A highland cerebral stroke early warning method and system based on image recognition

Abstract

The application discloses a kind of highland cerebral stroke early warning method and system based on image recognition, belong to medical image processing and intelligent diagnosis auxiliary field, the method includes: obtaining the first data set of subject;Based on second observation data, the first feature vector quantified from first observation data is adaptively corrected to environment;Based on the blood concentration state index contained in third observation data;First correction vector and marker correction concentration set are input into multidimensional nonlinear fusion processing, and comprehensive risk index is obtained, comprehensive risk index is positively correlated with the possibility that subject occurs cerebral stroke event;And risk grade label is output to early warning display terminal.The application can more accurately depict the clinical phenomenon that old brain injury patient is highly sensitive to acute hypoxia.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and intelligent diagnostic assistance, and in particular to a method and system for early warning of high-altitude stroke based on image recognition. Background Technology

[0002] Stroke is an acute neurological deficit caused by cerebrovascular disease. Its occurrence is closely related to pathological changes at multiple levels, including cerebral small vessel disease, degeneration of white matter connectivity, and damage to glial cells. Epidemiological studies have shown that people who are chronically exposed to high-altitude, low-pressure, and hypoxic environments have a significantly higher risk of stroke than those in plains areas. Early risk warning for this population is of significant clinical importance.

[0003] Current stroke risk assessment methods typically employ the following technical approach: automatically quantifying the imaging burden parameters of cerebral small vessel disease based on multimodal magnetic resonance imaging data, combining this with laboratory measurements of peripheral blood neurological injury markers, and then weighting the scores according to preset weights to output a risk level. However, directly applying this method to high-altitude populations presents the following technical limitations:

[0004] First, there is the problem of false positives in imaging assessment. In the hypoxic environment of high altitude, subjects exhibit compensatory responses such as increased cerebral blood flow, compensatory dilation of microvessels, and physiological expansion of perivascular spaces. These physiological changes appear highly similar to pathological small vessel disease damage on conventional MRI, making it difficult for traditional methods to distinguish between the two and easily misdiagnosing physiological compensation as pathological damage. Second, there is the problem of false elevations in biochemical markers. Long-term exposure to high altitudes leads to a large secretion of erythropoietin, resulting in an abnormally high hematocrit and a relative passive contraction of plasma volume. This causes the concentration of neurological damage markers dissolved in the plasma, and conventional measurements cannot accurately reflect the level of neuronal damage in the subjects. Third, there is the problem of insufficient early signal detection capability. Relying solely on macroscopic imaging lesions makes it difficult to capture early white matter micro-connectivity degradation signals in the subclinical stage before identifiable lesions have formed, leading to a delay in early warning.

[0005] Therefore, there is an urgent need for an efficient image recognition-based early warning method for high-altitude stroke. Summary of the Invention

[0006] One of the objectives of this invention is to provide an image recognition-based method for early warning of high-altitude stroke, in order to solve the problems of false positives in the assessment of cerebrovascular images of high-altitude populations, the loss cascade amplification effect of simple linear weighting of multi-source heterogeneous data, and the unavailability of the system in scenarios where some modal data is missing in the existing technology.

[0007] This invention is achieved through the following technical solution: a method for early warning of high-altitude stroke based on image recognition, comprising the following steps: acquiring pathological load parameters and environmental exposure data of the subject, wherein the pathological load parameters are parameters used to characterize the degree of pathological load obtained by processing image data of the target site of the subject, and the environmental exposure data are used to indicate the exposure intensity and cumulative exposure duration of the environment in which the subject is located; performing baseline correction on the pathological load parameters based on the environmental exposure data to obtain corrected pathological load parameters, wherein the baseline correction is used to subtract the physiological compensation benchmark corresponding to the exposure intensity and cumulative exposure duration from the pathological load parameters; determining a risk warning index based on the corrected pathological load parameters; comparing the risk warning index with a preset risk threshold to obtain and output the warning result.

[0008] Furthermore, the imaging data includes multimodal magnetic resonance imaging data acquired from the target site of the subject, and the pathological burden parameters include a first pathological component, a second pathological component, a third pathological component, and a fourth pathological component. The first pathological component, the second pathological component, the third pathological component, and the fourth pathological component are used to characterize lacunar lesion burden, white matter high signal burden, perivascular space burden, and microbleeding burden, respectively.

[0009] Furthermore, the exposure intensity is a discrete interval divided according to the intensity of the low-pressure hypoxic environment, and the cumulative exposure duration includes the native category and the relocated category, wherein the relocated category is further divided into at least one of multiple subcategories according to the relocation duration.

[0010] Further, baseline correction is performed on the pathological load parameters based on the environmental exposure data to obtain corrected pathological load parameters. This includes: using the exposure intensity and the cumulative exposure duration as query indexes, matching and retrieving the corresponding baseline pathological load parameters from the baseline parameter library, whereby the baseline pathological load parameters characterize the physiological compensation baseline corresponding to the query index; performing difference processing on the pathological load parameters and the baseline pathological load parameters to obtain difference results; and performing non-negative rectification on the difference results to obtain the corrected pathological load parameters. The baseline parameter library uses the exposure intensity and the cumulative exposure duration as a joint index to store the baseline pathological load parameters obtained from statistical analysis of individuals without stroke symptoms under different combinations of exposure intensity and cumulative exposure duration.

[0011] Further, the baseline pathological burden parameter is determined as follows: the exposure intensity is input into a stimulation function to obtain a compensatory driving force; the cumulative exposure duration is input into a time-adaptive factor to obtain a compensatory adjustment force; the compensatory driving force and the compensatory adjustment force are coupled term by term to obtain the baseline pathological burden parameter; the stimulation function is configured such that the compensatory driving force increases with the increase of the exposure intensity and tends to a saturation upper bound after the exposure intensity exceeds a preset threshold; the time-adaptive factor is configured such that the compensatory adjustment force decays with the increase of the cumulative exposure duration, and the decay includes acute compensatory components and chronic adaptive components with different decay rates.

[0012] Furthermore, the method for early warning of high-altitude stroke also includes: obtaining the network topology parameters of the subject.

[0013] The network topology parameters are obtained by performing topological measurements on the reconstructed structural connectivity network after processing the image data of the target area of ​​the subject. These parameters characterize the information transmission capability of the structural connectivity network. The network topology parameters include a first topology parameter and a second topology parameter. Determining the risk warning index based on the corrected pathological load parameter includes determining the risk warning index based on the corrected pathological load parameter and the network topology parameters.

[0014] Furthermore, the first topology parameter is used to characterize the global information transmission capability of the structured network, and increases as the cost of the shortest connection path between each pair of nodes in the structured network decreases; the second topology parameter is used to characterize the degree of local clustering of nodes in the structured network.

[0015] Furthermore, the structural connectivity network is reconstructed as follows: fiber tracing is performed on the diffusion tensor imaging data in the image data, the fiber tracing including at least one of continuous tracing and streamline iterative tracing; the target area of ​​the subject is divided into multiple network nodes according to a brain region segmentation template, the brain region segmentation template including at least one of various standardized partition maps; the fiber bundle connections detected between the network nodes are used as connection edges of the structural connectivity network, and a connection relationship matrix is ​​constructed using the connection weights of the connection edges; the connection weights are positively correlated with the number of fiber bundles contained in the corresponding connection edge, and negatively correlated with the average length of the fiber bundles contained in the connection edge.

[0016] Furthermore, the high-altitude stroke early warning method also includes: obtaining the concentrations of a first biomarker, a second biomarker, and blood concentration parameters of the subject; performing concentration correction on the first biomarker concentration and the second biomarker concentration according to the blood concentration parameters to obtain a first corrected concentration and a second corrected concentration; the step of determining the risk warning index based on the corrected pathological load parameter and the network topology parameter includes: determining the risk warning index based on the corrected pathological load parameter, the network topology parameter, the first corrected concentration, and the second corrected concentration.

[0017] Furthermore, the concentration correction includes: restoring the concentrations of the first marker and the second marker to their equivalent concentrations at a standard plasma volume, based on the relationship between the blood concentration parameters and a preset standard blood concentration benchmark; wherein, the higher the degree of blood concentration represented by the blood concentration parameters, the greater the upward adjustment corresponding to the restoration; the standard blood concentration benchmark includes at least one of a full-population statistical benchmark and a stratified population statistical benchmark.

[0018] Furthermore, the concentration correction also includes: when the blood concentration parameter exceeds a preset concentration threshold, applying attenuation compensation to the equivalent concentration that increases with the amount of excess; when the blood concentration parameter does not exceed the concentration threshold, the attenuation compensation does not change the equivalent concentration.

[0019] Further, determining the risk warning index includes: weighting and summing the corrected pathological load parameters according to a first weighting parameter to obtain an additive base quantity; determining a constraint weight based on the first corrected concentration and the second corrected concentration, and multiplicatively coupling the constraint weight with a microscopic degradation metric to obtain a synergistic amplification quantity, wherein the microscopic degradation metric increases as the first topological parameter decreases and as the second topological parameter decreases; and summing the additive base quantity with the synergistic amplification quantity to obtain the risk warning index.

[0020] Furthermore, the constraint weight is a weighted linear combination of the first correction concentration and the second correction concentration, or the product of the first correction concentration and the second correction concentration.

[0021] Further, the risk warning index is obtained by summing the additive basic quantity and the cooperative amplification quantity, which is replaced by: constructing a logarithmic cascade amplification coefficient based on the additive basic quantity, wherein the logarithmic cascade amplification coefficient increases with the increase of the additive basic quantity; amplifying the cooperative amplification quantity according to the logarithmic cascade amplification coefficient and superimposing it on the additive basic quantity to obtain the risk warning index; wherein the cooperative amplification quantity also includes a neuro-glial synchronous damage cooperative component and a structure-biochemical coupling component, wherein the structure-biochemical coupling component is determined based on a comprehensive network degradation factor, wherein the comprehensive network degradation factor increases with the increase of the degradation degree of the first topology parameter and the second topology parameter relative to their respective reference benchmarks.

[0022] Furthermore, the high-altitude stroke early warning method also includes: determining whether there are effective concentrations of the first biomarker and the second biomarker; if so, determining the risk warning index based on the corrected pathological load parameter, the network topology parameter, the first corrected concentration, and the second corrected concentration; if not, switching to a backup assessment path, wherein the backup assessment path determines the risk warning index solely based on the corrected pathological load parameter and the network topology parameter.

[0023] Furthermore, switching to the backup assessment path includes: selecting an assessment path corresponding to the current availability combination from multiple preset assessment paths based on the availability combination of the data availability of the corrected pathological load parameters, the data availability of the network topology parameters, and the data availability of the first correction concentration and the second correction concentration, in order to determine the risk warning index; wherein, when any parameter is unavailable, the assessment path sets the operation item corresponding to the unavailable parameter to zero or replaces it with a preset compensation item.

[0024] Furthermore, obtaining and outputting the early warning result includes: classifying the subject into a risk level of low risk, medium risk, or high risk based on the comparison result of the risk warning index relative to the first risk threshold and the second risk threshold; obtaining intervention prompt information corresponding to the classified risk level; and combining the risk level and the intervention prompt information and outputting them to the early warning display terminal, wherein the early warning display terminal includes at least one of a medical workstation, a portable diagnostic terminal, a mobile terminal, and a remote consultation terminal.

[0025] Another aspect of the present invention provides an image recognition-based early warning system for high-altitude stroke, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements any of the image recognition-based high-altitude stroke early warning methods described above.

[0026] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0027] 1. This invention introduces an image load benchmark correction mechanism that uses altitude exposure intensity and cumulative exposure duration as a joint index. By performing differential non-negative rectification between the actual image load of the subject and the physiological compensation benchmark under the corresponding environmental exposure conditions, the physiological compensation component of high altitude can be eliminated, thereby eliminating the false positive interpretation bias produced by traditional cerebral small vessel disease assessment methods in high-altitude populations.

[0028] 2. This invention introduces a dilution effect correction mechanism based on blood concentration indicators to restore the measured concentration of neurodegenerative markers to the equivalent concentration under standard plasma volume. This avoids misjudging the increase in nominal marker concentration caused by blood concentration as aggravation of neuronal damage, thus improving the accuracy of biochemical assessment. Furthermore, by introducing brain white matter structural network topology analysis and utilizing topological measures that reflect the microscopic connectivity and local compactness of the whole brain, it can capture early microscopic connectivity degradation signals in the subclinical stage where macroscopic lesions cannot be identified by conventional imaging, thereby improving the detection capability of ultra-early hypoxic injury.

[0029] 3. This invention introduces a multidimensional nonlinear fusion mechanism consisting of additive basic terms and multiplicative coupling terms, unifying macroscopic imaging load, microscopic network topology, and peripheral blood biochemical markers into the same risk assessment framework. This avoids the loss of cascade amplification effects caused by simple linear weighting, and more accurately characterizes the clinical phenomenon of patients with old brain injuries being highly sensitive to acute hypoxia. At the same time, by introducing an adaptive fault-tolerant switching mechanism based on modal availability, it automatically switches to the corresponding backup assessment path when some modal data is missing, avoiding the unavailability of the entire early warning system due to the lack of some detection data, and improving the robustness and consistency of the early warning system under various clinical conditions. Attached Figure Description

[0030] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0031] Figure 1 This is a flowchart of the method provided in Embodiment 1 of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0033] Example 1

[0034] This embodiment discloses a method for early warning of high-altitude stroke based on image recognition. Figure 1 The flowchart of the method in this embodiment is shown. As can be seen from the flowchart, this embodiment includes the following steps:

[0035] Step 1: Obtain the first dataset. The first dataset refers to a set of multi-source heterogeneous medical data used to fully describe the cerebrovascular pathological state and neurobiochemical damage level of subjects under high-altitude exposure conditions. Each sub-data in the first dataset must correspond to the same clinical assessment window in the time dimension to ensure the synchronization and fusion of multimodal data.

[0036] The first dataset may include: multimodal magnetic resonance imaging data of the subjects, altitude exposure parameters, and blood biomarker data.

[0037] Multimodal magnetic resonance imaging (MRI) data refers to a set of multidimensional medical images of the same subject's brain acquired using MRI equipment under different imaging sequences. This multimodal MRI data must be acquired continuously within the same scan cycle, and inter-slice registration and geometric alignment must be completed to ensure voxel-by-voxel correspondence between the different sequences in spatial location. For example, multimodal MRI data may include T1-weighted imaging (T1WI) sequences, T2-weighted imaging (T2WI) sequences, fluid attenuation inversion recovery (FLAIR) sequences, susceptibility-weighted imaging (SWI) sequences, and diffusion tensor imaging (DTI) sequences.

[0038] Altitude exposure parameters refer to a set of physiological-geographical parameters used to characterize the intensity of the low-pressure hypoxic environment at high altitudes where subjects are located and the cumulative duration of exposure. These altitude exposure parameters include the subject's current altitude gradient and the time spent at high altitudes.

[0039] Among them, the altitude gradient refers to the discrete altitude range defined based on the differences in the intensity of the human body's physiological compensatory response to low-pressure hypoxia environment. For example, the altitude gradient can be divided into the first gradient (below 2500 meters), the second gradient (2500 meters to 3500 meters), the third gradient (3500 meters to 4500 meters), and the fourth gradient (above 4500 meters).

[0040] High-altitude residence time refers to the cumulative length of time a subject stays in a high-altitude region and maintains metabolic-circulatory-nervous system adaptation. This residence time can be divided into two main categories: long-term residence and migration. Long-term residence indicates that the subject and their family have long resided in the high-altitude region and have completed intergenerational hereditary high-altitude adaptation. Migration indicates the stage of adaptation the subject is in after migrating from a low-altitude area to the high-altitude region. For example, migration can be further divided into subcategories based on duration: less than 1 year, 1 to 5 years, and more than 5 years.

[0041] Blood biomarker data refers to a set of molecular concentration parameters obtained by peripheral venous blood collection and laboratory biochemical detection methods, which can reflect the degree of neuronal damage, astrocyte activation, and blood concentration in a subject. These blood biomarker data include neurofilament light chain protein (NFL) concentration, glial fibrillary acidic protein (GFAP) concentration, and hematocrit (HCT) value.

[0042] Among them, the NFL concentration refers to the actual measured concentration of neurofilament light chain protein in plasma; the NFL concentration is used to characterize the level of axonal cytoskeleton protein released into the peripheral blood system after neuronal axonal structure is damaged, and is a sensitive molecular indicator of early neuronal axonal degeneration.

[0043] GFAP concentration refers to the actual measured concentration of glial fibrillary acidic protein in plasma. This GFAP concentration is used to characterize the level of glial cytoskeleton proteins released after astrocyte damage or activation, and is a sensitive molecular indicator of astrocyte response and changes in glial unit function.

[0044] The hematocrit (HCT) value refers to the proportion of red blood cells in the blood volume, usually expressed as a decimal. This HCT value is used to characterize the degree of compensatory increase in red blood cells and the relative level of blood concentration in subjects under hypoxic conditions at high altitudes.

[0045] In this embodiment, the subject's multimodal MRI data can be directly acquired by the magnetic resonance imaging equipment and transmitted to the data acquisition module of the early warning system via the DICOM protocol; altitude exposure parameters can be self-reported by the subject or automatically acquired through GPS positioning equipment and electronic record system; blood biomarker data can be measured by a laboratory biochemical analyzer and then accessed to the early warning system through the hospital information system (HIS).

[0046] By acquiring the aforementioned multi-source heterogeneous datasets encompassing imagery, geography, and biochemistry, a data foundation can be laid for subsequent altitude adaptive correction, network topology analysis, and multi-dimensional fusion, avoiding false positives in high-altitude assessments caused by single-modal evaluation.

[0047] Step 2: Process the first dataset to generate the second dataset. The second dataset refers to a set of derived feature data that can be directly used for risk warning determination, generated by performing operations such as altitude adaptive correction, dilution effect correction, network topology quantization, and multidimensional nonlinear fusion on the original multi-source data in the first dataset. This second dataset is based on a comprehensive stroke risk warning index. As the final output.

[0048] Specifically, this step may include the following sub-steps:

[0049] Sub-step 2.1: Altitude-adaptive CSVD image load baseline correction.

[0050] Cerebral small vessel disease (CSVD) imaging burden parameters refer to a set of characteristic parameters automatically identified and quantified from multimodal MRI data using medical image segmentation algorithms, used to characterize the severity of cerebral small vessel pathological burden in a subject. These CSVD imaging burden parameters constitute the subject's actual CSVD imaging burden parameter vector. .

[0051] In this embodiment, the CSVD image burden parameters may include the volume of lacunar lesions, the volume of high signal in white matter, the number of perivascular spaces, and the number of cerebral microbleeds.

[0052] Lacunar infarct volume refers to the total volume of small vessel occlusive infarcts located in the deep white matter or basal ganglia region, identified by combined threshold segmentation and morphological filtering of T1WI and FLAIR sequences. This lacunar infarct volume is used to characterize the cumulative burden of the subject's previous chronic small artery occlusive lesions.

[0053] The white matter high-signal volume refers to the total volume of non-specific high-signal lesions distributed in the periventricular and deep white matter regions of the brain, identified by automatic high-signal segmentation of FLAIR sequences. This white matter high-signal volume is used to characterize the extent of accumulation of chronic ischemic demyelinating lesions in the subject's white matter.

[0054] The number of perivascular spaces refers to the total number of dilated perivascular spaces along the course of perforating arteries identified through automated tubular structure detection on T2WI sequences. This number of perivascular spaces is used to characterize the degree of impaired interstitial fluid return and microvascular lymphatic drainage in the subject's brain tissue.

[0055] The number of cerebral microbleeds refers to the total number of hemosiderin deposits less than 10 mm in diameter identified by automatic punctate low-signal detection on SWI sequences. This number of cerebral microbleeds is used to characterize the cumulative burden of past microvascular rupture events in the subject.

[0056] In the hypoxic environment of high altitude and low pressure, subjects will exhibit a series of physiological compensatory responses of cerebrovascular system, such as compensatory increase in cerebral blood flow, compensatory dilation of microvessels, and physiological expansion of perivascular spaces. These physiological changes are highly similar to the imaging manifestations of pathological CSVD damage on conventional MRI images.

[0057] Understandably, by introducing an altitude-adaptive baseline correction step, it is possible to avoid misjudging the aforementioned physiological compensation as pathological damage when assessing the risk of stroke in high-altitude populations, thereby eliminating the false positive interpretation bias produced by the traditional CSVD assessment method in high-altitude populations.

[0058] In this embodiment, to achieve altitude-adaptive CSVD image load baseline correction, step S2.1 includes:

[0059] Obtain the actual CSVD image load parameter vector of the subject. ;

[0060] The altitude gradient and time spent living at high altitudes of the subjects were used as query indexes;

[0061] By matching and retrieving a pre-defined altitude-CSVD baseline database, the CSVD baseline feature vectors corresponding to the altitude gradient and the time spent living on the plateau are extracted. ;

[0062] The actual CSVD image load parameter vector Compared with CSVD baseline eigenvectors Perform vector difference;

[0063] The difference results are non-negative rectified to obtain the corrected CSVD pathological burden vector. .

[0064] The altitude-CSVD baseline database is a parametric database established through multimodal MRI scans and long-term clinical follow-up of asymptomatic stroke patients stratified by different altitude gradients and altitude residence times. It is used to characterize the physiological compensatory baseline of cerebral small vessel disease imaging features under various altitude-residence time combinations. This database uses altitude gradient and altitude residence time as a joint index and stores the baseline feature vectors of CSVD for each combination. The statistical mean and standard deviation.

[0065] CSVD baseline eigenvectors This refers to the column vector composed of the estimated physiological compensation baseline values ​​of four CSVD indicators—lacunar foci, white matter hyperintensity, perivascular space, and cerebral microbleeds—found through matching and retrieval, under a specific altitude-time of residence combination. It can be represented as: The values ​​represent, in order, the baseline values ​​for the volume of lacunar lesions, the baseline values ​​for the volume of white matter high signal, the baseline values ​​for the number of perivascular spaces, and the baseline values ​​for the number of cerebral microbleeds under a specific physiological compensation level.

[0066] For example, in this embodiment, the corrected CSVD pathological burden vector It can be calculated using the following formula:

[0067]

[0068] in, This is the corrected pathological burden vector for CSVD; The actual CSVD image load parameter vector for the subject; To match the retrieved CSVD baseline feature vectors from the elevation-CSVD baseline database; It is a rectification operator that takes non-negative values ​​for each element.

[0069] It should be noted that in the above formula, Operators act on difference vectors For each component, negative components are truncated to zero, while non-negative components retain their original values, resulting in a corrected pathological load vector. All components are in the non-negative real number field.

[0070] To analyze the baseline feature vector of CSVD The internal generation mechanism is parameterized and modeled to improve the extrapolability of the baseline database in sparse regions. In this embodiment, the CSVD baseline feature vector... It can also be dynamically calculated through the coupling of the altitude stimulus function vector and the residence time adaptive factor vector. The specific steps include:

[0071] Determine the actual altitude of the subjects With cumulative residence time ;

[0072] altitude Substitute the altitude stimulus function vector The physiological compensatory driving force of each CSVD index as altitude increases was calculated item by item.

[0073] Time of residence Substitute the residence time adaptive factor vector The compensatory attenuation adjustment of each CSVD indicator as it evolves with the length of residence is calculated item by item.

[0074] altitude stimulus function vector Adaptive factor vector of residence time Element-wise multiplication yields the CSVD baseline feature vector. .

[0075] The altitude stimulus function vector refers to a parameterized function vector used to characterize the physiological compensation of each CSVD indicator as a function of altitude. The design basis of this altitude stimulus function vector is that as altitude increases, the intensity of physiological compensation induced by hypoxia gradually increases, but due to the biological limits of human physiological functions, this increase will not continue indefinitely but will tend to saturate.

[0076] The residence time adaptive factor vector is a parameterized function vector used to characterize the biphasic decay regulation of physiological compensation quantities of various CSVD indicators as residence time at high altitude increases. The design basis of this residence time adaptive factor vector is that the human body gradually transitions from an acute compensatory response upon initial arrival at high altitude to a chronic adaptation state after long-term residence. The acute compensatory response is strong but decays rapidly, while the chronic adaptation process is relatively slow but lasts longer.

[0077] For example, in this embodiment, the CSVD baseline feature vector It can be calculated using the following formula:

[0078]

[0079] in, The altitude stimulus function vector; This is a vector of adaptive factors for residence time; The Hadamard product operator represents the element-wise multiplication operation between two vectors of the same dimension; The actual altitude at which the subject was currently exposed is shown in meters. The cumulative time spent by the subjects in high-altitude areas is expressed in years.

[0080] For example, in this embodiment, the altitude stimulus function vector Each component can be calculated using the following formula:

[0081]

[0082] in, The category index for CSVD indicators corresponds to lacunar lesions, white matter hyperintensities, perivascular spaces, and cerebral microbleeds, respectively. For the first The upper limit of the maximum physiological compensation of each indicator under extreme hypoxia; For the first The compensatory growth rate constant of each indicator; The altitude threshold for physiological compensation activation.

[0083] It should be noted that in the above formula, the denominator adopts the form of... The logical staging structure makes the function value in When it approaches zero, Time equals ,exist Time approaches the upper bound ; constant term in the numerator Located at the molecule position, it imposes amplitude constraints on the function value.

[0084] For example, in this embodiment, the residence time adaptive factor vector Each component can be calculated using the following formula:

[0085]

[0086] in, The weight of the acute compensatory phase component in the total compensation; This represents the decay rate constant during the acute compensatory phase. The decay rate constant during the chronic adaptation period; The cumulative time spent by the subjects in high-altitude areas is expressed in years.

[0087] It should be noted that in the above formula, the first term... With the second item It constitutes a biphasic exponential decay structure, in which Typically, the value is greater than This causes the first term to decay faster than the second term; when hour, ;when hour, .

[0088] Sub-step 2.2: Calculation of topological attribute parameters of brain white matter structural network.

[0089] The white matter structural network is a weighted undirected graph structure reconstructed from diffusion tensor imaging (DTI) data to characterize the anatomical connections between different brain regions via white matter fiber bundles. This white matter structural network uses brain regions as nodes, inter-brain fiber bundle connections as edges, and the strength of these connections as the weights of the edges.

[0090] Network topology attribute parameters refer to a set of topology measurement parameters calculated using graph theory methods based on the white matter structure network of the brain, used to quantitatively characterize the information transmission capability of the network at both global and local levels. These network topology attribute parameters constitute the network topology attribute vector. The terms represent, in turn, the global efficiency reflecting the microscopic connectivity of the whole brain and the local clustering coefficient reflecting the local density.

[0091] Understandably, by introducing topological analysis of brain white matter structural networks, early signs of white matter micro-connectivity degradation can be captured in the subclinical stage, where macroscopic lesions cannot be identified by conventional MRI, thereby improving the early warning system's ability to detect ultra-early hypoxic injury.

[0092] In this embodiment, in order to construct the brain white matter structural network and calculate its topological attribute parameters, sub-step 2.2 includes:

[0093] Deterministic fiber tracing of DTI data in multimodal MRI datasets;

[0094] Using an automated anatomical marker template as a reference, the subject's brain was divided into sections. Each brain region serves as a network node;

[0095] The fiber bundle connections between brain regions detected by the fiber tracing algorithm are used as the edges of the network;

[0096] A weighted adjacency matrix is ​​constructed using the fiber bundle connection density between nodes as the weight of the connecting edges. ;

[0097] Based on weighted adjacency matrix Calculate the global efficiency separately and local clustering coefficient This constitutes the network topology attribute vector. .

[0098] Deterministic fiber tracing refers to a fiber tract imaging method that reconstructs the trajectory of white matter fiber tracts by progressively integrating along the main diffusion direction of each voxel in DTI data. This deterministic fiber tracing can be implemented using the FACT algorithm or an iterative tracing algorithm based on streamlines.

[0099] An automated anatomical marker template is a standardized regional atlas used to divide the whole brain into several anatomically significant brain regions. This automated anatomical marker template can be the AAL90 brain region template, or it can be the AAL116 brain region template, Brainnetome246 brain region template, etc.

[0100] Weighted adjacency matrix This refers to the storage of connection weights between pairs of nodes in the brain's white matter structural network. A 3D square matrix, whose elements Represents a node With nodes Connection weights between The weighted adjacency matrix satisfies the symmetry constraint. and nonnegativity constraints .

[0101] For example, in this embodiment, the node With nodes Connection weights between It can be calculated using the following formula:

[0102]

[0103] in, For nodes With nodes The number of fiber bundles detected by the fiber tracking algorithm; The average length of these fiber bundles is given in millimeters.

[0104] It should be noted that in the above formula, Located at the molecular position, Located in the denominator, connecting the weights With the number of fiber bundles Positively correlated with average fiber bundle length Negative correlation.

[0105] To base on the weighted adjacency matrix Calculate the shortest topological path distance between node pairs. In this embodiment, the node... With nodes Shortest topological path distance between It can be calculated using the following formula:

[0106]

[0107] in, For the node To the node The set of all possible paths; For path Each edge Connection weights; This means taking the minimum sum of costs among all paths.

[0108] It should be noted that in the above formula, the path cost is the reciprocal of the edge weight. By accumulating the weights, the edge with the larger connection weight has a smaller cost.

[0109] For example, in this embodiment, global efficiency It can be calculated using the following formula:

[0110]

[0111] in, This represents the total number of network nodes. A set of network nodes; For nodes With nodes Shortest topological path distance between; summation Iterate through all ordered node pairs.

[0112] It should be noted that in the above formula, The inverse of the shortest path distance between node pairs, with pre-normalized coefficients. Located outside the summation, The range of values ​​for is non-negative real numbers.

[0113] For example, in this embodiment, the local clustering coefficient It can be calculated using the following formula:

[0114]

[0115] in, , , They are nodes , , The weights of the pairwise connections between the three; The geometric mean of the three connection weights; For nodes The degree, that is, the degree with the node The number of other nodes with valid connections can be determined by... The calculation yielded, where It is a symbolic function.

[0116] It should be noted that in the above formula, the molecule is represented by nodes. The sum of the weighted strengths of all triangle structures at the vertices, with the denominator being... For nodes The maximum possible number of triangle structures in the fully connected case; outer layer The coefficients are calculated as the arithmetic mean of the local clustering coefficients of all nodes.

[0117] Sub-step 2.3: Correction for the dilution effect of blood biomarkers.

[0118] Blood biomarker dilution effect correction refers to the process of adjusting the measured values ​​of NFL and GFAP concentrations in plasma using the subject's actual hematocrit (HCT) value, restoring them to equivalent concentrations at a standard plasma volume. The design of this correction process is based on the fact that prolonged exposure to high-altitude hypoxic environments stimulates the secretion of large amounts of erythropoietin (EPO), leading to an abnormally high hematocrit in subjects, and in some cases, even high-altitude polycythemia. This results in a relative passive contraction of plasma volume, passively concentrating neurodegenerative markers dissolved in the plasma.

[0119] Understandably, by introducing a dilution effect correction step for HCT values, it is possible to avoid misjudging the increase in nominal concentration of markers caused by blood concentration effect as an aggravation of substantial neurological damage in the brain when assessing the level of neurological damage in high-altitude populations.

[0120] In this embodiment, in order to achieve dilution effect correction of blood biomarkers, step S2.3 includes:

[0121] Obtain the actual measured NFL concentration of the subjects. With GFAP concentration ;

[0122] Obtain the actual hematocrit value of the subject ;

[0123] Obtain the preset plain-standardized hematocrit constant. ;

[0124] Substituting the above parameters into the correction formula, the corrected NFL concentration values ​​were calculated respectively. and GFAP concentration value .

[0125] in, This refers to the raw concentration value obtained by directly measuring the NFL content in the peripheral plasma of a subject using a laboratory biochemical analyzer.

[0126] This refers to the raw concentration value obtained by directly measuring the GFAP content in the peripheral plasma of a subject using a laboratory biochemical analyzer.

[0127] This refers to the actual hematocrit value measured in the subject, expressed as a decimal.

[0128] This refers to the preset baseline constant for hematocrit in a normal plain population. It can be expressed as the statistical mean of the entire population, or as the mean of a subgroup stratified by gender and age.

[0129] For example, in this embodiment, the corrected NFL concentration value It can be calculated using the following formula:

[0130]

[0131] Similarly, the corrected GFAP concentration value It can be calculated using the following formula:

[0132]

[0133] in, , These represent the actual NFL and GFAP concentrations measured in the subjects' blood, respectively. The actual hematocrit value of the subject; The pre-defined plain-standardized hematocrit constant; This indicates the percentage of the subject's plasma volume in whole blood; This indicates the percentage of plasma volume in whole blood in a normal population from the plains.

[0134] It should be noted that in the above formula, Located at the molecular position, Located in the denominator position, when The ratio is less than 1 at that time.

[0135] In this embodiment, to further characterize the current The nonlinear effect of microcirculatory stasis after significantly exceeding the physiological threshold on the marker clearance rate, and the corrected NFL concentration value. With GFAP concentration value A microcirculation stagnation compensation item can also be introduced.

[0136] For example, in this embodiment, the corrected NFL concentration after introducing a microcirculation stagnation compensation term. It can be calculated using the following formula:

[0137]

[0138]

[0139] Similarly, the corrected GFAP concentration after introducing a microcirculation stagnation compensation term. It can be calculated using the following formula:

[0140]

[0141]

[0142] in, The critical hematocrit threshold is used to distinguish between physiological compensation at high altitudes and pathological microcirculatory stasis. The microcirculation stagnation index; This is a non-negative rectification operator. In the above equation, when... hour, The function outputs zero, and the exponent term is zero. This compensation term does not produce a numerical correction to the original concentration; when At that time, the exponent term is between 0 and 1.

[0143] Sub-step 2.4: Nonlinear fusion of multidimensional heterogeneous data and calculation of risk warning index.

[0144] Comprehensive Stroke Risk Warning Index This refers to the corrected CSVD pathological burden vector Network topology attribute vector and corrected blood biomarker concentrations , A scalar index obtained through multidimensional nonlinear fusion computation, used to comprehensively characterize the risk of altitude-related stroke in subjects. This comprehensive stroke risk warning index The higher the value of , the higher the likelihood of a subject experiencing a stroke. By introducing a multidimensional nonlinear fusion mechanism, the previously fragmented macroscopic imaging load, microscopic network topology, and peripheral blood biochemical markers can be unified under the same risk assessment framework, avoiding the loss of cascade amplification effects caused by simple linear weighting.

[0145] In this embodiment, the specific steps of multidimensional nonlinear fusion may include:

[0146] Network topology attribute parameters and Risk index is introduced as a principal component feature in the calculation;

[0147] Corrected NFL concentration and GFAP concentration It is introduced as a constraint weighting factor in the calculation of the risk index;

[0148] Corrected CSVD pathological burden vector By pre-setting weight vector Weighted summation yields the additive basic term of the macroscopic pathological burden;

[0149] By nonlinearly coupling the principal component features with the constraint weighting factor, a micro-biochemical synergistic amplification term is obtained;

[0150] Summing the additive basic terms and the micro-biochemical synergistic amplification terms yields the comprehensive stroke risk warning index. .

[0151] Among them, the preset weight vector This refers to a pre-labeled weighted vector based on the differences in the contribution of various CSVD pathological features to stroke events in retrospective clinical studies, which can be represented as: The components are, in order, the weighting coefficients of four pathological burdens: lacunar foci, white matter hyperintensity, perivascular space, and cerebral microbleeds. This is the transpose symbol for a matrix.

[0152] The constraint weighting factor, composed of the corrected NFL and GFAP concentrations, is used to adjust the magnitude of the micro-risk level reflected by network topology degradation. This constraint weighting factor can be expressed as a weighted linear combination of the two concentrations or as a product of the two concentrations.

[0153] For example, in this embodiment, the comprehensive stroke risk warning index It can be calculated using the following formula:

[0154]

[0155] in, This is an additive basic term, representing the weighted sum of the pathological burdens of each item in CSVD; The preset weight vector for each pathological burden of CSVD; This is the corrected pathological burden vector for CSVD. This is the sensitivity adjustment coefficient for the corrected NFL concentration; This is the sensitivity adjustment coefficient for the corrected GFAP concentration; For the global efficiency of the brain's white matter structural network; The local clustering coefficients of the brain's white matter structural network; This is the reciprocal of the overall efficiency. This is a complementary term to the local clustering coefficients.

[0156] It should be noted that in the above formula, middle Located in the denominator position, Item It appears in the form of a subtraction term, with the two terms added together to form the micro-degradation measure term within the square brackets; The relationship between the item in square brackets and the item in square brackets is multiplicative.

[0157] In this embodiment, to further characterize the cascading amplification effect between macroscopic pathological burden and microscopic-biochemical risk, a comprehensive stroke risk warning index is used. An extended fusion form combining logarithmic cascade amplification coefficients and synergistic damage terms can also be introduced.

[0158] For example, in this embodiment, the comprehensive risk warning index after introducing a logarithmic concatenation amplification factor is used. It can be calculated using the following formula:

[0159]

[0160] in, The macroscopic pathological damage weight vector can be represented as: ; Weights of the synergistic factors for synchronous neuro-glial injury; Weights for the structure-biochemical coupling factor; This refers to the microscopic degradation factors of the integrated network.

[0161] in, It can be calculated using the following formula:

[0162]

[0163] in, The global efficiency mean of the white matter network in a healthy control group matched for age and sex is used as a reference. The weighted clustering coefficients of the white matter network in a healthy control group matched for age and sex are the reference mean. and This is the topology sensitivity scaling factor.

[0164] It should be noted that in the above formula, pass The formal construction, when two terms When all function outputs are zero ; and The scaling factors, which serve as the global efficiency degradation term and the clustering coefficient degradation term respectively, are located inside the exponential function.

[0165] Step 3: Output the early warning results; the early warning results refer to the comprehensive stroke risk early warning index. The discretized risk level label, obtained by comparing the risk level with a preset safety risk threshold, is used to directly guide clinical intervention decisions. The warning result is output and displayed to the user in the form of a risk level string and its corresponding clinical intervention recommendations.

[0166] Safety risk threshold refers to a pre-defined index used to classify the comprehensive risk of stroke. A set of threshold parameters with a range of values. This security risk threshold includes a first risk threshold. Second risk threshold And satisfy .

[0167] In this embodiment, the specific steps for outputting the warning level may include:

[0168] Integrating the stroke risk warning index Respectively with the first risk threshold Second risk threshold Perform numerical comparisons;

[0169] Based on the comparison results, the subjects were classified into one of low-risk, medium-risk, or high-risk categories.

[0170] Obtain clinical intervention recommendations corresponding to the risk level;

[0171] The risk level and clinical intervention recommendations are combined and output to the early warning display terminal.

[0172] Among them, the early warning display terminal can be a medical workstation display screen, a portable diagnostic instrument display screen, a mobile terminal APP interface, a hospital information system prompt window, and a remote consultation terminal, etc.

[0173] For example, in this embodiment, the output logic for the warning level is as follows:

[0174] when At that time, output 'low risk, maintain physiological compensation observation';

[0175] when At that time, the output was 'medium risk, indicating the presence of subclinical microscopic hypoxia injury, and recommending high-altitude deacclimatization protective intervention';

[0176] when When the time comes, the system outputs "High risk, indicating the presence of organic CSVD pathological lesions, and activates the red clinical warning for stroke".

[0177] Understandably, discretizing the continuous early warning index into three-level risk labels can provide clinicians with a direct and actionable basis for intervention decisions, avoiding delays in treatment caused by the ambiguity of the clinical interpretation of continuous values.

[0178] Adaptive fault-tolerant processing: In clinical emergency or large-scale screening scenarios, subjects may experience partial modality data loss due to implanted devices preventing DTI sequence scanning or the absence of blood tests in the emergency room. To ensure that the early warning algorithm can still output usable and consistent risk assessment results under incomplete data conditions, this embodiment also includes an adaptive fault-tolerant processing step for missing blood biomarker data. This adaptive fault-tolerant processing step may include:

[0179] Determine whether valid blood biomarker data exist in the first dataset;

[0180] If the judgment result is yes, then calculate the comprehensive stroke risk warning index according to the multidimensional nonlinear fusion formula in sub-step 2.4. ;

[0181] If the judgment result is negative, the fusion calculation logic in sub-step 2.4 will be switched to the backup image single-dimensional evaluation path.

[0182] The early warning index calculated from the switched backup image single-dimensional evaluation path Step 3 is used to output the warning level.

[0183] Among them, the alternative imaging single-dimensional assessment path refers to a simplified risk assessment formula constructed using only the macroscopic CSVD pathological burden vector and brain white matter structural network topology parameters when blood biomarker data are missing.

[0184] For example, in this embodiment, the comprehensive stroke risk warning index under the backup image unidimensional assessment path It can be calculated using the following formula:

[0185]

[0186] in, This is a pre-defined weight vector for the pathological burdens of each item in CSVD. This is the corrected pathological burden vector for CSVD. Image topology feature compensation factor; For overall efficiency; represents the local clustering coefficient. In the above formula, Replaced the complete formula The corresponding position serves as the image independent compensation coefficient when biochemical markers are missing; the mathematical structure of the remaining terms remains consistent with the complete formula.

[0187] In this embodiment, the adaptive fault-tolerant processing steps can also be extended to a more general multimodal availability combination scenario, that is, simultaneously considering any subset of availability of the three modalities: conventional MRI data, DTI data, and blood biomarker data.

[0188] For example, in this embodiment, a binary vector of availability is defined. ,in This indicates the availability of conventional MRI and CSVD computational data. Indicates the availability of DTI topology calculation data. This indicates the availability of blood biomarker detection data. A value of 1 indicates that the data is complete, and a value of 0 indicates that the data is missing.

[0189] Adaptive Early Warning Index It can be calculated using the following formula:

[0190]

[0191] Among them, macro-micro amplification factor It can be calculated using the following formula:

[0192]

[0193] Microscopic biochemical fusion project It can be calculated using the following formula:

[0194]

[0195]

[0196]

[0197]

[0198] in, This is a Kronecker indicator function, which takes the value 1 when the condition in parentheses is true and 0 otherwise. For independent risk mapping coefficients of simple network topology degradation; For independent risk mapping coefficients of simple biochemical markers; For logical AND operator.

[0199] It should be noted that in the above formula, the three The condition corresponding to the function , , exist The four possible combinations of values ​​are mutually exclusive; when Three The function outputs zero. The total is zero; middle The position of the coefficient before the logarithmic term, when hour .

[0200] Understandably, by introducing an adaptive fault-tolerant processing mechanism, the situation where the entire early warning system becomes unusable due to the lack of some test data in clinical emergency or screening scenarios can be avoided, thereby improving the robustness and consistency of the early warning system under various clinical conditions.

[0201] Example 2

[0202] This embodiment discloses an image recognition-based early warning system for high-altitude stroke.

[0203] Specifically, the image recognition-based high-altitude stroke early warning system can be integrated into electronic devices, such as terminals, servers, medical imaging workstations, hospital information servers, regional telemedicine platforms, and mobile screening terminals. The terminals can be tablets, laptops, personal computers, portable medical image analysis terminals, etc.; the servers can be single servers or server clusters composed of multiple servers.

[0204] In this embodiment, the image recognition-based high-altitude stroke early warning system can also be integrated into multiple electronic devices. For example, the image recognition-based high-altitude stroke early warning system can be integrated into the hospital image archiving and communication system PACS server, the laboratory information system LIS server, and the clinical decision support server, with multiple servers working together to implement the method in embodiment 1 of this application.

[0205] In this embodiment, the server can also be implemented in the form of a terminal; for example, in plateau primary medical institutions, mobile medical examination vehicles, border patrol stations or medical stations in high-altitude mining areas, multimodal MRI image processing, biochemical marker correction and risk warning output can be completed through locally deployed image analysis workstations.

[0206] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for early warning of high-altitude stroke based on image recognition, characterized in that, The high-altitude stroke early warning method includes: The pathological burden parameters and environmental exposure data of the subjects were obtained. The pathological burden parameters were obtained by processing the imaging data of the target sites of the subjects and were used to characterize the degree of pathological burden. The environmental exposure data were used to indicate the exposure intensity and cumulative exposure duration of the environment in which the subjects were located. Based on the environmental exposure data, the pathological load parameters are baseline-corrected to obtain corrected pathological load parameters. The baseline correction is used to subtract the physiological compensation benchmark corresponding to the exposure intensity and cumulative exposure duration from the pathological load parameters. The risk warning index is determined based on the corrected pathological load parameters; The risk warning index is compared with a preset risk threshold to obtain and output the warning result.

2. The image recognition-based early warning method for high-altitude stroke according to claim 1, characterized in that, The step of baseline correction of the pathological load parameters based on the environmental exposure data to obtain corrected pathological load parameters includes: Using the exposure intensity and the cumulative exposure duration as query indexes, the corresponding baseline pathological load parameters are retrieved from the baseline parameter library. The baseline pathological load parameters are used to characterize the physiological compensation baseline corresponding to the query index. The pathological load parameters are compared with the baseline pathological load parameters to obtain the difference results. The difference results are non-negative rectified to obtain the corrected pathological load parameters; The baseline parameter library uses the exposure intensity and the cumulative exposure duration as a joint index to store the baseline pathological burden parameters obtained from populations without stroke symptoms under different combinations of exposure intensity and cumulative exposure duration.

3. The image recognition-based early warning method for high-altitude stroke according to claim 2, characterized in that, The baseline pathological burden parameters are determined in the following manner: The exposure intensity is input into the stimulus function to obtain the compensatory driving force; Input the cumulative exposure duration into the time adaptive factor to obtain the compensatory adjustment amount; The compensatory driving force and the compensatory adjustment force are coupled item by item to obtain the baseline pathological load parameters; The stimulation function is configured such that the compensatory driving amount increases with the increase of the exposure intensity, and tends to the upper bound of saturation after the exposure intensity exceeds a preset threshold; the time adaptive factor is configured such that the compensatory adjustment amount decreases with the increase of the cumulative exposure duration, and the decrease includes acute compensatory components and chronic adaptive components with different decrease rates.

4. The image recognition-based early warning method for high-altitude stroke according to claim 1, characterized in that, The high-altitude stroke early warning method further includes: obtaining the network topology parameters of the subject. The network topology parameters are obtained by performing topological measurements on the reconstructed structural connectivity network after processing the image data of the target part of the subject. They are parameters used to characterize the information transmission capability of the structural connectivity network. The network topology parameters include a first topology parameter and a second topology parameter. The step of determining the risk warning index based on the corrected pathological burden parameters includes: The risk warning index is determined based on the corrected pathological load parameters and the network topology parameters.

5. The image recognition-based early warning method for high-altitude stroke according to claim 4, characterized in that, The structural connection network is reconstructed in the following manner: Fiber tracing is performed on the diffusion tensor imaging data in the image data, and the fiber tracing includes at least one of continuous tracing and streamline iterative tracing. The target region of the subject is divided into multiple network nodes according to a brain region segmentation template, which includes at least one of a variety of standardized partition maps. The fiber bundle connections detected between the network nodes are used as the connection edges of the structural connection network, and a connection relationship matrix is ​​constructed using the connection weights of the connection edges. The connection weight is positively correlated with the number of fiber bundles contained in the corresponding connection edge, and negatively correlated with the average length of the fiber bundles contained in the connection edge.

6. The image recognition-based early warning method for high-altitude stroke according to claim 4, characterized in that, The high-altitude stroke early warning method also includes: Obtain the concentrations of the first biomarker, the second biomarker, and blood concentration parameters of the subjects; The concentrations of the first biomarker and the second biomarker are corrected according to the blood concentration parameters to obtain the first corrected concentration and the second corrected concentration. The step of determining the risk warning index based on the corrected pathological load parameters and the network topology parameters includes: The risk warning index is determined based on the corrected pathological load parameters, the network topology parameters, the first corrected concentration, and the second corrected concentration.

7. The image recognition-based early warning method for high-altitude stroke according to any one of claims 1 to 6, characterized in that, The high-altitude stroke early warning method also includes: Determine whether there are valid concentrations of the first biomarker and the second biomarker; If present, the risk warning index is determined based on the corrected pathological load parameter, the network topology parameter, the first correction concentration, and the second correction concentration. If not, the system switches to the backup assessment path, which determines the risk warning index solely based on the corrected pathological load parameters and the network topology parameters.

8. The image recognition-based early warning method for high-altitude stroke according to claim 7, characterized in that, The switch to the backup evaluation path includes: Based on the availability of the data for the corrected pathological load parameters, the availability of the data for the network topology parameters, and the availability of the data for the first and second corrected concentrations, an assessment path corresponding to the current availability combination is selected from multiple preset assessment paths to determine the risk warning index. When any parameter is unavailable, the evaluation path sets the operation term corresponding to the unavailable parameter to zero or replaces it with a preset compensation term.

9. The image recognition-based early warning method for high-altitude stroke according to claim 1, characterized in that, The process of obtaining and outputting the early warning results includes: Based on the comparison results of the risk warning index relative to the first risk threshold and the second risk threshold, the subjects are divided into one of the risk levels: low risk, medium risk, and high risk. The system acquires intervention prompt information corresponding to the classified risk level, and combines the risk level with the intervention prompt information to output to the early warning display terminal. The early warning display terminal includes at least one of a medical workstation, a portable diagnostic terminal, a mobile terminal, and a remote consultation terminal.

10. A high-altitude stroke early warning system based on image recognition, characterized in that, The high-altitude stroke early warning system includes: processor; The memory stores a computer program that, when executed by a processor, implements the image recognition-based early warning method for high-altitude stroke as described in any one of claims 1 to 9.

Images