detailed Implementation

[0017] like Figure 1 As shown, an embodiment of the present invention provides an MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU, comprising the following steps: S101: Acquire magnetic resonance flow velocity data of the cerebral aqueduct region of the patient and generate a flow velocity time series; identify the flow reversal moment in the flow velocity time series, and use the flow reversal moment as the dividing point to divide the flow velocity time series into a systolic outflow sequence and a diastolic reflux sequence.

[0018] It's important to understand that the pulsatile flow of cerebrospinal fluid (CSF) within the cerebral aqueduct is influenced by two key factors: the physical patency of the aqueduct itself and the elastic compliance of the cranial cavity and brain tissue. If the aqueduct experiences physical narrowing or obstruction, the CSF return pathway is mechanically restricted, resulting in a significant decrease in the amplitude of the return flow velocity—a "channel-restricted" disorder. Conversely, if brain tissue compliance decreases due to edema or swelling, the elastic recoil capacity of the cranial cavity weakens, leading to a time delay in the CSF return response relative to the driving input—an overall waveform lag—a "damped-enhanced" disorder. Both pathological mechanisms can cause insufficient CSF return, but their physical nature differs: the former results in an amplitude deficit due to channel truncation, while the latter results in a phase lag due to viscous damping.

[0019] This embodiment provides an MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU. This method is primarily applied to patients with external ventricular drainage who are scheduled for weaning assessment. During the stable period before weaning decision, patients can be transferred to the MRI room for a single scan to obtain cerebrospinal fluid flow velocity data. Based on this method, the intracranial compensatory space status is quantitatively assessed, providing clinicians with objective auxiliary decision-making information.

[0020] In this embodiment, phase-contrast magnetic resonance imaging (PCI) is used to position the scanning plane on a cross-section perpendicular to the long axis of the cerebral aqueduct, and to collect flow velocity data of the cerebral aqueduct region of the patient within multiple cardiac cycles. ECG gating technology is used to simultaneously acquire ECG signals, with the R-wave peak of the ECG signal as the start time of the cardiac cycle. The flow velocity data of multiple cardiac cycles are aligned according to the cardiac cycle and an arithmetic mean is calculated to obtain a flow velocity time series reflecting the changes in cerebrospinal fluid flow velocity within a cardiac cycle.

[0021] Because the cerebral aqueduct is a narrow passage connecting the third and fourth ventricles, the pulsatile flow of cerebrospinal fluid is most pronounced here, sensitively reflecting dynamic changes in intracranial pressure. Therefore, the cerebral aqueduct was selected as the region of interest for collecting flow velocity data. Specifically, phase-contrast magnetic resonance imaging (PCI) was used to acquire flow velocity data in the patient's cerebral aqueduct region. PCI can quantitatively measure the velocity of flowing protons. By setting a flow velocity encoding threshold, the flow velocity change data of the aqueduct cross-section within one cardiac cycle can be obtained. This technique is a publicly disclosed method in the art and can be directly applied in this embodiment, so it will not be described in detail here.

[0022] It is understandable that during the data acquisition process, ECG gating technology is used to synchronously record the patient's ECG signal, with the R-wave peak of the ECG signal serving as the start of the cardiac cycle to ensure that the flow rate data is synchronized with the heart rhythm.

[0023] To eliminate random errors from a single measurement, flow velocity data from multiple cardiac cycles were collected. After aligning each cardiac cycle according to the R-wave apex, the flow velocity values ​​at the same time point were calculated using an arithmetic mean to generate a representative average cardiac cycle flow velocity time series, i.e., the flow velocity time series.

[0024] Since the raw flow velocity data may contain high-frequency noise, the flow velocity time series needs to be smoothed and filtered to remove noise interference. Specifically, smoothing filtering can be achieved using low-pass filtering or moving average filtering, which are common existing technologies and will not be elaborated here.

[0025] It's important to understand that under the periodic pumping action of the heart, cerebrospinal fluid (CSF) exhibits a reciprocating flow characteristic within the cerebral aqueduct: During cardiac contraction, cerebral arteries dilate and fill, compressing the ventricular system, and CSF flows from the third ventricle through the aqueduct to the fourth ventricle; at this time, the flow velocity is positive, known as positive flow. During cardiac diastole, cerebral arteries recoil, and the cranial cavity elastically shrinks, causing CSF to flow back from the fourth ventricle through the aqueduct to the third ventricle; at this time, the flow velocity is negative, known as negative flow. Therefore, there is a critical moment when CSF transitions from positive to negative flow.

[0026] To accurately identify the moment of flow reversal in a flow velocity time series, as an example, the moment corresponding to the maximum value of the positive flow velocity in the flow velocity time series is determined as the peak moment of the positive flow velocity; after the peak moment of the positive flow velocity, the first time point at which the flow velocity value turns from positive to negative is searched and taken as the moment of flow reversal.

[0027] In a flow velocity time series, the positive velocity portion corresponds to the cardiac systole period, during which the flow velocity value gradually rises from zero to a peak and then gradually decreases. Therefore, the maximum flow velocity is typically searched within the cardiac systole period, and the moment corresponding to the maximum flow velocity is taken as the peak moment of the positive velocity.

[0028] Since the minor fluctuations in the early systolic phase are filtered out by the peak value, the point where the velocity changes from positive to negative after the peak of the positive flow velocity accurately reflects the physiological boundary between the end of systole and the beginning of diastole. Therefore, the first time point after the peak of the positive flow velocity where the velocity value changes from positive to negative will be searched as the flow reversal point. It should be noted that, in extreme cases, if there is no time point after the peak of the positive flow velocity where the velocity value changes from positive to negative, the time point with the minimum velocity value after the peak of the positive flow velocity can be obtained and this time point will be used as the flow reversal point.

[0029] In this embodiment, the moment of flow reversal is used as the dividing line. The part of the flow velocity time series before the moment of flow reversal is taken as the systolic outflow sequence, and the part after the moment of flow reversal is taken as the diastolic return sequence.

[0030] The systolic efflux sequence is the flow velocity data from the start of the cardiac cycle to the moment of flow reversal; the diastolic reflux sequence is the flow velocity data from the moment of flow reversal to the end of the cardiac cycle.

[0031] S102: Construct a reference baseline sequence based on the systolic effluent sequence, and align and normalize the diastolic reflux sequence with the reference baseline sequence to obtain a comparable reference data sequence and observation data sequence.

[0032] It is important to understand that since the systolic outflow sequence records the process of cerebrospinal fluid flowing out of the cranial cavity under the drive of cardiac systole, and the diastolic refluxing sequence records the process of cerebrospinal fluid flowing back into the cranial cavity under the elastic recoil of cardiac diastole, physiologically, there is an intrinsic correlation between systolic outflow and diastolic refluxing. That is, during systole, the dilation of cerebral arteries "push" cerebrospinal fluid out of the cranial cavity, and during diastole, the recoil of cerebral arteries "absorbs" cerebrospinal fluid back into the cranial cavity. In order to extract the asymmetric features between the systolic and diastolic waveforms, the systolic outflow sequence can be time-flipped to obtain a reference sequence for comparison with the diastolic refluxing sequence.

[0033] It should be noted that the reference baseline sequence is obtained by rearranging the data points in the condensation outflow sequence in reverse chronological order. Specifically, the first data point in the original sequence is moved to the last position, the last data point is moved to the first position, and so on, to form the reference baseline sequence.

[0034] It's important to understand that the different durations of systole and diastole lead to inconsistencies in the number of raw data points between the reference sequence and the diastolic refluxing sequence, making direct point-to-point numerical comparisons impossible. Furthermore, variations in aqueduct diameter, heart rate, and cardiac output among patients result in a lack of cross-patient comparability of the absolute values ​​of flow velocity amplitude. Therefore, it is necessary to perform temporal alignment and energy normalization on both the diastolic refluxing sequence and the reference sequence. This involves resampling to map both sequences onto the same temporal axis to eliminate duration differences, and energy normalization scaling the two sequences uniformly based on the signal energy of the reference sequence to eliminate individual variations.

[0035] In this embodiment, the absolute values ​​of each velocity value in the diastolic refluxing sequence are taken to obtain the positive diastolic sequence; the reference sequence and the positive diastolic sequence are phase-resampled to a preset number of sampling points; the velocity values ​​of each sampling point in the reference sequence are squared to obtain squared values; all squared values ​​are summed to obtain a summed result; the square root of the summed result is performed to obtain the signal energy value; the signal energy value is added to a preset minimum value to obtain a normalized reference value; the velocity values ​​of each sampling point in the reference sequence are divided by the normalized reference value to obtain the reference data sequence; the velocity values ​​of each sampling point in the positive diastolic sequence are divided by the normalized reference value to obtain the observation data sequence.

[0036] Since the original diastolic reflux sequence has a negative velocity value, representing the direction of cerebrospinal fluid reflux, a positive diastolic sequence is also needed for comparison with a positive reference sequence.

[0037] It should be noted that time alignment is achieved using resampling technology. Specifically, the reference sequence and the forward diastolic sequence are mapped onto a unified time axis using interpolation, ensuring that both have the same number of data points. Linear interpolation or spline interpolation can be used, which are conventional techniques in signal processing and can be directly applied in this embodiment.

[0038] It is understandable that after time-phase alignment processing, the reference data sequence and the observation data sequence have the same number of sampling points, with each sampling point corresponding to a time phase.

[0039] It should be noted that the specific value of the preset quantity can be determined according to actual needs, and this embodiment does not impose a specific limitation. For example, the preset quantity is preferably 100.

[0040] It is understandable that the resampled reference sequence and the positive diastolic sequence have the same sequence length, i.e., a preset number of sampling points.

[0041] The signal energy value represents the total driving energy outflow during the contraction phase.

[0042] It should be noted that the specific value of the preset minimum involved in this embodiment is determined based on engineering experience, and this embodiment does not impose any specific limitations. For example, the preset minimum value is 10 to the power of negative six, and the preset minimum value has the same dimensions as the corresponding denominator.

[0043] S103: Construct a coordinate graph with the reference data sequence as the horizontal axis and the observed data sequence as the vertical axis, and extract a reference line on the coordinate graph; use the flow velocity values ​​corresponding to the same sampling point in the reference data sequence and the observed data sequence as coordinate points, and connect them sequentially to form a flow velocity change trajectory; determine the amplitude deviation based on the distance of each point on the flow velocity change trajectory to the reference line; determine the waveform hysteresis based on the flow velocity change trajectory.

[0044] As a preferred implementation, the reference line is a straight line in the coordinate graph where the horizontal axis value is equal to the vertical axis value.

[0045] The reference line represents the ideal state in which the flow velocity at a certain sampling point in the observed data sequence is equal to the flow velocity at the same sampling point in the reference data sequence. That is, when the actual return flow completely follows the theoretical expectation, the coordinate point falls on the reference line.

[0046] For example, a two-dimensional coordinate graph is constructed with the flow velocity values ​​of the reference data sequence on the horizontal axis and the flow velocity values ​​of the observed data sequence on the vertical axis. For each sampling point, the flow velocity value of that sampling point in the reference data sequence is used as the horizontal axis, and the flow velocity value in the observed data sequence is used as the vertical axis, resulting in a coordinate point. Since the reference data sequence and the observed data sequence have the same number of sampling points, a total of coordinate points are obtained, equal to the number of sampling points (and the preset number). Finally, according to the time sequence of the sampling points, from the first sampling point to the last sampling point, the coordinate points are connected sequentially to form a continuous trajectory line, i.e., the flow velocity change trajectory.

[0047] The velocity change trajectory reflects the dynamic relationship between the actual return velocity and the theoretical expected velocity within a complete cardiac cycle: if the actual return completely follows the theoretical expectation, the coordinate point falls on the diagonal line (i.e., the reference line) where the horizontal axis equals the vertical axis; if the actual return exhibits amplitude deficit or phase lag, the coordinate point will deviate from the diagonal line, forming a specific trajectory shape.

[0048] It can be understood that there is a one-to-one correspondence between sampling points, coordinate points, and time phases, that is, one time phase corresponds to one sampling point, one sampling point corresponds to one coordinate point, and the order of coordinate points is consistent with the time order of sampling points.

[0049] The process of determining the amplitude deviation includes the following steps: 1) For each coordinate point on the velocity change trajectory, obtain the x-coordinate and y-coordinate values ​​of the coordinate point.

[0050] 2) Calculate the absolute difference between the x-coordinate and y-coordinate values, and use it as the coordinate difference.

[0051] Observed flow velocity value: the flow velocity value at a certain sampling point in the observed data sequence; Reference flow velocity value: the flow velocity value at the same sampling point in the reference data sequence.

[0052] The coordinate difference reflects the absolute deviation between the observed flow velocity value and the reference flow velocity value at a certain coordinate point, that is, it reflects the absolute deviation between the observed flow velocity value and the reference flow velocity value at a certain time phase.

[0053] 3) Divide the coordinate difference by the square root of two to obtain the vertical distance from the coordinate point to the reference line.

[0054] Since the perpendicular distance from a coordinate point to the reference line is not simply the difference between the horizontal and vertical coordinates, it needs to be converted into a geometric distance. Therefore, according to the formula for the distance from a point to a line, when the reference line is a line with the horizontal axis equal to the vertical axis, the perpendicular distance from the point to that line is equal to the absolute value of the coordinate difference divided by the square root of two. Specifically, dividing by the square root of two converts the coordinate difference into a standard geometric distance, giving the magnitude deviation a clear geometric meaning.

[0055] 4) Use the vertical distance as the amplitude deviation of the sampling point corresponding to the coordinate point.

[0056] Amplitude deviation reflects the degree of deviation of the observed flow velocity value at a certain sampling point from the reference flow velocity value. The larger the amplitude deviation, the greater the difference between the actual return flow velocity and the theoretical expected flow velocity.

[0057] It's important to understand that in the velocity comparison graph, the horizontal axis represents the theoretically expected reference velocity value, and the vertical axis represents the actual measured observed velocity value. If the actual return flow perfectly follows the theoretical expectation, the observed velocity value equals the reference velocity value, and the coordinate point falls exactly on the reference line where the horizontal axis equals the vertical axis. If the actual return flow velocity is lower than the theoretical expectation, the coordinate point shifts towards the horizontal axis, deviating from the reference line. The vertical distance from the coordinate point to the reference line is a geometric measure of the degree of deviation; the larger this distance, the greater the difference between the actual velocity and the theoretical expectation. Since this distance considers deviations in both the horizontal and vertical axes, it comprehensively reflects the degree of amplitude deviation at that moment, and is therefore used as the amplitude deviation rate.

[0058] The process of determining waveform hysteresis includes the following steps: 1) Connect the last coordinate point of the velocity change trajectory with the origin to form the first closed line segment; connect the origin with the first coordinate point of the velocity change trajectory to form the second closed line segment.

[0059] Since the velocity change trajectory is an open line extending from the first coordinate point to the last, its area cannot be directly calculated. Therefore, to obtain an area index that reflects the waveform's hysteresis characteristics, the velocity change trajectory is converted into a closed graph.

[0060] 2) The closed figure is formed by the trajectory of the change in flow velocity, the first closed line segment, and the second closed line segment.

[0061] From a geometric perspective, if two waveforms are perfectly synchronized, the coordinate points move back and forth along the reference line, the rising path coincides with the falling path, and the closed figure has no enclosed area. However, if there is a phase lag, the rising path and falling path no longer coincide, forming a gap between them, enclosing a ring-shaped region, i.e., a closed figure. Therefore, the area of ​​this ring-shaped region intuitively reflects the degree of asynchrony between the two waveforms.

[0062] 3) The polygon area calculation method is used to calculate the area enclosed by the closed figure and use it as the waveform hysteresis. In the case where the flow velocity change trajectory generates self-intersection and divides the closed figure into multiple sub-regions, the area enclosed by each sub-region is calculated separately, and the absolute value is accumulated. The accumulated area is used as the waveform hysteresis.

[0063] It should be noted that a closed figure is a polygon formed by connecting multiple vertices in sequence, and its area can be obtained using polygon area calculation methods. Polygon area calculation methods divide the polygon into several triangles and obtain the area through cross-calculation of the coordinate values ​​of each vertex. This is a common technique in computational geometry and will not be elaborated upon here.

[0064] Under normal physiological conditions, the trajectory of flow velocity changes is usually a simple loop that does not self-intersect. However, under certain pathological conditions or noise interference, the trajectory may self-intersect (e.g., the shape intersects). In this case, the closed shape is no longer a simple polygon but is composed of multiple sub-regions. Therefore, if the flow velocity change trajectory self-intersects, causing the closed shape to split into multiple sub-regions, the area calculation of the simple polygon may only yield an algebraic sum of the areas of each sub-region (with positive and negative values ​​canceling each other out), which cannot accurately reflect the actual waveform hysteresis. In this case, the area enclosed by each sub-region can be calculated separately, and the absolute values ​​of each area can be summed. The total summed area can then be used as the waveform hysteresis.

[0065] Waveform hysteresis reflects the overall lag of the observed waveform relative to the reference waveform. The greater the waveform hysteresis, the more significant the phase lag of the actual return waveform.

[0066] It's important to understand that in a coordinate graph, if the actual return waveform perfectly follows the reference waveform (i.e., the outflow waveform) without phase lag, the velocity change trajectory closely follows the reference line. The trajectory is a line segment that travels back and forth along the reference line, and the area of ​​the closed shape is close to zero. However, if the actual return waveform exhibits phase lag, the change in the observed velocity value lags behind the change in the reference velocity value. This causes the coordinate point to follow different paths in the rising and falling segments, forming a loop trajectory. After closing, this loop trajectory encloses an area of ​​a certain size. The size of this area is positively correlated with the degree of lag; that is, the more severe the lag, the wider the loop trajectory and the larger the area of ​​the closed shape.

[0067] S104: The reflux deficit index is calculated based on the amplitude deviation; the reflux hysteresis index is calculated based on the waveform hysteresis; the reflux deficit index and the reflux hysteresis index are fused to obtain the dynamic risk score, and the score is used as the result of MRI image feature extraction for dynamic assessment of intracranial pressure.

[0068] It is important to understand that amplitude deviation only reflects the instantaneous deviation at a single sampling point and cannot comprehensively assess the overall reflux deficit throughout the entire cardiac cycle. Furthermore, small instantaneous deviations may be physiological fluctuations, while large, sustained deviations are more likely to indicate channel obstruction. Therefore, it is necessary to weight and accumulate the amplitude deviations from each sampling point and normalize them into a single indicator to obtain a reflux deficit indicator that reflects the overall patency of the cerebrospinal fluid reflux channel.

[0069] To comprehensively obtain the reflux deficit index, as an example, for each sampling point, the velocity value of that sampling point in the reference data sequence is recorded as the reference velocity value; the velocity value of that sampling point in the observed data sequence is recorded as the observed velocity value; if the reference velocity value is greater than the observed velocity value, the difference between the reference velocity value and the observed velocity value is calculated as the single-point deficit; if the reference velocity value is not greater than the observed velocity value, the single-point deficit is set to zero; a preset first constant is multiplied by the amplitude deviation to obtain the product value; one is added to the product value to obtain the weighting coefficient; the single-point deficit is multiplied by the weighting coefficient to obtain the weighted deficit; all reference velocity values ​​in the reference data sequence are summed to obtain the total reference velocity value; the weighted deficit of all sampling points is summed to obtain the accumulated deficit; the sum of the total reference velocity value and the preset minimum value is calculated as the total velocity value; the accumulated deficit is divided by the total velocity value to obtain the reflux deficit index.

[0070] If the reference velocity value at a certain sampling point (i.e., a certain time phase) is greater than the observed velocity value, it indicates that the actual return velocity at that time phase is lower than the theoretical expectation, and there is a deficit. In this case, the difference between the reference velocity value and the observed velocity value is directly used as the single-point deficit. Conversely, if the reference velocity value is not greater than the observed velocity value, it indicates that there is no deficit, and the single-point deficit of that sampling point is directly set to zero.

[0071] Deficit refers to the phenomenon where the actual reflux velocity is lower than the theoretically expected velocity. Specifically, for any sampling point, if the observed velocity value (representing the actual reflux velocity) in the observed data sequence is less than the reference velocity value (representing the theoretically expected velocity) in the reference data sequence, the difference between the two is the deficit at that sampling point. Deficit reflects the energy dissipation or channel interruption that occurs during the cerebrospinal fluid reflux process. The larger the deficit at a certain sampling point (i.e., at a certain time phase), the more severe the actual reflux loss relative to the theoretical expectation at that time phase, suggesting a greater likelihood of physical stenosis, obstruction, or other pathological changes hindering reflux.

[0072] It should be noted that the specific value of the preset first constant can be obtained through calibration using historical case data, and those skilled in the art can adjust this constant according to the actual application scenario. Specifically, flow velocity data from patients diagnosed with aqueductal stenosis are collected, and the distribution characteristics of their amplitude deviation are statistically analyzed. This ensures that if the amplitude deviation is in the low range of normal fluctuation, the weighting coefficient is close to 1; if the amplitude deviation enters the high range of significant deviation, the weighting coefficient increases significantly, thereby amplifying the deficit of significant deviations (i.e., a sampling point with a large single-point deficit). For example, if the preset first constant is 2.0, and the amplitude deviation is 0.5, then the weighting coefficient is equal to 2.0, meaning the corresponding single-point deficit is amplified by one time.

[0073] Since small, instantaneous deviations may be physiological fluctuations or measurement noise, while large, sustained deviations are more likely to indicate pathological channel obstruction, they need to be given higher weights to enhance sensitivity to pathological features. Therefore, weighting coefficients are used to adjust the contribution of different amplitude deviations to the deficit; that is, if the amplitude deviation is smaller, the weighting coefficient is close to 1, and the single-point deficit is counted at its original value; if the amplitude deviation is larger, the weighting coefficient is greater than 1, and the single-point deficit is amplified and included.

[0074] The total reference velocity value represents the theoretically expected total return energy.

[0075] The reflux deficit index reflects the patency of the cerebrospinal fluid reflux pathway; the higher the index value, the more severe the obstruction or narrowing of the reflux pathway.

[0076] It is important to understand that waveform hysteresis measures the overall lag of a waveform only from a geometric area perspective, without considering the specific offset of the waveform on the time axis. Furthermore, a simple increase in waveform area can be caused by multiple factors. Therefore, based on the centroid positions of the observed and reference data sequences, the centroid hysteresis is determined. Then, the centroid hysteresis is combined with the waveform hysteresis, and the waveform hysteresis is used to correct the centroid hysteresis, resulting in a reflux hysteresis index. This allows for a more accurate assessment of the impact of decreased brain tissue compliance or increased viscous damping on cerebrospinal fluid reflux.

[0077] The calculation process for the reflux hysteresis index includes the following steps: 1) The centroid calculation method is used to calculate the centroid positions of the reference data sequence and the observed data sequence respectively.

[0078] It should be noted that the centroid position is calculated using a weighted average method. The weighted average method involves: summing the products of the indices of each sampling point and the absolute values ​​of their corresponding flow velocities as the numerator; and calculating the denominator as the sum of the absolute values ​​of all flow velocities plus a preset minimum value. The ratio of the numerator to the denominator is then calculated. The result of the weighted average method is the centroid position. This centroid calculation method is a standard technique in signal processing and will not be elaborated upon here.

[0079] The position of the center of gravity reflects the energy center of the flow velocity waveform on the phase axis.

[0080] 2) Subtract the centroid position of the reference data sequence from the centroid position of the observed data sequence to obtain the centroid position difference; if the centroid position difference is positive, use the centroid position difference as the centroid lag; if the centroid position difference is not greater than zero, set the centroid lag to zero.

[0081] The difference in the center of gravity position reflects the direction and extent of the deviation of the observed waveform relative to the reference waveform on the time axis.

[0082] The reference waveform refers to the theoretical reflux waveform obtained by time reversal of the systolic outflow sequence, representing the expected morphology of cerebrospinal fluid reflux in an ideal elastic system; the observed waveform refers to the actual reflux waveform obtained by taking the absolute value of the diastolic reflux sequence, representing the actual cerebrospinal fluid reflux morphology in the patient.

[0083] If the difference in the center of gravity position is positive, it means that the center of gravity of the observed waveform lags behind the reference waveform, and the difference in the center of gravity position is directly used as the lag value; if the difference in the center of gravity position is not greater than zero, it means that the observed waveform has no lag or leads the reference waveform, and the lag value is set to zero.

[0084] 3) Multiply the preset second constant by the waveform hysteresis to obtain the product result; then add one to the product result to obtain the correction coefficient.

[0085] It should be noted that the specific value of the preset second constant can be set according to the typical range of waveform hysteresis, so that when the waveform hysteresis is small, the correction coefficient is close to 1, and when the waveform hysteresis increases significantly, the correction coefficient increases accordingly, thereby appropriately enhancing the correction of the centroid hysteresis. The preset second constant is used to control the amplification sensitivity, and its specific value can be adjusted by those skilled in the art according to the actual application scenario. For example, the preset second constant can be set to 5.

[0086] It is important to understand that if the waveform lag is smaller, the correction coefficient is closer to 1, which means that the waveform lag is less significant and the centroid lag is counted at its original value; if the waveform lag is larger, the correction coefficient is larger, which means that the waveform lag is more significant and the centroid lag is amplified to a greater extent.

[0087] 4) Calculate the product of the center of gravity lag and the correction coefficient, and use it as the reflux lag index.

[0088] The reflux hysteresis index reflects the degree to which brain tissue compliance affects cerebrospinal fluid reflux. A larger reflux hysteresis index indicates a more significant lag in the observed waveform relative to the reference waveform, suggesting a more severe degree of brain tissue edema or decreased compliance.

[0089] To accurately obtain the dynamic risk score, as an example, the backflow deficit index and backflow hysteresis index are weighted and summed according to preset weight coefficients to obtain a weighted sum value; the weighted sum value is compared with a preset upper limit value; if the weighted sum value is less than or equal to the preset upper limit value, the weighted sum value is retained; if the weighted sum value is greater than the preset upper limit value, the weighted sum value is truncated to the preset upper limit value; the weighted sum value is multiplied by a preset coefficient to obtain the dynamic risk score.

[0090] It should be noted that the preset weight coefficient includes a first preset weight and a second preset weight, wherein the sum of the first preset weight and the second preset weight is one.

[0091] The specific values ​​of the preset weighting coefficients can be determined based on the statistical results of clinical data. These coefficients reflect the different levels of importance of the reflux deficit and reflux retardation indicators in intracranial pressure assessment. Specifically, the weighting percentages of the reflux deficit and reflux retardation indicators can be set according to the correlation strength between these two indicators and intracranial pressure abnormalities under different pathological states. This embodiment does not impose specific limitations. For example, the first preset weighting value is 0.6, and the second preset weighting value is 0.4.

[0092] The weighted sum is obtained as follows: multiply the first preset weight by the return deficit indicator to obtain the first weighted value; multiply the second preset weight by the return lag indicator to obtain the second weighted value; add the first weighted value and the second weighted value to obtain the weighted sum.

[0093] The weighted sum reflects the overall degree of abnormality in the dynamics of cerebrospinal fluid refluxing in patients. The larger the weighted sum, the more severe the deficit and stagnation of cerebrospinal fluid refluxing, and the higher the risk of increased intracranial pressure and related complications. The smaller the weighted sum, the closer the dynamics of cerebrospinal fluid refluxing are to the normal state, and the lower the risk of intracranial pressure.

[0094] Because severe abnormalities in a single indicator (such as the reflux deficit indicator) can lead to an excessively high weighted sum, exceeding the reasonable range for clinical assessment and affecting the reference value of the score, it is necessary to set a preset upper limit value to avoid extreme outliers interfering with the assessment results.

[0095] It should be noted that the specific value of the preset upper limit can be determined based on clinical trials, and this embodiment does not impose a specific limitation. Specifically, sufficient data on reflux deficit and reflux retardation indicators are collected from ICU patients with normal and abnormal intracranial pressure. The distribution range of their weighted sum values ​​is statistically analyzed, and the preset upper limit is determined in conjunction with clinical experience to ensure that the preset upper limit can cover the normal and mildly abnormal conditions of the vast majority of patients. For example, the preset upper limit can be set to 1.

[0096] It should be noted that the preset coefficient is used to standardize the weighted sum value after truncation by a preset upper limit to a range that is easy for clinical medical staff to judge and compare, so as to facilitate rapid assessment of the patient's condition. Its specific value can be adjusted according to actual needs. For example, the preset coefficient value is 100.

[0097] The dynamic risk score is used to quantitatively assess the dynamic state of cerebrospinal fluid; the higher the score, the greater the risk of reflux obstruction.

[0098] It is important to understand that the dynamic risk score, as a result of MRI image feature extraction, corresponds to the degree of intracranial pressure-related cerebrospinal fluid dynamic abnormalities. It is used by medical staff to quickly assess the dynamic changes in intracranial pressure and the compensatory reserve of patients, and to assist in making decisions on weaning patients undergoing external ventricular drainage.

[0099] This embodiment also provides an MRI image feature extraction system for dynamic assessment of intracranial pressure in the ICU. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU described above.

Claims

1. A method for extracting MRI image features for dynamic assessment of intracranial pressure in the ICU, characterized in that, include: Magnetic resonance flow velocity data of the cerebral aqueduct region of the patient were acquired and a flow velocity time series was generated. The flow reversal time point in the flow velocity time series was identified, and the flow velocity time series was divided into systolic outflow sequence and diastolic reflux sequence with the flow reversal time point as the dividing point. A reference baseline sequence was constructed based on the systolic effluent sequence, and the diastolic reflux sequence was aligned and normalized with the reference baseline sequence to obtain a comparable reference data sequence and observation data sequence. Construct a coordinate graph with the reference data sequence as the horizontal axis and the observed data sequence as the vertical axis, and extract a reference line from the coordinate graph; use the velocity values ​​corresponding to the same sampling point in the reference data sequence and the observed data sequence as coordinate points, and connect them sequentially to form a velocity change trajectory; determine the amplitude deviation based on the distance of each point on the velocity change trajectory to the reference line; and determine the waveform hysteresis based on the velocity change trajectory. The reflux deficit index is calculated based on the amplitude deviation; the reflux hysteresis index is calculated based on the waveform hysteresis; the reflux deficit index and the reflux hysteresis index are fused to obtain a dynamic risk score, and the score is used as the result of MRI image feature extraction for dynamic assessment of intracranial pressure.

2. The MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU according to claim 1, characterized in that, The flow velocity time series generation process includes: Phase-contrast magnetic resonance imaging was used to locate the scanning plane on a cross section perpendicular to the long axis of the cerebral aqueduct, and to collect flow velocity data of the cerebral aqueduct region of the patient during multiple cardiac cycles. ECG gating technology is used to synchronously acquire ECG signals, with the peak of the R wave in the ECG signal as the start time of the cardiac cycle; The flow velocity data from multiple cardiac cycles are aligned according to the cardiac cycle and then arithmetically averaged to obtain a flow velocity time series reflecting the changes in cerebrospinal fluid flow velocity within a cardiac cycle.

3. The MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU according to claim 1, characterized in that, The identification of the flow direction reversal moment in the flow velocity time series includes: The moment corresponding to the maximum value of the forward velocity in the velocity time series is determined as the peak moment of the forward velocity. After the peak of the positive flow velocity, search for the first time point when the flow velocity value turns from positive to negative, and use this as the moment of flow reversal.

4. The MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU according to claim 1, characterized in that, The reference baseline sequence is obtained by rearranging the data points in the systolic outflow sequence in the reverse chronological order; The process of aligning and normalizing the diastolic reflux sequence with the reference baseline sequence to obtain comparable reference data sequences and observation data sequences includes: The absolute value of each velocity value in the diastolic refluxing sequence is taken to obtain the positive diastolic sequence; The reference sequence and the positive diastolic sequence are phase-resampled to a preset number of sampling points respectively; The flow velocity values ​​at each sampling point in the reference sequence are squared to obtain the squared values; all squared values ​​are summed to obtain the sum; the square root of the sum is taken to obtain the signal energy value; the signal energy value is added to the preset minimum value to obtain the normalized reference value. The reference data sequence is obtained by dividing the flow velocity values ​​at each sampling point in the reference baseline sequence by the normalized baseline value; the observation data sequence is obtained by dividing the flow velocity values ​​at each sampling point in the positive diastolic sequence by the normalized baseline value.

5. The MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU according to claim 1, characterized in that, The reference line is a straight line in the coordinate graph where the horizontal axis value is equal to the vertical axis value; The amplitude deviation determination process includes: For each coordinate point on the trajectory of the flow velocity change, obtain the x-coordinate and y-coordinate values ​​of the coordinate point; Calculate the absolute difference between the x-coordinate and y-coordinate values, and use this as the coordinate difference. Divide the coordinate difference by the square root of two to get the vertical distance from the coordinate point to the reference line. The vertical distance is used as the amplitude deviation of the sampling point corresponding to the coordinate point.

6. The MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU according to claim 1, characterized in that, The waveform hysteresis determination process includes: Connect the last coordinate point of the velocity change trajectory with the origin to form the first closed line segment; connect the origin with the first coordinate point of the velocity change trajectory to form the second closed line segment. The closed figure is formed by the trajectory of the change in flow velocity, the first closed line segment, and the second closed line segment. The polygon area calculation method is used to calculate the area enclosed by the closed figure, which is used as the waveform hysteresis. In the case where the flow velocity change trajectory generates self-intersection and divides the closed figure into multiple sub-regions, the area enclosed by each sub-region is calculated separately, and the absolute values ​​are accumulated. The accumulated area is used as the waveform hysteresis.

7. The MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU according to claim 5, characterized in that, The calculation of the reflux deficit index based on the amplitude deviation includes: For each sampling point, the flow velocity value of that sampling point in the reference data sequence is recorded as the reference flow velocity value; the flow velocity value of that sampling point in the observation data sequence is recorded as the observation flow velocity value. If the reference velocity value is greater than the observed velocity value, the difference between the reference velocity value and the observed velocity value is calculated as the single-point deficit; if the reference velocity value is not greater than the observed velocity value, the single-point deficit is set to zero. Multiply the preset first constant by the amplitude deviation to obtain the product value; add one to the product value to obtain the weighting coefficient; multiply the single-point deficit by the weighting coefficient to obtain the weighted deficit. The total reference flow rate value is obtained by summing up all the reference flow rate values ​​in the reference data sequence. The weighted deficit of all sampling points is summed to obtain the cumulative deficit; the sum of the total reference flow rate and the preset minimum value is calculated as the total flow rate; the cumulative deficit is divided by the total flow rate to obtain the reflux deficit index.

8. The MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU according to claim 6, characterized in that, The calculation of the return hysteresis index based on waveform hysteresis includes: The centroid calculation method is used to calculate the centroid positions of the reference data sequence and the observed data sequence, respectively. The centroid position difference is obtained by subtracting the centroid position of the reference data sequence from the centroid position of the observed data sequence. If the centroid position difference is positive, it is used as the centroid lag. If the centroid position difference is not greater than zero, the centroid lag is set to zero. Multiply the preset second constant by the waveform hysteresis to obtain the product result; then add one to the product result to obtain the correction coefficient; The product of the center of gravity lag and the correction coefficient is calculated as the reflux hysteresis index.

9. The MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU according to claim 1, characterized in that, The fusion of the reflux deficit index and the reflux hysteresis index yields a dynamic risk score, including: Based on preset weighting coefficients, a weighted summation is performed on the return deficit indicator and the return lag indicator to obtain a weighted sum value. Compare the weighted sum with the preset upper limit; if the weighted sum is less than or equal to the preset upper limit, retain the weighted sum; if the weighted sum is greater than the preset upper limit, truncate the weighted sum to the preset upper limit. The weighted sum is multiplied by a preset coefficient to obtain the dynamic risk score.

10. An MRI image feature extraction system for dynamic assessment of intracranial pressure in the ICU, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the MRI image feature extraction method for dynamic assessment of intracranial pressure in the ICU as described in any one of claims 1 to 9.

Images