Preoperative prediction system for cerebral hyperperfusion syndrome based on multi-modal data

Abstract

The present application relates to the technical field of medical data processing, in particular to a brain hyperperfusion syndrome preoperative prediction system based on multi-modal data. The system standardizes and quantizes image data to obtain the cerebral blood flow time series pairs of each voxel in the region; constructs a difference matrix according to the differences between the cerebral blood flow time series pairs of each voxel in the region; searches for the optimal matching path in the difference matrix; separates the blood flow delay time and the contrast agent retention degree according to the optimal matching path; determines the vascular permeability of each voxel in the region according to the blood flow delay time, the contrast agent retention degree and the vascular wall state index; and determines the vascular tolerance pressure value in combination with the pre-set theoretical basis tolerance pressure and the blood pressure load index; simulates the corresponding theoretical risk brain volume under a plurality of pre-set postoperative blood pressure target values based on the vascular tolerance pressure value to predict the postoperative blood pressure that the patient can safely tolerate, thereby accurately predicting the CHS incidence risk of different brain regions.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data. Background Technology

[0002] Cerebral hyperperfusion syndrome (CHS) is a potentially catastrophic complication following carotid artery revascularization. Its core pathology lies in the failure of cerebral autoregulation and damage to vascular wall structure caused by prolonged hypoperfusion. The high-pressure impact created by the sudden restoration of blood flow postoperatively easily triggers cerebral edema and hemorrhage, resulting in extremely high rates of disability and mortality. Therefore, accurate preoperative prediction of the patient's tolerable postoperative blood pressure is crucial for preventing CHS, ensuring surgical safety, and improving patient prognosis.

[0003] Currently, preoperative assessment in existing technologies primarily relies on imaging techniques, particularly computed tomography perfusion (CTP). The perfusion status of brain tissue is then assessed by analyzing the time-density curves generated by CTP and calculating hemodynamic parameters such as time to peak flow and mean transit time. However, due to the significant slowing of blood flow velocity distal to stenosis and the delayed arrival of contrast agent, the morphology of the time-density curve changes, exhibiting a broad and flattened characteristic. Furthermore, it is noteworthy that malignant extravascular leakage of contrast agent caused by blood-brain barrier disruption also manifests as waveform broadening on the CTP time-density curve. Therefore, in complex situations involving severe vascular stenosis, existing methods struggle to distinguish between the similar curve broadening exhibited by these two different physiological or pathological processes, resulting in an inability to fully reflect the true "pressure resistance" of the vessel wall and reducing the certainty of surgical risk prediction. Summary of the Invention

[0004] To address the technical problem of existing methods' inability to distinguish between "benign blood flow delay caused by vascular stenosis" and "malignant contrast agent extravasation caused by blood-brain barrier disruption," which leads to difficulties in accurately predicting the true tolerance pressure of the vessel wall, this invention provides a preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data. The specific technical solution adopted is as follows: This invention proposes a preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data. The system includes: The acquisition module is used to acquire the patient's preoperative imaging data, vascular wall status indicators, and blood pressure load indicators. The vascular wall status indicators are determined based on the patient's preoperative serum biomarker data, and the blood pressure load indicators are determined based on the patient's preoperative blood pressure monitoring data. The processing module is used to identify the affected side region of interest in the image data; and to standardize and quantize the image data to obtain the cerebral blood flow time series pairs of each voxel in the region. The separation module is used to construct a difference matrix based on the differences between the cerebral blood flow time series pairs of each voxel in the region; search for the optimal matching path in the difference matrix; and separate the blood flow delay time, which characterizes the blood flow transmission delay, and the contrast agent retention degree, which characterizes the abnormal retention of contrast agent, based on the optimal matching path. The determination module is used to determine the vascular permeability of each voxel in the region based on blood flow delay time, contrast agent retention, and vascular wall condition indicators; and to determine the vascular tolerance pressure value of each voxel in the region based on preset theoretical tolerance pressure, vascular permeability, and blood pressure load indicators. The prediction module simulates the theoretical risk brain volume corresponding to multiple preset postoperative blood pressure target values ​​based on the vascular tolerance pressure values ​​of each voxel in the region, in order to predict the postoperative blood pressure that the patient can safely tolerate.

[0005] Furthermore, the preoperative serum biomarker data includes specific protein concentration data, wherein the specific proteins include at least one biomarker selected from glial fibrillary acidic protein and β-amyloid protein; The process for determining the vessel wall condition indicators includes: Obtain the concentration measurement value of at least one biomarker from the specific protein concentration data; At least one concentration measurement is mapped to a scalar value ranging from 0 to 1 as an indicator of blood vessel wall condition.

[0006] Furthermore, the patient's preoperative blood pressure monitoring data includes a sequence of systolic blood pressure measurements over time during a specific preoperative period; The process for determining the blood pressure load index includes: Obtain the preset safe blood pressure baseline value and the time interval between adjacent measurement times in the systolic blood pressure measurement sequence; Iterate through each systolic blood pressure measurement in the sequence and determine whether it is greater than the safe blood pressure baseline. For each target measurement value that exceeds the safe blood pressure baseline, the difference between the target measurement value and the safe blood pressure baseline is calculated as the excess value; The blood pressure load index is obtained by summing the products of all excess values ​​and their corresponding time intervals.

[0007] Furthermore, the cerebral blood flow time series includes a first time series from any voxel in the affected side region of interest, and a second time series from a voxel at a corresponding mirror anatomical location in the contralateral cerebral hemisphere. The standardization and quantization of the image data to obtain time-series pairs of cerebral blood flow for each voxel within the region includes: For each voxel in the region of interest on the affected side, the original curves of contrast agent concentration changes over time at each voxel are extracted; the original curves are then resampled onto a preset standard time axis using mathematical interpolation to obtain a time axis-aligned sequence for the affected side; simultaneously, the same operation is performed at the corresponding location in the contralateral cerebral hemisphere to obtain a time axis-aligned reference sequence. Check whether the peak concentration of the reference sequence is higher than the preset background noise threshold; if not, use a preset standard reference curve to replace the reference sequence. If so, divide each data point in the affected side sequence by its own peak concentration value to obtain the first concentration baseline value; at the same time, divide each data point in the reference sequence by its own peak concentration value to obtain the second concentration baseline value. Arrange the first concentration baseline values ​​in chronological order to obtain the first time series; arrange the second concentration baseline values ​​in chronological order to obtain the second time series; the two together form a voxel-based cerebral blood flow time series pair.

[0008] Furthermore, the difference matrix construction process includes: Create a two-dimensional array with the number of rows equal to the length of the first time series and the number of columns equal to the length of the second time series, and set the initial value of all elements in the array to zero; Go through each time point in the first time series one by one and record it as the first time point; for the first time point, go through each time point in the second time series one by one and record it as the second time point; Read the first value of the first time series at the first time point, read the second value of the second time series at the second time point, and calculate the square of the difference between the first value and the second value as the squared difference. The squared differences are filled into a two-dimensional array with the first time point as the row index and the second time point as the column index. After traversing all time points in the first and second time series, the filled two-dimensional array is used as the difference matrix.

[0009] Furthermore, the search for the optimal matching path in the difference matrix includes: Set a maximum allowed time offset as a physiological constraint for the search; create a cumulative cost matrix with the same size as the difference matrix, and initialize all elements in the cumulative cost matrix to a special value representing infinity; Traverse each position of the cumulative cost matrix in order from the start point to the end point; for the current position, check whether it meets the physiological constraint condition, which indicates whether the absolute value of the difference between the row index and the column index represented by the current position does not exceed the maximum allowed time offset; if it does not meet the constraint condition, mark the current position as unreachable. If satisfied, select the best precursor point that satisfies the physiological constraints and has the minimum cumulative cost from the left, top, and upper left adjacent positions; Add the squared difference of the current position in the difference matrix to the cumulative cost of the best predecessor point to obtain the minimum cumulative cost of the current position and record it. Start backtracking from the endpoint; sequentially determine the cost source point for each location point, and set the source point as the previous location point on the path, while moving the location point to the source point; Continue this process until the starting position is reached; connect all the positions passed during the entire movement in sequence, and the resulting continuous trajectory is the optimal matching path.

[0010] Further, the separation of blood flow delay time (characterizing blood flow transmission delay) and contrast agent retention degree (characterizing abnormal contrast agent retention) based on the optimal matching path includes: Read each location point recorded on the optimal matching path sequentially, extract the target row index and target column index of the location point at the same position in the difference matrix, and the corresponding squared difference; For each location point on the path, calculate the absolute difference between the target row index and the target column index as the time offset; calculate the arithmetic mean of the time offsets calculated for all location points on the path to obtain the average time offset. Multiply the average time offset by the time sampling interval of the image data to obtain the blood flow delay time; Calculate the arithmetic mean of the squared differences corresponding to all locations along the path, and use it as the average difference value; The contrast agent retention rate is obtained by performing a square root operation on the average difference value.

[0011] Furthermore, the process of determining vascular permeability includes: For each voxel within the region, the blood flow delay time is divided by a preset normalized time constant to obtain a dimensionless ratio; the dimensionless ratio is then subjected to a hyperbolic tangent function operation to obtain the function output value; the function output value is added to the value of 1 to obtain the first sum; the first sum is divided by the value of 2 to obtain the final saturation coefficient. Multiply the preset delay gain coefficient by the saturation coefficient to obtain the first product value; add 1 to the first product value to obtain the delay amplification factor. The blood vessel wall condition index is added to a preset minimum positive compensation constant to obtain a second sum; the second sum is then subjected to a preset power mathematical operation to obtain a biochemical weighting coefficient. The vascular permeability of voxels within the region is obtained by multiplying the contrast agent retention rate, the delay amplification factor, and the biochemical weighting coefficient.

[0012] Furthermore, the process for determining the vascular tolerance pressure value includes: For each voxel within the region, the vascular permeability is multiplied by a preset damage sensitivity coefficient to obtain an intermediate product value; the exponential function value of the negative intermediate product value with the natural constant as the base is calculated as the damage attenuation coefficient. Divide the blood pressure load index by the preset load reference value to obtain the dimensionless relative load ratio; perform a natural logarithmic operation on the sum of the logarithmic value and the relative load ratio to obtain the logarithmic value; calculate the product of the logarithmic value and the preset fatigue sensitivity coefficient as the second product value; calculate the difference between the logarithmic value and the second product value, and perform non-negative truncation on the difference as the initial fatigue attenuation coefficient. The initial fatigue attenuation coefficient is compared with the preset strength minimum coefficient, and the larger of the two values ​​is taken as the final historical fatigue attenuation coefficient. The vascular tolerance pressure value of the voxel in the region is obtained by multiplying the preset theoretical tolerance pressure value, the damage attenuation coefficient, and the historical fatigue attenuation coefficient.

[0013] Furthermore, the process of obtaining the theoretical risk brain volume includes: Obtain the preset postoperative blood pressure target value to be simulated; Traverse all voxels within the affected side's area of ​​interest; for each postoperative blood pressure target, compare the vascular tolerance pressure value of the current voxel with the postoperative blood pressure target value. If the vascular tolerance pressure value is less than the postoperative blood pressure target value, the current voxel is marked as a high-risk voxel; otherwise, it is marked as a low-risk voxel. After all voxels in the affected side’s area of ​​concern have been compared, the total number of voxels marked as high-risk is counted. Multiply the total number by the actual physical volume represented by a single voxel in the image space to obtain the theoretical risk brain volume simulated under the postoperative blood pressure target value.

[0014] The present invention has the following beneficial effects: This invention overcomes the limitations of traditional single-data prediction by acquiring three core multimodal data types: preoperative imaging data, vascular wall status indicators, and blood pressure load indicators. It achieves a comprehensive assessment of cerebral vascular structure, vascular wall function, and blood pressure load status. Specifically, vascular wall status indicators accurately reflect vascular wall integrity, blood pressure load indicators quantify the cumulative effect of preoperative blood pressure abnormalities, and imaging data captures cerebral hemodynamic characteristics. These three factors work synergistically to provide comprehensive and accurate foundational data for subsequent predictions, effectively avoiding prediction biases caused by single-data assessments. Furthermore, by searching for the optimal matching path, it effectively distinguishes the different physiological meanings behind all "waveform broadening" patterns in the time-density curve, thereby specifically extracting "contrast agent retention," fundamentally solving the problem that waveform broadening is caused by "blood..." This study addresses the challenge of confusing "benign blood flow delay caused by vascular stenosis" with "malignant contrast agent extravasation caused by blood-brain barrier disruption," significantly reducing the misjudgment rate of benign delays while enhancing the detection capability of early leakage signals. By fusing blood flow delay time, contrast agent retention, and vascular wall status indicators, the study accurately calculates the vascular permeability of each voxel within the region. Furthermore, by combining blood pressure load indicators, it quantifies the vascular tolerance pressure value of each voxel, achieving a precise characterization of individual cerebrovascular tolerance. This breakthrough transforms multiple factors such as vascular function and blood pressure load into quantifiable pressure indicators, namely "vascular tolerance pressure values," enabling precise reflection of the differences in CHS risk in different brain regions and providing a scientific basis for subsequent prediction of postoperative blood pressure risk. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data, provided in one embodiment of the present invention. Figure 2 This is an example diagram illustrating the process of obtaining cerebral blood flow time series pairs according to an embodiment of the present invention; Figure 3 This is an example diagram illustrating the process of determining vascular tolerance pressure values ​​according to an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data provided by this invention.

[0020] Please see Figure 1 The diagram illustrates a preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data, according to an embodiment of the present invention. The system includes: The acquisition module 101 is used to acquire the patient's preoperative imaging data, vascular wall status indicators and blood pressure load indicators. The vascular wall status indicators are determined based on the patient's preoperative serum biomarker data, and the blood pressure load indicators are determined based on the patient's preoperative blood pressure monitoring data.

[0021] Preoperative imaging data refers to computed tomography perfusion imaging (CTP) data used to assess cerebral hemodynamics.

[0022] Acquiring perfusion imaging (CTP) data is a common technique in the medical field, and will not be described in detail in this embodiment.

[0023] For example, after intravenous injection of iodine contrast agent, a continuous and rapid CT scan is performed on a selected brain layer, ultimately obtaining a four-dimensional data volume composed of multiple voxels. The four-dimensional data volume contains three-dimensional spatial information (which can be represented by (x, y, z)) and a time dimension. The three-dimensional spatial coordinates (x, y, z) locate the position of the voxel in space, and the fourth dimension (time) records the CT values ​​of the voxel scanned at multiple consecutive time points after the injection of the contrast agent. Then, fixing the spatial coordinates (x, y, z) of a voxel, the curve formed by connecting its CT values ​​at different times along the time dimension is the time-density curve (TDC) corresponding to each voxel.

[0024] It is important to understand that, based on existing medical research, blood-specific proteins closely related to neurovascular unit damage and blood-brain barrier disruption can be selected as biomarkers. Typical biomarkers include glial fibrillary acidic protein (GFAP): an intermediate filament protein mainly expressed in astrocytes. If the blood-brain barrier is disrupted or glial cells are damaged, GFAP can be released into the blood. Its elevated serum concentration is associated with the severity and prognosis of various cerebrovascular diseases. β-amyloid protein: its metabolic abnormalities are associated with cerebral vascular amyloidosis and may affect the elasticity and fragility of the blood vessel wall.

[0025] It should be noted that the preoperative serum biomarker data includes specific protein concentration data, wherein the specific proteins include at least one biomarker among glial fibrillary acidic protein and β-amyloid protein.

[0026] For example, fasting venous blood is drawn from the patient in the morning before surgery, and the serum concentration of the above-mentioned specific proteins is quantitatively measured using standardized clinical immunological detection methods (such as enzyme-linked immunosorbent assay), to obtain the concentration measurement value of biomarkers, such as the concentration measurement value of GFAP and the concentration measurement value of β-amyloid protein.

[0027] In this embodiment, the concentration measurement value of at least one biomarker in the specific protein concentration data is obtained; the at least one concentration measurement value is mapped to a scalar value between 0 and 1 as an indicator of blood vessel wall status.

[0028] It is important to understand that, based on medical research, abnormally elevated concentrations of selected serum biomarkers (such as GFAP and β-amyloid) provide objective biochemical evidence of neurovascular unit damage, blood-brain barrier disruption, or cerebrovascular pathological changes. For example, GFAP, an astrocyte cytoskeleton protein, directly indicates glial cell damage and increased blood-brain barrier permeability when it enters the bloodstream. Abnormal accumulation of β-amyloid is associated with cerebral vascular amyloidosis and can weaken vascular wall elasticity. Therefore, the concentration of biomarkers is negatively correlated with the structural integrity of the vascular wall. Higher concentrations indicate more severe microscopic damage and poorer macroscopic "stability" of the vascular wall in response to hemodynamic shocks. Therefore, the mapping of concentration measurements to vascular wall status indicators should follow a core principle: higher input biomarker concentrations indicate more significant vascular wall-related pathological changes, and the larger the mapped scalar value (i.e., vascular wall status indicator) should be, closer to 1, indicating poorer vascular stability. Conversely, lower concentrations should result in smaller output scalar values, closer to 0, indicating relatively better vascular stability.

[0029] For example, assuming we obtain concentration measurements of two biomarkers, x1 (representing GFAP concentration measurement) and x2 (representing β-amyloid protein concentration measurement), then the vascular wall condition index can be expressed by the following formula: Wherein, α represents the blood vessel wall condition index; , , represents the model parameters obtained through training with historical data; x1 represents the measured value of GFAP concentration; x2 represents the measured value of β-amyloid protein concentration; exp() represents an exponential function with the natural constant as the base.

[0030] It should be noted that, , , The specific values ​​can be obtained by training with historical preoperative serum biomarker data, and this embodiment does not impose specific limitations. For example, historical preoperative serum biomarker data can be input into a logistic regression model. Using optimization algorithms such as maximum likelihood estimation, the computer automatically and iteratively adjusts the values. , , The value of is determined by adjusting the model so that it calculates the correct value for all patients who have "CHS" (represented by label = 1). The value should be as close to 1 as possible; for all patients who "did not have CHS" (which can be represented by label = 0), the calculated value is... The values ​​should be as close to 0 as possible; ultimately, a set of predictions should be found that best matches the actual clinical outcomes. , , Value; after training, this set of optimal parameters is fixed as... , , The specific value, for example =-5.0; =0.02; =0.01.

[0031] It should be noted that the patient's preoperative blood pressure monitoring data includes a sequence of systolic blood pressure measurements over a specific preoperative period.

[0032] It should be noted that the specific preoperative time period can be determined based on the actual situation, and this embodiment does not impose a specific limitation. For example, it is usually within 24 hours before the patient's surgery.

[0033] For example, patients wear a portable ambulatory blood pressure monitor before surgery. The monitor automatically and non-invasively measures and records systolic blood pressure over 24 hours (usually every 15-30 minutes), ultimately forming a sequence of systolic blood pressure measurements.

[0034] In this embodiment, a preset safe blood pressure baseline value and the time interval between adjacent measurement times in the systolic blood pressure measurement value sequence are obtained; each systolic blood pressure measurement value in the systolic blood pressure measurement value sequence is traversed to determine whether it is greater than the safe blood pressure baseline value; for each target measurement value that is greater than the safe blood pressure baseline value, the difference between the target measurement value and the safe blood pressure baseline value is calculated as the excess value; the products of all excess values ​​and their corresponding time intervals are accumulated to obtain the blood pressure load index.

[0035] It should be noted that the specific value of the preset safe blood pressure baseline can be determined based on authoritative clinical guidelines and expert consensus on hypertension management, and this embodiment does not impose specific limitations. For example, a common value is 140 mmHg. For higher-risk patients or those requiring stricter management, it can be set to 130 mmHg; for relatively young patients without complications, 150 mmHg may also be used.

[0036] It's important to understand that, from a vascular biomechanical perspective, blood pressure is a continuous stress acting on the blood vessel wall. Long-term or repeated blood pressure exceeding the blood vessel's self-regulating capacity (i.e., above the safe baseline blood pressure) can cause microscopic damage to the vascular endothelium and smooth muscle. This damage is cumulative over time; a single, brief overpressure episode may be repaired, but repeated, continuous overpressure leads to cumulative damage, vascular remodeling, decreased elasticity, and ultimately increased vascular fragility and a lower tolerance threshold. Therefore, calculating and summing "excess value × corresponding time interval" is precisely to quantify the severity of this harmful pressure. A higher blood pressure load index indicates a higher cumulative mechanical load on the blood vessel wall in the recent preoperative period.

[0037] It is important to clarify that the vascular wall status index and blood pressure load index are single, global values ​​calculated based on the patient's overall condition (i.e., each patient has two specific vascular wall status indices and blood pressure load indices). In subsequent calculations, the patient's vascular wall status index and blood pressure load index will be applied to each voxel in their brain imaging that needs to be analyzed.

[0038] The processing module 102 is used to identify the affected side region of interest in the image data; and to standardize and quantify the image data to obtain the cerebral blood flow time series pairs of each voxel in the region.

[0039] It should be noted that the affected side region of interest can be identified using two well-known existing technologies in the field of medical image processing: "image segmentation" and "atlas registration".

[0040] For example, using vascular segmentation technology, the patient's preoperative imaging data is analyzed to identify the responsible artery (such as the left internal carotid artery) with severe stenosis or occlusion; a pre-set standard cerebral artery blood flow domain atlas is retrieved, which records the blood supply correspondence between each major cerebral artery and brain tissue regions; based on the identified responsible artery, the corresponding standard blood supply area template is found and extracted from the standard cerebral artery blood flow domain atlas; the standard blood supply area template is mapped to the patient's brain imaging space using image registration technology to obtain a three-dimensional brain tissue region; the three-dimensional brain tissue region is defined as the affected side region of interest.

[0041] The process of identifying the responsible artery using vascular segmentation technology is a publicly available technique in the field of medical image processing, and this embodiment can be directly applied. For example, from a patient's computed tomography angiography (CTA) image, a complete three-dimensional model of the brain's vascular tree can be automatically extracted using region growing, level set methods, or deep learning-based vascular segmentation networks. Then, on this model, by measuring the percentage of luminal stenosis in each segment of the vessel and comparing it with the normal proximal diameter, the target vessel with significant hemodynamic stenosis (e.g., a luminal stenosis percentage greater than 70%) can be automatically identified and labeled as the responsible artery.

[0042] It should be noted that the standard cerebral artery blood flow domain atlas is a public research resource in the field of neuroimaging, and will not be described in detail in this embodiment.

[0043] As an example, the cerebral blood flow time series pair includes a first time series from any voxel in the affected side region of interest, and a second time series from a voxel at the corresponding mirror anatomical location in the contralateral cerebral hemisphere.

[0044] Voxel corresponding to the mirror anatomical location of the contralateral cerebral hemisphere: a voxel located in the healthy cerebral hemisphere that is precisely symmetrical in three-dimensional space with respect to any voxel in the region about the midsagittal plane of the brain (i.e. the boundary between the left and right hemispheres).

[0045] It should be noted that the specific process of locating the corresponding mirror anatomical position voxel of the contralateral cerebral hemisphere is as follows: Each voxel has a unique spatial coordinate (x, y, z); during localization, the coordinates of the voxel on the affected side are directly mirrored, that is, the y-axis (anteroposterior direction) and z-axis (vertical direction) coordinates are usually kept unchanged, while the x-axis (left-right direction) coordinates are reversed (i.e., from +x to -x, or vice versa); the voxel corresponding to the spatial coordinates (-x, y, z) obtained after the transformation is the corresponding mirror anatomical position voxel of the contralateral cerebral hemisphere.

[0046] The process of obtaining cerebral blood flow time series is as follows Figure 2 As shown, it includes: S101-1: For each voxel in the area of ​​interest on the affected side, extract the original curve of the contrast agent concentration changing over time at each voxel; resample the original curve to a preset standard time axis using mathematical interpolation to obtain a time axis-aligned sequence for the affected side; at the same time, perform the same operation at the corresponding position in the contralateral cerebral hemisphere to obtain a time axis-aligned reference sequence.

[0047] The original curve specifically refers to the discrete data sequence of contrast agent concentration (CT value, unit HU) that is directly extracted from a specific voxel in CT perfusion imaging data and shows the change of contrast agent concentration over scan time.

[0048] A preset standard time axis refers to a time scale with fixed duration and fixed time intervals. For example, from 0 seconds before the injection of contrast agent to 60 seconds after the injection, this total duration is evenly divided into dozens of discrete time points at fixed and equal intervals (e.g., one point per second). This means that the original curves of all voxels will be resampled to generate a discrete sequence of length 61, with corresponding CT values ​​at 0 seconds, 1 second, 2 seconds... up to 60 seconds.

[0049] S101-2: Check whether the peak concentration of the reference sequence is higher than the preset background noise threshold; if not, use a preset standard reference curve to replace the reference sequence.

[0050] It should be noted that the specific value of the preset background noise threshold can be determined based on the system noise level of CT imaging and the range of background CT values ​​of brain tissue; this embodiment does not impose a specific limitation. For example, in brain tissue CTP scanning, the baseline CT value of brain parenchyma (gray matter, white matter) is usually between 20-40 HU, while the CT value of air or noise may fluctuate close to 0 HU. Therefore, the preset background noise threshold can be set at around 10 HU to reliably distinguish between real, meaningful brain tissue perfusion signals and invalid signals generated by equipment noise, partial volume effects, or cerebrospinal fluid regions.

[0051] It should be noted that the preset standard reference curve is a standardized time-density curve representing the idealized perfusion response of healthy brain tissue. For example, it can be generated by statistical modeling of CTP data from a large number of patients clearly unaffected by cerebrovascular disease. The specific process includes: collecting CTP data from a large number of normal individuals; performing the same standardization and quantification steps as in this invention on all CTP data (such as time axis alignment and peak normalization); calculating the average CT value of the processed curves for all individuals at each time point in a specific region of interest within standard brain space (such as the typical location of the middle cerebral artery supply area); and connecting these average CT values ​​in chronological order to form a preset standard reference curve.

[0052] S101-3: If so, divide each data point in the affected side sequence by its own peak concentration value to obtain the first concentration baseline value; at the same time, divide each data point in the reference sequence by its own peak concentration value to obtain the second concentration baseline value.

[0053] It should be noted that the peak concentration value refers to the maximum value found by searching for the contrast agent concentration values ​​at all time points of a time-density curve that has been time-aligned (whether it is the affected side sequence or the reference sequence).

[0054] S101-4: Arrange the first concentration benchmark values ​​in chronological order to obtain the first time series; arrange the second concentration benchmark values ​​in chronological order to obtain the second time series; the two together form a voxel cerebral blood flow time series pair.

[0055] The separation module 103 is used to construct a difference matrix based on the differences between the cerebral blood flow time series pairs of each voxel in the region; search for the optimal matching path in the difference matrix; and separate the blood flow delay time, which characterizes the blood flow transmission delay, and the contrast agent retention degree, which characterizes the abnormal retention of the contrast agent, based on the optimal matching path.

[0056] It is important to understand that the goal of this invention is to find the optimal non-linear temporal correspondence between the affected side and the reference sequence (i.e., a match that allows for time axis scaling), rather than a simple point-by-point alignment. Therefore, a difference matrix is ​​constructed by systematically evaluating the differences under all possible time-point pairing methods. This difference matrix is ​​used to quantify the "matching cost" search space, where each element represents the degree of mismatch under the time alignment assumption.

[0057] In this embodiment, a two-dimensional array is created with the number of rows equal to the length of the first time series and the number of columns equal to the length of the second time series. The initial value of all elements in the array is set to zero. Each time point in the first time series is iterated through and recorded as the first time point. For the first time point, each time point in the second time series is iterated through and recorded as the second time point. The first value of the first time series at the first time point is read, and the second value of the second time series at the second time point is read. The square of the difference between the first value and the second value is calculated as the squared difference. The squared difference is filled into the two-dimensional array at the position with the first time point as the row index and the second time point as the column index. After traversing all time points in the first and second time series, the filled two-dimensional array is used as the difference matrix.

[0058] The squared difference is the result obtained by subtracting the concentration value of the affected sequence at a certain time point (represented by i) from the concentration value of the reference sequence at a certain time point (represented by j) in the difference matrix, and then squaring the difference.

[0059] Specifically, the smaller the squared difference between the affected sequence at time i and the reference sequence at time j, the closer the concentration values ​​of the two sequences are, assuming that the state at time i on the affected side corresponds to the state at time j on the reference side. This usually means that the time correspondence is more reasonable and more likely to reflect the true time delay relationship. On the other hand, the larger the squared difference between the affected sequence at time i and the reference sequence at time j, the greater the difference in the states of the two curves under the above assumption. This may mean that the time correspondence is more unreasonable (e.g., attempting to match the rising segment of the affected side to the falling segment of the reference), or more likely to reflect inherent morphological differences that cannot be explained by time offset (e.g., curve morphological distortion caused by leakage).

[0060] It is understandable that the row indices in the difference matrix correspond to the time points of the affected side sequence, and the column indices correspond to the time points of the reference sequence.

[0061] The difference matrix quantifies the degree of mismatch between the first and second time series under all possible time point alignment methods.

[0062] It is important to understand that because the affected side's sequences may experience overall delay, partial delay, or waveform distortion, simple linear comparisons cannot capture this complex temporal distortion relationship. Therefore, a matching scheme that minimizes the overall morphological differences and allows for non-uniform scaling of the time axis can be found. This involves searching for the optimal matching path in the difference matrix to simulate an alignment that can best explain the difference between the two curves using "simple time delay." The optimal matching path itself defines the magnitude of the "benign delay," while the residual differences that the path fails to eliminate are defined as "malignant extravasation" signals that cannot be explained by delay. This achieves a mathematical decoupling between the "benign delay caused by vascular stenosis" and the "malignant extravasation of contrast agent caused by blood-brain barrier disruption."

[0063] It should be noted that a constrained dynamic time warping algorithm is used to search for an optimal matching path from the upper left corner to the lower right corner of the difference matrix, which will not be described in detail in this embodiment.

[0064] As an example, searching for the optimal matching path in the difference matrix includes the following steps: 1) Set the maximum allowed time offset as a physiological constraint for the search; create a cumulative cost matrix with the same size as the difference matrix, and initialize all elements in the cumulative cost matrix to a special value representing infinity.

[0065] 2) Traverse each position of the cumulative cost matrix in order from the starting point to the ending point; for the current position, check whether it meets the physiological constraint condition, where the physiological constraint condition is used to indicate whether the absolute value of the difference between the row index and the column index represented by the current position does not exceed the maximum allowed time offset; if it does not meet the constraint condition, mark the current position as unreachable.

[0066] 3) If satisfied, select the best precursor point from the left, top, and upper left adjacent positions that meet the physiological constraints and have the lowest cumulative cost.

[0067] 4) Add the squared difference of the current position in the difference matrix to the cumulative cost of the best predecessor point to obtain the minimum cumulative cost of the current position and record it.

[0068] 5) Start backtracking from the endpoint; determine the cost source point for each position point in sequence, and set the source point as the previous position point on the path, while moving the position point to the source point.

[0069] 6) Continue this process until you reach the starting position; connect all the positions you have passed through in sequence, and the resulting continuous trajectory is the optimal matching path.

[0070] The optimal matching path refers to a continuous sequence of points from the start to the end of the difference matrix, which is found under the premise of satisfying the physiological time offset constraint. The optimal matching path defines which time point on the reference side's time-density curve (i.e., the first time series) should be best matched with each time point on the affected side's time-density curve (i.e., the second time series) so that the overall morphological difference between the two curves is minimized over the entire time course.

[0071] It should be noted that the specific value of the maximum permissible time offset can be determined based on the maximum physiological delay time required for the establishment of cerebral collateral circulation, and this embodiment does not impose a specific limitation. For example, in cases of severe carotid artery stenosis, blood flow compensates through primary collaterals (such as the Circle of Willis) or secondary collaterals. This process leads to a delay in blood flow arrival. Based on extensive clinical research and physiological consensus, this delay is usually within a few seconds (e.g., no more than 6-8 seconds). Therefore, the maximum permissible offset can be set within this physiological limit (e.g., 4 seconds) to prevent excessive matching that goes against physiological common sense, such as "matching the 5th second of the affected side to the reference 20th second" in order to mathematically find the optimal matching path.

[0072] It is important to understand that, even after optimal time alignment, the morphological differences between the affected and reference curves cannot be eliminated and therefore cannot be explained by the "delay in blood flow." Thus, by quantifying these "inescapable morphological differences" separately, contrast agent retention is obtained. Contrast agent retention measures these residual morphological differences, characterizing malignant pathological changes caused by blood-brain barrier disruption, contrast agent extravasation, and abnormal retention. It is a direct radiographic indicator for predicting the risk of vascular wall integrity collapse (such as hemorrhage).

[0073] In this embodiment, each location point recorded on the optimal matching path is read sequentially, and the target row index and target column index of the location point at the same position in the difference matrix, as well as the corresponding squared difference, are extracted. For each location point on the path, the absolute difference between the target row index and the target column index is calculated as the time offset. The arithmetic mean of the time offsets calculated for all location points on the path is calculated to obtain the average time offset. The average time offset is multiplied by the time sampling interval of the image data to obtain the blood flow delay time. The arithmetic mean of the squared differences corresponding to all location points on the path is calculated as the average difference value. The square root of the average difference value is then performed to obtain the contrast agent retention rate.

[0074] The same position of a location point in the difference matrix means that the row index and column index of a coordinate point recorded on the optimal matching path correspond exactly to the element with the same row index and column index in the previously constructed difference matrix.

[0075] It should be noted that if the time offset at a certain point on the path is larger, it means that in order to align the two curves (i.e., the time-density curve of the affected side and the time-density curve of the reference side), the time axis needs to be offset to a greater extent. This may mean that the blood flow delay represented at that point is more significant. If the time offset at a certain point on the path is smaller, it means that the time points of the two curves can be aligned more easily in the local tissue at that point, and the blood flow delay represented is less obvious.

[0076] The average time offset is the arithmetic mean of the time offsets of all points along the path. A larger average time offset indicates a greater degree of time distortion required for the two curves to align, resulting in a more severe overall blood flow delay.

[0077] The time sampling interval refers to the fixed time length represented between two adjacent data points in a cerebral blood flow time series pair after resampling. For example, assuming the preset standard time axis is from 0 to 60 seconds, with one point per second, then the time sampling interval is 1 second.

[0078] It's important to understand that blood flow delay time refers to the physical time required to align the time-density curve of the affected brain tissue as closely as possible with the time-density curve of the reference side. A larger blood flow delay time, calculated to align the time-density curve of the affected brain tissue with the reference side for a given voxel, indicates a greater stretching of the time axis required to match the two curves. This means the affected curve lags significantly behind the reference curve, strongly suggesting severe stenosis or obstruction in the upstream blood vessels (such as the carotid artery), resulting in slower contrast agent delivery to the brain region – a typical hemodynamic delay. Conversely, a smaller blood flow delay time indicates closer proximity of the two curves on the time axis, with less time distortion required for alignment. This suggests relatively unobstructed blood supply to the local area represented by the voxel, and less significant delay in contrast agent arrival compared to the normal side.

[0079] It should be noted that a larger average difference value indicates that even with optimal time alignment, the concentration values ​​of the two curves at various points still generally differ significantly. This strongly suggests the presence of inherent morphological anomalies that cannot be explained by time delay, i.e., a higher probability of vascular leakage. On the other hand, a smaller average difference value indicates that after optimal time alignment, the two curves match well at most points, with a higher degree of morphological similarity. This strongly suggests that the integrity of the vessel wall may be better, and the leakage signal may be weak.

[0080] It's important to understand that directly using the average difference value can amplify calculation errors due to the squaring operation. Therefore, by taking the square root of the average difference value, we obtain the contrast agent retention rate, ensuring its dimensions are consistent with the original concentration value, making it easier to understand intuitively.

[0081] The determination module 104 is used to determine the vascular permeability of each voxel in the region based on blood flow delay time, contrast agent retention degree and vascular wall status indicators; and to determine the vascular tolerance pressure value of each voxel in the region based on preset theoretical tolerance pressure, vascular permeability and blood pressure load indicators.

[0082] It is important to understand that the risk of vascular rupture in CHS is the result of multiple synergistic factors, and a single-dimensional indicator is insufficient for a comprehensive assessment. Therefore, a comprehensive indicator integrating multidimensional information can be constructed, with vascular permeability as an intermediate variable. This indicator uses a nonlinear model to integrate contrast agent retention (direct damage to blood vessels), blood flow delay time (the environment in which the damaged blood vessel is located), and vascular wall condition indicators (the vulnerability of the blood vessel itself). This allows for a more comprehensive and stable quantification of the overall pathological damage to the vascular wall in its current state.

[0083] In this embodiment, for each voxel within the region, the blood flow delay time is divided by a preset normalized time constant to obtain a dimensionless ratio; a hyperbolic tangent function is applied to the dimensionless ratio to obtain a function output value; the function output value is added to the function output value to obtain a first sum; the first sum is divided by the value 2 to obtain the final saturation coefficient; a preset delay gain coefficient is multiplied by the saturation coefficient to obtain a first product value; the first product value is added to the value 1 to obtain a delay amplification factor; the vascular wall state index is added to a preset minimum positive compensation constant to obtain a second sum; a preset power operation is performed on the second sum to obtain a biochemical weighting coefficient; the contrast agent retention, delay amplification factor, and biochemical weighting coefficient are multiplied together to obtain the vascular permeability of the voxels within the region.

[0084] It should be noted that the specific value of the preset normalized time constant can be determined based on the analysis of the blood flow delay time distribution range in typical clinical data, and this embodiment does not impose a specific limitation. For example, a typical preset normalized time constant can be set to 10 seconds.

[0085] It should be noted that the hyperbolic tangent function has the following characteristics: if the input value is 0, the output is 0; if the input value is a large positive number, the output approaches 1 infinitely but is always less than 1; if the input value is a large negative number, the output approaches -1 infinitely but is always greater than -1. Therefore, the output range of the hyperbolic tangent function is between -1 and +1, meaning the function output value ranges from -1 to +1. However, the required saturation coefficient ranges from 0 to 1. Therefore, the following formula can be used for adjustment: Saturation coefficient = .

[0086] Because blood flow delay (stasis) prolongs the contact time between the contrast agent and the diseased vessel wall, it may exacerbate observed leakage. However, this exacerbation effect gradually approaches a physiological upper limit as the delay increases. Therefore, this invention does not simply allow the delay time to linearly amplify the risk indefinitely, but rather aims to simulate this "risk amplification with an upper limit." The original blood flow delay time is converted into a saturation coefficient between 0 and 1. A larger saturation coefficient indicates a longer blood flow delay time, reaching or approaching its maximum potential risk to the vessel wall. In this case, the "boosting" effect of the delay on leakage risk is more fully accounted for. Conversely, a smaller saturation coefficient (closer to 0) indicates a shorter blood flow delay time, with a negligible amplification effect on leakage risk.

[0087] It should be noted that the specific value of the preset delay gain coefficient can be determined through regression training of historical clinical data, and this embodiment does not impose a specific limitation. For example, a typical preset delay gain coefficient can be set to 0.8.

[0088] It should be noted that the preset coefficients involved in this invention (such as delay gain coefficient 0.8, damage sensitivity coefficient 2.0, power exponent, etc.) are not fixed physical constants. Their actual values ​​are obtained by training and calibrating through machine learning algorithms (such as logistic regression, support vector machine) based on historical case databases and labeled with the actual postoperative CHS incidence rate. The values ​​in the embodiments are only examples.

[0089] It's important to understand that directly multiplying contrast agent retention by (delay gain factor × saturation factor) results in a product of (delay gain factor × saturation factor) approaching 0 when the delay is small. This would lead to excessive suppression of vascular permeability. However, even without delay, significant leakage is inherently high-risk, which doesn't align with pathological reality. Therefore, by adding a value of 1, a baseline value of 1 for the delay amplification factor is ensured.

[0090] It should be noted that the purpose of the preset minimum positive compensation constant is to mathematically prevent the entire biochemical weighting coefficient calculation result from being 0 when the blood vessel wall state index is the theoretical perfect value of 0. The specific value of the preset minimum positive compensation constant can be determined based on industry experience, and this embodiment does not impose a specific limitation. For example, a typical preset minimum positive compensation constant can be set to 1e-2.

[0091] It's important to understand that the risk of blood vessel wall rupture is the result of both macroscopic "image-visible damage" and microscopic "biochemical structural fragility," and these two factors are not independent but rather mutually influential. Therefore, blood vessel wall condition indicators should not be simply regarded as an independent risk factor, but rather as a modulator used to adjust the risk weight of image-related damage. The biochemical weighting coefficient obtained through blood vessel wall condition indicators is precisely such a modulator; that is, if the blood vessel wall condition indicators show poor blood vessel stability (high blood vessel wall condition indicator value), the biochemical weighting coefficient will also increase accordingly, thereby amplifying the risk represented by contrast agent retention. This aligns more closely with the pathophysiological principle that "the higher the fragility of the blood vessel itself, the more easily it is damaged under the same external force."

[0092] The biochemical weighting coefficient quantifies the microscopic fragility of the vascular wall as reflected by indicators of the patient's vascular wall condition, and modulates the risk of macroscopic damage (i.e., contrast agent retention) detected by imaging. A larger biochemical weighting coefficient (closer to 1) indicates poorer vascular wall microscopic stability based on serum marker assessments. In this case, the same degree of contrast agent retention will be considered a higher overall risk because the fragile vascular wall is more prone to rupture under pressure.

[0093] It should be noted that the biochemical weighting coefficients can be represented by the following predefined powers in mathematical operations: Where W represents the biochemical weighting coefficient; Indicators representing the condition of the blood vessel wall; represents the preset minimum positive compensation constant; r represents the power exponent.

[0094] It should be noted that the specific value of the power exponent (r) can be obtained based on the machine learning training process of historical clinical data, and this embodiment does not impose a specific limitation. For example, a typical preset power exponent r can be set to 0.5.

[0095] It is important to understand that since cerebral hyperperfusion syndrome is essentially caused by postoperative blood pressure exceeding the physical tolerance limit of the diseased cerebral blood vessels, resulting in bleeding or leakage, preoperative risk assessment should convert the risk into a quantitative physical indicator that can be directly linked to clinical blood pressure management, namely the vascular tolerance pressure value.

[0096] The process of determining vascular tolerance pressure value is as follows: Figure 3 As shown, it includes: S102-1: For each voxel in the region, multiply the vascular permeability by a preset damage sensitivity coefficient to obtain an intermediate product value; calculate the exponential function value of the negative intermediate product value with the natural constant as the base, as the damage attenuation coefficient.

[0097] It's important to understand that, as a biological material, the strength loss of the blood vessel wall does not decrease linearly with damage, but rather "accelerates failure" after the damage accumulates to a certain level. Therefore, a mathematical model is needed to simulate this nonlinear collapse trend. The exponential decay function, exp(-damage sensitivity coefficient × vascular permeability), is a classic model describing this nonlinear collapse process. When the nonlinear collapse is small, the strength decreases gradually; however, if the nonlinear collapse exceeds a certain critical point, the strength drops sharply. In summary, calculating the damage decay coefficient is to quantify this nonlinearly accelerated strength reduction process as damage intensifies.

[0098] It should be noted that the specific value of the preset damage sensitivity coefficient can be determined based on machine learning training of historical clinical data, and this embodiment does not impose a specific limitation. For example, a typical preset damage sensitivity coefficient can be set to 2.0.

[0099] The damage attenuation coefficient represents the percentage of the theoretically based tolerance pressure remaining after the current damage to the blood vessel wall structure.

[0100] Among them, the larger the damage attenuation coefficient (closer to 1), the lower the vascular permeability, that is, the more minor the pathological damage, the smaller the weakening effect on vascular strength, and the closer the vascular strength is to the healthy baseline; the smaller the damage attenuation coefficient (closer to 0), the higher the vascular permeability, the greater the weakening effect on vascular strength, the more significant the vascular strength attenuation, and the closer to the failure edge.

[0101] S102-2: Divide the blood pressure load index by the preset load reference value to obtain the dimensionless relative load ratio; perform a natural logarithmic operation on the sum of the logarithmic value 1 and the relative load ratio to obtain the logarithmic value; calculate the product of the logarithmic value and the preset fatigue sensitivity coefficient as the second product value; calculate the difference between the logarithmic value 1 and the second product value, and perform non-negative truncation on the difference as the initial fatigue attenuation coefficient.

[0102] It should be noted that the specific value of the preset load reference value can be determined based on statistical analysis of the typical distribution of blood pressure load in clinical data, and this embodiment does not impose a specific limitation. For example, the median of the common load range can be used as a reference value, and a typical preset load reference value can be set to 100 mmHg·h.

[0103] It should be noted that a higher relative load ratio indicates that the patient's blood pressure load is significantly higher than the set load reference value, suggesting that their blood vessels have recently been subjected to abnormally heavy and continuous overpressure, and that the accumulation of historical "fatigue damage" may be more severe.

[0104] It should be noted that the non-negative truncation process is as follows: if the difference between the value 1 and the second product value is less than 0, then the initial fatigue decay coefficient is 0.

[0105] It is important to understand that since the initial high load causes the greatest damage to blood vessels, the incremental damage from subsequent equal loads will gradually decrease. This conforms to the "diminishing marginal effect" law of blood pressure load on vascular damage, which is a saturation characteristic of biological damage. Therefore, this invention uses a logarithmic function to simulate this "diminishing marginal effect" law. The characteristics of the logarithmic function are: if the input value is very small, its growth is almost linearly related to the input value (i.e., it represents large initial damage); if the input value is very large, its growth is much slower than the input value (i.e., it represents small additional damage from subsequent equal loads).

[0106] It should be noted that the specific value of the preset fatigue sensitivity coefficient can be obtained through training with historical clinical data, and this embodiment does not impose a specific limitation. For example, a typical preset fatigue sensitivity coefficient can be set to 0.1.

[0107] Since the damage to the blood vessel wall caused by historical blood pressure load is a gradual, cumulative process that weakens its strength, and this weakening should have an upper limit (i.e., the intensity of blood pressure load damage to the blood vessel cannot be reduced to a negative value), a decreasing mapping relationship can be constructed from the load (i.e., the blood pressure load index) to the fatigue attenuation coefficient. This is achieved by subtracting an attenuation amount determined by both the logarithmic value and a preset fatigue sensitivity coefficient, using 1 (representing no attenuation) as a baseline. Specifically, as the load (i.e., the blood pressure load index) increases, the logarithmic value increases, the attenuation amount increases, and thus the fatigue attenuation coefficient decreases.

[0108] The fatigue decay coefficient represents the percentage of the theoretically basic tolerance pressure value of blood vessels remaining due to "fatigue damage" caused by long-term blood pressure load accumulation. A larger fatigue decay coefficient (closer to 1) indicates a lower blood pressure load, a smaller weakening effect of historical load on vascular strength, and a less severe "fatigue damage" to the blood vessels.

[0109] S102-3: Compare the initial fatigue attenuation coefficient with the preset strength minimum coefficient, and take the larger value of the two as the final historical fatigue attenuation coefficient.

[0110] It should be noted that the preset strength guarantee factor is set based on the following: even under extremely severe historical loads, the blood vessel wall, as a biological material, is considered to retain a minimum inherent structural strength and will not completely lose its tensile capacity. Its specific value can be determined based on industry experience, and this embodiment does not impose any specific limitations. For example, a typical preset strength guarantee factor can be set to 0.4.

[0111] S102-4: Multiply the preset theoretical tolerance pressure value, damage attenuation coefficient, and historical fatigue attenuation coefficient to obtain the vascular tolerance pressure value of the voxel in the region.

[0112] It should be noted that the specific value of the preset theoretical tolerance pressure can be determined based on the consensus of literature on vascular biomechanics research and clinical experience. It represents the theoretical upper limit of static pressure that the walls of cerebral blood vessels (such as arterioles and precapillary resistance vessels) can withstand without rupture under ideal and healthy physiological conditions. For example, a typical preset theoretical tolerance pressure value can be set at 220 mmHg.

[0113] The vascular tolerance pressure value combines the pathological damage state (i.e., vascular permeability) and historical load accumulation (i.e., historical fatigue decay coefficient) of the local blood vessel wall represented by a particular voxel to calculate the theoretical tensile strength limit of the local blood vessel wall represented by that voxel under the current state. A higher vascular tolerance pressure value for a particular voxel indicates higher overall strength of the local blood vessel wall, enabling it to withstand higher intraluminal pressures without easily rupturing, and resulting in a relatively lower risk of postoperative hyperperfusion hemorrhage.

[0114] It should be noted that the vascular tolerance pressure value is essentially a quantitative score that characterizes the overall tolerance of the vascular wall. For the sake of clinical understanding, it is mapped to an equivalent pressure unit (mmHg), rather than being a directly measured physical burst pressure.

[0115] The prediction module 105 simulates the theoretical risk brain volume corresponding to multiple preset postoperative blood pressure target values ​​based on the vascular tolerance pressure values ​​of each voxel in the region, so as to predict the postoperative blood pressure that the patient can safely tolerate.

[0116] In this embodiment, a preset postoperative blood pressure target value to be simulated is obtained; each voxel in the affected side's area of ​​interest is traversed; for each postoperative blood pressure target value, the vascular tolerance pressure value of the traversed current voxel is numerically compared with the postoperative blood pressure target value; if the vascular tolerance pressure value is less than the postoperative blood pressure target value, the current voxel is marked as a high-risk voxel; otherwise, it is marked as a low-risk voxel; after all voxels in the affected side's area of ​​interest have been traversed and compared, the total number of all voxels marked as high-risk voxels is counted; the total number is multiplied by the actual physical volume represented by a single voxel in the image space to obtain the theoretical risk brain volume simulated under the postoperative blood pressure target value.

[0117] It should be noted that the specific value of the preset postoperative blood pressure target can be set based on the common practice range of postoperative blood pressure management in clinical practice, and this embodiment does not impose a specific limitation. For example, a series of values ​​from 100 mmHg to 160 mmHg in 5 mmHg increments, each postoperative blood pressure target value represents a hypothetical postoperative blood pressure control plan.

[0118] It should be noted that if the vascular tolerance pressure value of a voxel is less than a postoperative blood pressure target value, this means that if the postoperative blood pressure is controlled at the postoperative blood pressure target value, the pressure applied to the vascular wall of that voxel has exceeded its theoretical tolerance limit. Therefore, this voxel is marked as a high-risk voxel.

[0119] It should be noted that if the vascular tolerance pressure value of a certain voxel is greater than or equal to a postoperative blood pressure target value, it means that the blood vessel wall strength is sufficient to withstand the pressure at that postoperative blood pressure target value. Therefore, this voxel is marked as a low-risk voxel.

[0120] The actual physical volume represented by a single voxel in image space is directly determined by the scanning parameters and reconstruction algorithm of the CT image; it is a fixed and known physical quantity. Specifically, at the beginning of image processing, the pixel size and slice thickness parameters are directly read from the header information of the image file. These parameters are then multiplied to obtain the actual physical volume of a single voxel. This is a well-known technique and will not be described in detail here.

[0121] The theoretical risk brain volume simulates the total volume of space occupied by all voxels in the affected brain region where the vascular tolerance pressure is lower than the target postoperative blood pressure. A larger theoretical risk brain volume indicates that the estimated strength of the blood vessel walls in a larger area of ​​the brain is insufficient to withstand the target blood pressure, meaning a greater spatial extent of overall hemorrhage risk, and the current postoperative blood pressure control protocol (i.e., at the target postoperative blood pressure) is considered less safe. Conversely, a smaller theoretical risk brain volume indicates that the vascular strength in a smaller number of brain regions is lower than the target postoperative blood pressure, meaning a more limited spatial extent of overall hemorrhage risk, and the use of the target postoperative blood pressure is relatively safer.

[0122] It's important to understand that after obtaining a series of correspondences between "postoperative blood pressure target values ​​and theoretical risk brain volume," a preset risk volume tolerance value (e.g., 3 mL) can be used as a safety standard. From a series of "postoperative blood pressure target values ​​- theoretical risk brain volume" data, the highest blood pressure target value that satisfies the safety condition of "theoretical risk brain volume not exceeding the risk volume tolerance value" is identified and output. Therefore, the highest blood pressure target value is the predicted, patient-specific, and safe upper limit of postoperative blood pressure. If all simulated postoperative blood pressure target values ​​fail to meet the safety condition, an extremely high-risk warning is issued.

[0123] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0124] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data, characterized in that, The system includes: The acquisition module is used to acquire the patient's preoperative imaging data, vascular wall status indicators, and blood pressure load indicators. The vascular wall status indicators are determined based on the patient's preoperative serum biomarker data, and the blood pressure load indicators are determined based on the patient's preoperative blood pressure monitoring data. The processing module is used to identify the affected side region of interest in the image data; to standardize and quantify the image data to obtain the cerebral blood flow time series pairs of each voxel in the region; wherein, the cerebral blood flow time series pair includes a first time series obtained from any voxel in the affected side region of interest, and a second time series obtained from the corresponding mirror anatomical position voxel in the contralateral cerebral hemisphere. The separation module is used to construct a difference matrix based on the differences between the cerebral blood flow time series pairs of each voxel in the region; search for the optimal matching path in the difference matrix; and separate the blood flow delay time, which characterizes the blood flow transmission delay, and the contrast agent retention degree, which characterizes the abnormal retention of contrast agent, based on the optimal matching path. The determination module is used to determine the vascular permeability of each voxel in the region based on blood flow delay time, contrast agent retention, and vascular wall condition indicators; and to determine the vascular tolerance pressure value of each voxel in the region based on preset theoretical tolerance pressure, vascular permeability, and blood pressure load indicators. The prediction module simulates the theoretical risk brain volume corresponding to multiple preset postoperative blood pressure target values ​​based on the vascular tolerance pressure values ​​of each voxel in the region, in order to predict the postoperative blood pressure that the patient can safely tolerate.

2. The preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data according to claim 1, characterized in that, The preoperative serum biomarker data includes specific protein concentration data, wherein the specific proteins include at least one biomarker selected from glial fibrillary acid protein and β-amyloid protein. The process for determining the vessel wall condition indicators includes: Obtain the concentration measurement value of at least one biomarker from the specific protein concentration data; At least one concentration measurement is mapped to a scalar value ranging from 0 to 1 as an indicator of blood vessel wall condition.

3. The preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data according to claim 1, characterized in that, The patient's preoperative blood pressure monitoring data includes a sequence of systolic blood pressure measurements over a specific preoperative period. The process for determining the blood pressure load index includes: Obtain the preset safe blood pressure baseline value and the time interval between adjacent measurement times in the systolic blood pressure measurement sequence; Iterate through each systolic blood pressure measurement in the sequence and determine whether it is greater than the safe blood pressure baseline. For each target measurement value that exceeds the safe blood pressure baseline, the difference between the target measurement value and the safe blood pressure baseline is calculated as the excess value; The blood pressure load index is obtained by summing the products of all excess values ​​and their corresponding time intervals.

4. The preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data according to claim 1, characterized in that, The standardization and quantization of the image data to obtain time-series pairs of cerebral blood flow for each voxel within the region includes: For each voxel in the region of interest on the affected side, the original curves of contrast agent concentration changes over time at each voxel are extracted; the original curves are then resampled onto a preset standard time axis using mathematical interpolation to obtain a time axis-aligned sequence for the affected side; simultaneously, the same operation is performed at the corresponding location in the contralateral cerebral hemisphere to obtain a time axis-aligned reference sequence. Check whether the peak concentration of the reference sequence is higher than the preset background noise threshold; if not, use a preset standard reference curve to replace the reference sequence. If so, divide each data point in the affected side sequence by its own peak concentration value to obtain the first concentration baseline value; at the same time, divide each data point in the reference sequence by its own peak concentration value to obtain the second concentration baseline value. Arrange the first concentration baseline values ​​in chronological order to obtain the first time series; arrange the second concentration baseline values ​​in chronological order to obtain the second time series; the two together form a voxel-based cerebral blood flow time series pair.

5. The preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data according to claim 4, characterized in that, The difference matrix construction process includes: Create a two-dimensional array with the number of rows equal to the length of the first time series and the number of columns equal to the length of the second time series, and set the initial value of all elements in the array to zero; Go through each time point in the first time series one by one and record it as the first time point; for the first time point, go through each time point in the second time series one by one and record it as the second time point; Read the first value of the first time series at the first time point, read the second value of the second time series at the second time point, and calculate the square of the difference between the first value and the second value as the squared difference. The squared differences are filled into a two-dimensional array with the first time point as the row index and the second time point as the column index. After traversing all time points in the first and second time series, the filled two-dimensional array is used as the difference matrix.

6. The preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data according to claim 5, characterized in that, The process of searching for the optimal matching path in the difference matrix includes: Set a maximum allowed time offset as a physiological constraint for the search; create a cumulative cost matrix with the same size as the difference matrix, and initialize all elements in the cumulative cost matrix to a special value representing infinity; Traverse each position of the cumulative cost matrix in order from the start point to the end point; for the current position, check whether it meets the physiological constraint condition, which indicates whether the absolute value of the difference between the row index and the column index represented by the current position does not exceed the maximum allowed time offset; if it does not meet the constraint condition, mark the current position as unreachable. If satisfied, select the best precursor point that satisfies the physiological constraints and has the minimum cumulative cost from the left, top, and upper left adjacent positions; Add the squared difference of the current position in the difference matrix to the cumulative cost of the best predecessor point to obtain the minimum cumulative cost of the current position and record it. Start backtracking from the endpoint; sequentially determine the cost source point for each location point, and set the source point as the previous location point on the path, while moving the location point to the source point; Continue this process until the starting position is reached; connect all the positions passed during the entire movement in sequence, and the resulting continuous trajectory is the optimal matching path.

7. The preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data according to claim 6, characterized in that, The process of separating the blood flow delay time (characterizing blood flow transmission delay) and the contrast agent retention degree (characterizing abnormal contrast agent retention) based on the optimal matching path includes: Read each location point recorded on the optimal matching path sequentially, extract the target row index and target column index of the location point at the same position in the difference matrix, and the corresponding squared difference; For each location point on the path, calculate the absolute difference between the target row index and the target column index as the time offset; calculate the arithmetic mean of the time offsets calculated for all location points on the path to obtain the average time offset. Multiply the average time offset by the time sampling interval of the image data to obtain the blood flow delay time; Calculate the arithmetic mean of the squared differences corresponding to all locations along the path, and use it as the average difference value; The contrast agent retention rate is obtained by performing a square root operation on the average difference value.

8. The preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data according to claim 1, characterized in that, The process of determining vascular permeability includes: For each voxel within the region, the blood flow delay time is divided by a preset normalized time constant to obtain a dimensionless ratio; the dimensionless ratio is then subjected to a hyperbolic tangent function operation to obtain the function output value; the function output value is then added to the number 1 to obtain the first sum; the first sum is then divided by the number 2 to obtain the final saturation coefficient. Multiply the preset delay gain coefficient by the saturation coefficient to obtain the first product value; add the number 1 to the first product value to obtain the delay amplification factor; The blood vessel wall condition index is added to a preset minimum positive compensation constant to obtain a second sum; the second sum is then subjected to a preset power mathematical operation to obtain a biochemical weighting coefficient. The vascular permeability of voxels within the region is obtained by multiplying the contrast agent retention rate, the delay amplification factor, and the biochemical weighting coefficient.

9. A preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data according to claim 8, characterized in that, The process for determining the vascular tolerance pressure value includes: For each voxel within the region, the vascular permeability is multiplied by a preset damage sensitivity coefficient to obtain an intermediate product value; the exponential function value of the negative intermediate product value with the natural constant as the base is calculated as the damage attenuation coefficient. Divide the blood pressure load index by the preset load reference value to obtain the dimensionless relative load ratio; perform a natural logarithmic operation on the sum of the number 1 and the relative load ratio to obtain the logarithmic value; calculate the product of the logarithmic value and the preset fatigue sensitivity coefficient as the second product value; calculate the difference between the number 1 and the second product value, and perform non-negative truncation on the difference as the initial fatigue attenuation coefficient. The initial fatigue attenuation coefficient is compared with the preset strength minimum coefficient, and the larger of the two values ​​is taken as the final historical fatigue attenuation coefficient. The vascular tolerance pressure value of the voxel in the region is obtained by multiplying the preset theoretical tolerance pressure value, the damage attenuation coefficient, and the historical fatigue attenuation coefficient.

10. A preoperative prediction system for cerebral hyperperfusion syndrome based on multimodal data according to claim 1, characterized in that, The process of obtaining the theoretical risk brain volume includes: Obtain the preset postoperative blood pressure target value to be simulated; Traverse all voxels within the affected side's area of ​​interest; for each postoperative blood pressure target, compare the vascular tolerance pressure value of the current voxel with the postoperative blood pressure target value. If the vascular tolerance pressure value is less than the postoperative blood pressure target value, the current voxel is marked as a high-risk voxel; otherwise, it is marked as a low-risk voxel. After all voxels in the affected side’s area of ​​concern have been compared, the total number of voxels marked as high-risk is counted. Multiply the total number by the actual physical volume represented by a single voxel in the image space to obtain the theoretical risk brain volume simulated under the postoperative blood pressure target value.

Images