A system for predicting prognosis of craniocerebral trauma based on a three-dimensional model

Abstract

This invention relates to the field of neurotrauma prognostic image analysis technology, and discloses a three-dimensional model-based system for predicting and analyzing the prognosis of craniocerebral trauma. The system involves multi-scale segmentation and topology construction of three-dimensional images of the patient's brain, tracking the morphological evolution rate of the traumatic region, and combining this with edema signal change curves to accurately identify key signal events in the trauma progression process. Furthermore, when these signal events occur, the system quantifies and analyzes the dynamic deviation of the integrity of fiber bundles associated with normal brain tissue, thereby generating damage propagation paths and mapping them to functional network nodes in a standard brain atlas, ultimately identifying key prognostic brain networks. This technical solution achieves the capture of key events and prospective prediction of pathways in the dynamic propagation process of secondary injuries after craniocerebral trauma, improving the accuracy of prognostic assessment.

Description

Technical Field

[0001] This invention relates to the field of neurotrauma prognostic image analysis technology, specifically a craniocerebral trauma prognostic prediction and analysis system based on a three-dimensional model. Background Technology

[0002] Currently, prognostic assessment of patients with traumatic brain injury relies on sequential observation of computed tomography (CT) or magnetic resonance imaging (MRI) images. Existing technologies can separately measure the volume of the traumatic core and quantify the signal intensity of the surrounding edema area. However, these methods typically analyze morphological changes and imaging signal variations independently, providing only isolated, descriptive indicators. Due to the lack of analysis of the dynamic correlations between multiple parameters, it is difficult to accurately identify key time points reflecting pathophysiological transitions in trauma from complex temporal changes, resulting in delayed and inaccurate capture of critical events in the progression of the disease.

[0003] In assessing the impact of trauma on brain networks, conventional techniques employ diffusion tensor imaging (DTI) to statically evaluate the damage status of local or global white matter fiber tracts at specific time points. These methods either perform simple pre- and post-injury comparisons or focus only on areas where significant structural changes have occurred. Their limitations include the possibility that the analysis time point may not match the dynamic activity phase of the trauma, and the assessment criteria are often based on absolute damage values ​​of fiber tracts, failing to effectively reveal how damage propagates early and covertly along intact structural connectivity networks, thus hindering early warning of secondary damage. Summary of the Invention

[0004] The purpose of this invention is to provide a three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury, the system comprising:

[0006] The image segmentation and topology construction module is used to perform multi-scale segmentation of the three-dimensional images of the patient's brain, identify the core trauma region, edema region and surrounding normal brain tissue region, and construct a spatial topology connection map between each region.

[0007] The trauma evolution analysis module is used to track the volume, shape parameters and diffusion trend of the trauma core region at different time points based on the spatial topology connection map, and to calculate the morphological evolution rate of the trauma region in continuous time intervals.

[0008] The signal event recognition module is used to identify the inflection point of edema signal change based on the morphological evolution rate and the signal intensity change curve of the edema region within the same time interval, and to mark the signal event that matches the mutation point of the trauma core evolution rate.

[0009] The structural connectivity analysis module is used to extract diffusion tensor imaging data of nerve fiber bundles associated with normal brain tissue regions when the signal event occurs, quantify the degree of deviation of fiber bundle integrity, and generate damage propagation paths based on structural connectivity.

[0010] The functional network mapping module is used to map functional network nodes to standard brain atlases based on the damage propagation path based on structural connectivity, identify the set of functional network nodes affected by the damage propagation path, and generate key prognostic brain network identifiers.

[0011] Preferably, the spatial topology connectivity graph includes a region adjacency matrix, boundary contact area, and centroid distance vector; the morphological evolution rate includes volume expansion acceleration, surface irregularity growth rate, and vector direction of invasion into surrounding regions; the signal event includes signal intensity peak time, signal attenuation start time, and signal plateau duration; the damage propagation path based on structural connectivity includes a list of damaged fiber bundle numbers, integrity deviation value, and path hop count; and the prognostic key brain network identifier includes the affected node number, the functional network type to which the node belongs, and the estimated loss of connectivity strength within the network.

[0012] Preferably, the image segmentation and topology construction module performs multi-scale segmentation on the three-dimensional image of the patient's brain, identifies the core trauma region, edema region, and surrounding normal brain tissue region, and constructs a spatial topological connectivity map between the regions, specifically including:

[0013] Load the three-dimensional medical image sequence of the brain and use adaptive threshold segmentation and region growing algorithm to separate the voxel set of the core trauma region;

[0014] Around the voxel set in the core region of the trauma, a level set evolution profile based on signal intensity is applied to delineate the continuous boundary of the edema region.

[0015] The area of ​​normal brain tissue is defined as the area extending outward from the continuous boundary of the edema region to the tissues in which the image gray value is within the normal physiological range.

[0016] The spatial relationship between the trauma core region, the edema region and the surrounding normal brain tissue region is calculated. The region adjacency relationship is established with the region centroid as the reference point. The boundary contact area is calculated with the shared voxel surface, and a topological connection map describing the spatial relationship between the regions is generated.

[0017] Preferably, the trauma evolution analysis module, based on the spatial topological connectivity map, tracks the volume, shape parameters, and diffusion trend of the trauma core region at different time points, and calculates the morphological evolution rate of the trauma region over continuous time intervals, specifically including:

[0018] At multiple consecutive data acquisition time points, the total volume, surface area, and sphericity of the trauma core region were calculated respectively.

[0019] Based on the total volume, surface area and sphericity values ​​of the trauma core region at different time points, calculate the changes in each parameter between adjacent time points;

[0020] Based on the adjacency relationship between the trauma core region and the surrounding region in the aforementioned spatial topology connection diagram, the main spatial orientation of the trauma region volume growth is analyzed to determine the diffusion trend vector;

[0021] By dividing the changes and diffusion trend vectors of the aforementioned parameters by the corresponding time interval, a set of morphological evolution rates reflecting the dynamic changes of the trauma area is obtained.

[0022] Preferably, the signal event recognition module identifies inflection points in edema signal changes based on the morphological evolution rate and the signal intensity change curve of the edema region within the same time interval, and marks signal events that match the abrupt change points in the evolution rate of the trauma core, specifically including:

[0023] The average signal intensity value of the edema region at continuous time points is extracted to form an edema signal intensity curve that changes with time.

[0024] The edema signal intensity curve is smoothed and differentiated to locate the zero-crossing point of the first derivative of the signal intensity, which is identified as the inflection point of the signal change.

[0025] Compare the time point at which the inflection point of the signal change occurs with the time when the rate value in the morphological evolution rate set undergoes a sudden change;

[0026] Matching inflection points with time deviations less than a preset tolerance with abrupt change times, and marking the matching pairs as signal events with spatiotemporal correlation.

[0027] Preferably, the structural connectivity analysis module extracts diffusion tensor imaging data of neural fiber tracts associated with normal brain tissue regions when the signal event occurs, quantifies the degree of deviation from fiber tract integrity, and generates a damage propagation path based on structural connectivity, specifically including:

[0028] At the time point marked by the signal event, diffusion tensor imaging data of normal brain tissue regions that are spatially adjacent to the edema region and the core trauma region are acquired;

[0029] The main nerve fiber bundles passing through the normal brain tissue region are reconstructed from the diffusion tensor imaging data, and the anisotropy fraction and mean diffusion rate of each fiber bundle are calculated.

[0030] The calculated anisotropy score and average diffusion rate are compared with the corresponding values ​​in the standard template database of healthy individuals to calculate the deviation.

[0031] The fiber bundles are sorted according to the degree of deviation, and fiber bundles with deviation exceeding the threshold are selected and connected in series according to their spatial connection relationship to form a damage propagation path that describes the possible spread of damage along the white matter fiber pathway.

[0032] Preferably, the functional network mapping module maps the damage propagation path based on structural connectivity to functional network nodes in a standard brain atlas, identifies the set of functional network nodes affected by the damage propagation path, and generates key prognostic brain network identifiers, specifically including:

[0033] The fiber bundles contained in the damage propagation path based on structural connectivity are spatially registered with the standard white matter fiber atlas.

[0034] Identify the cerebral cortex or deep nucleus regions connected by the fiber bundles, map the cerebral cortex or deep nucleus regions to a standard brain functional network atlas, and obtain the corresponding functional network node numbers;

[0035] Collect all affected functional network node numbers, and form a set of functional network nodes after deduplication;

[0036] Query the preset functional network type to which each node in the set of functional network nodes belongs, estimate the connection strength loss ratio based on historical data, and generate a prognostic key brain network identifier.

[0037] Preferably, the loading of the three-dimensional medical image sequence of the cranium, and the separation of the voxel set of the core trauma region using an adaptive threshold segmentation and region growing algorithm, specifically includes:

[0038] Read the grayscale value data of each voxel in a three-dimensional medical image sequence of the brain;

[0039] A gray-level histogram is generated based on the gray-level value distribution of the entire image sequence, and an adaptive segmentation threshold is dynamically calculated based on the peak and valley values ​​of the histogram.

[0040] Using voxels with gray values ​​higher than the adaptive segmentation threshold as initial seed points, the region growing algorithm iteratively checks the adjacent voxels of each seed point. If the difference between the gray value of an adjacent voxel and the gray value of the seed point is less than a preset similarity tolerance, the adjacent voxels are merged into the current region.

[0041] Repeat the iterative process until no new voxels are added, and obtain the set of connected voxels in the core region of the trauma.

[0042] Preferably, the step of calculating the changes in each parameter between adjacent time points based on the total volume, surface area, and sphericity values ​​of the trauma core region at different time points specifically includes:

[0043] For each acquisition time point, the total volume is obtained by multiplying the total number of all voxels in the core trauma region by the volume of a single voxel.

[0044] The sum of the outer surface areas of the boundary voxels of the trauma core region is calculated as the surface area;

[0045] The sphericity index is calculated based on the volume and surface area of ​​the core trauma region, and is expressed as the ratio of the region's volume to the surface area of ​​a sphere of the same volume.

[0046] The difference in total volume between adjacent time points is extracted as the volume change, the difference in surface area is extracted as the surface area change, and the difference in sphericity is extracted as the sphericity change.

[0047] Preferably, the step of reconstructing the main nerve fiber bundles passing through the normal brain tissue region from the diffusion tensor imaging data, and calculating the anisotropy fraction and mean diffusion rate of each fiber bundle, specifically includes:

[0048] Preprocessing operations are applied to diffusion tensor imaging data, including eddy current distortion correction, head motion artifact correction, and brain tissue extraction.

[0049] Within the normal brain tissue region, the main diffusion direction is determined based on the eigenvector of the diffusion tensor matrix of each voxel, and a fiber tract tracing algorithm is used to gradually connect adjacent voxels along the main direction to generate a continuous fiber tract path.

[0050] For each reconstructed nerve fiber bundle, the eigenvalues ​​of the diffusion tensor are extracted to calculate the anisotropy fraction, which reflects the direction dependence of water molecule diffusion.

[0051] The arithmetic mean of the three eigenvalues ​​of the diffusion tensor is calculated as the average diffusion rate of the nerve fiber bundle.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] By calculating the morphological evolution rate of the core trauma region and identifying its abrupt change points, and then matching these abrupt change points with inflection points in the signal intensity change curves of the edema region, a signal event is only identified when the two coincide temporally. This cross-modal parameter temporal correlation rule enables precise identification of key physiological time points in the trauma evolution process, overcoming the reliability limitations of single-parameter analysis. It provides a highly specific temporal benchmark for all subsequent analyses and improves the accuracy of prognostic biomarker discovery.

[0054] At specific time points following the occurrence of the aforementioned signaling events, diffusion tensor imaging data of normal brain tissue regions spatially and topologically associated with the traumatic area are extracted, and the degree of deviation of their fiber tract integrity from the baseline state is quantified. This quantitative analysis of "state deviation" in "normal" regions, conducted during periods of active pathological activity, can sensitively capture the early trajectory of damage propagation along the white matter fiber tract network, dynamically revealing the path of damage propagation rather than merely showing the final damage outcome. This enables early identification of secondary neurological injuries and prediction of risk areas.

[0055] Based on dynamically generated lesion propagation paths, these paths are mapped to functional network nodes in a standard brain atlas, identifying the set of affected functional network nodes. This method directly links the structural dynamics of lesion propagation with the brain's functional network architecture, proactively identifying potentially damaged specific functional circuits. It provides an objective, connectivity-based basis for personalized prognostic assessment, elevating prognostic prediction from the traditional lesion localization level to the level of brain network function impact. Attached Figure Description

[0056] Figure 1 This is a schematic diagram illustrating the working principle of the three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury described in this invention.

[0057] Figure 2 A flowchart illustrating the operation of the image segmentation and topology construction module;

[0058] Figure 3 A flowchart illustrating the operation of the trauma evolution analysis module;

[0059] Figure 4 Line graph showing the rate of evolution of craniocerebral trauma;

[0060] Figure 5 This is a diagram showing the morphological rate changes in trauma evolution analysis. Detailed Implementation

[0061] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Please see Figure 1This invention provides a three-dimensional model-based prognostic prediction and analysis system for craniocerebral trauma. The system includes: an image segmentation and topology construction module, a trauma evolution analysis module, a signal event recognition module, a structural connectivity analysis module, and a functional network mapping module. The image segmentation and topology construction module performs multi-scale segmentation on the three-dimensional image of the patient's brain, identifying the trauma core region, edema region, and surrounding normal brain tissue regions, and constructs a spatial topological connectivity map between these regions. Based on this spatial topological connectivity map, the trauma evolution analysis module tracks the volume, shape parameters, and diffusion trend of the trauma core region at different time points, calculating the morphological evolution rate of the trauma region over consecutive time intervals. The signal event recognition module, based on the morphological evolution rate and combined with the signal intensity change curve of the edema region over the same time interval, identifies inflection points in edema signal changes and marks signal events matching abrupt changes in the trauma core evolution rate. The structural connectivity analysis module extracts diffusion tensor imaging data of neural fiber tracts associated with normal brain tissue regions when signal events occur, quantifies the deviation of fiber tract integrity, and generates damage propagation paths based on structural connectivity. The functional network mapping module maps the damage propagation path to functional network nodes in a standard brain atlas, identifies the set of functional network nodes affected by the damage propagation path, and generates key prognostic brain network identifiers.

[0063] Example 1: The spatial topology connectivity graph includes a region adjacency matrix, boundary contact area, and centroid distance vector. The morphological evolution rate includes volume expansion acceleration, surface irregularity growth rate, and vector direction of invasion into surrounding regions. The signal events include signal intensity peak time, signal attenuation start time, and signal plateau duration. The damage propagation path based on structural connectivity includes a list of damaged fiber bundle numbers, integrity deviation value, and path hop count. The key prognostic brain network identifiers include the affected node number, the functional network type to which the node belongs, and the estimated loss of connectivity strength within the network.

[0064] In practice, the region adjacency matrix is ​​a mathematical representation describing the spatial connectivity between the trauma core region, the edema region, and the surrounding normal brain tissue regions. The boundary contact area quantifies the size of the shared boundary between any two adjacent regions. The centroid distance vector defines the direction and distance from the geometric center of one region to the geometric center of another adjacent region. The morphological evolution rate includes the volume expansion acceleration, the surface irregularity growth rate, and the vector direction of intrusion into surrounding regions. The volume expansion acceleration describes the acceleration or deceleration of the volume change of the trauma core region over time. The surface irregularity growth rate reflects the rate of change in the complexity of the boundary shape of the trauma core region. The vector direction of intrusion into surrounding regions indicates the primary spatial orientation of the trauma region's expansion in three-dimensional space.

[0065] Signal events include the peak signal intensity time, the onset of signal decay, and the duration of the signal plateau. The peak signal intensity time is the point at which the signal intensity of the edema region reaches its highest point in the imaging sequence. The onset of signal decay marks the turning point where the edema signal begins to decline from its peak. The duration of the signal plateau is the length of time during which the signal intensity remains relatively stable before entering decay. Damage propagation pathways based on structural connectivity include a list of damaged fiber tract numbers, integrity deviation values, and pathway hop counts. The list of damaged fiber tract numbers records the identifiers of specific neural fiber tracts identified as having integrity impairment. The integrity deviation value quantifies the degree of difference between the anisotropy score of each damaged fiber tract and the healthy baseline value. The pathway hop count describes the number of intermediate regions traversed by the damage as it propagates from the origin region along the fiber tract to the distal functional node. Key prognostic brain network identifiers include the affected node number, the functional network type to which the node belongs, and the estimated loss of connectivity strength within the network. The affected node number is a unique identifier for the functional network node affected by the damage propagation pathway on a standard brain atlas. The functional network type to which the node belongs indicates the classification of the higher cognitive functional subsystem to which the affected node belongs. The network connectivity strength loss prediction is a numerical estimate of the degree of decline in information transmission efficiency within the functional network to which the node is located, based on the damage to structural connections.

[0066] In practical implementation, the composition of the spatial topological connectivity map is illustrated using continuous imaging data of a trauma patient as an example. At time point T1, the system-generated region adjacency matrix shows that the trauma core region is adjacent to the edema region, but not directly adjacent to the surrounding normal brain tissue region. The boundary contact area between the trauma core region and the edema region is calculated to be A1 square millimeters. The vector from the centroid of the trauma core region to the centroid of the edema region is V1 millimeters. At time point T2, the updated region adjacency matrix shows that the trauma core region has established adjacency with the surrounding normal brain tissue region, with a corresponding boundary contact area of ​​A2 square millimeters, and the centroid distance vector is updated to V2 millimeters. The calculation of the morphological evolution rate relies on the basic data provided by the above spatial topological connectivity map. The volume expansion acceleration is obtained by comparing the difference in the rate of volume change between adjacent time intervals. The surface irregularity growth rate is obtained by analyzing the slope of the change in the sphericity parameter of the trauma core region.

[0067] The system first calculates the sphericity index of the trauma core region at multiple acquisition time points. This index quantifies the degree to which the shape approximates a sphere by the ratio of the region's volume to its surface area. Subsequently, the system extracts the difference in sphericity values ​​between adjacent time points as the sphericity change, and divides this change by the corresponding time interval to obtain the rate of change of the sphericity parameter, i.e., the surface irregularity growth rate. The vector direction of intrusion into the surrounding area is determined by analyzing the displacement direction of the centroid of the trauma core region between consecutive time points and combining it with the change in the boundary contact area with adjacent regions. The identification of signal events requires time alignment with the morphological evolution rate. For example, the system identifies the peak signal intensity in the edema region at time point Tx, i.e., the peak signal intensity time. At time point Ty, the system identifies the start of signal attenuation. Between time points Tx and Ty, the system measures the duration of the signal plateau period. The system matches time point Tx with the moment when the acceleration of volume expansion in the trauma core region abruptly changes; if the time deviation is within a preset tolerance, the peak time is marked as a relevant signal event.

[0068] In some embodiments, the generation of damage propagation paths based on structural connectivity involves the processing of specific data. The list of damaged fiber tract numbers may include identifiers such as "AF_L" and "ILF_L". The integrity deviation value is a scalar value calculated for each fiber tract; for example, the integrity deviation value for fiber tract "AF_L" is 0.15, and for fiber tract "ILF_L" it is 0.22. The path hop count is determined based on the anatomical connections of the fiber tracts; for example, if the path from the trauma area to the default network node may pass through two intermediate brain regions, the path hop count is recorded as 2. The generation of prognostic key brain network identifiers is based on the above path information. Affected node numbers may be "Node_44" and "Node_87" in a standard atlas. The functional network type of the nodes is determined by querying the atlas database; "Node_44" belongs to the "default network," and "Node_87" belongs to the "salience network." The estimated loss of intra-network connectivity strength is derived by combining the integrity deviation value of the damaged fiber tract with historical statistical models; for example, the estimated loss proportion for "Node_44" is 25%. It is understandable that the specific calculation of volume expansion acceleration can be achieved using a mathematical model. In one implementation, volume expansion acceleration can be quantified by the following formula:

[0069]

[0070] in: Indicates within the time interval [ , The average volume expansion acceleration within ] , , These represent points in consecutive time. , , The measured volume of the core trauma region, This represents a fixed time interval between adjacent time points. The surface irregularity growth rate can be obtained by calculating the time derivative of the sphericity parameter. The vector direction of intrusion into the surrounding area is determined by the centroid coordinate difference of the trauma core region at consecutive time points. It can be understood that matching the time parameters of signal events with abrupt changes in morphological evolution rates is a step in the system's operation. The system records the peak time of signal intensity, the start time of signal decay, and the duration of the signal plateau period. The system locates abrupt changes in morphological evolution rates on the same time axis. The system pairs and marks inflection points with time deviations within a preset threshold with abrupt changes, forming a spatiotemporally correlated record of signal events for use by the structural connectivity analysis module.

[0071] In some embodiments, data comparison involves comparing the integrity deviation value of the damage propagation path with a preset threshold. The integrity deviation value calculated by the system, such as 0.15 and 0.22, is compared with a threshold pre-calculated from a database of healthy individuals. Fiber bundles with integrity deviation values ​​exceeding this threshold are filtered into a list of damaged fiber bundle numbers. The path hop count is calculated based on a brain connectivity map; for example, if there is a relay node on the anatomical path from the trauma area to node "Node_44", the path hop count is counted as 2. The prediction of intra-network connectivity strength loss in key prognostic brain network identifiers requires the integration of multi-source information. The system reads the affected node numbers "Node_44" and "Node_87". The system queries the connectivity map database to obtain the functional network type of the node, "Default Network" and "Salience Network". The intra-network connectivity strength loss prediction is calculated by weighting and fusing the integrity deviation values ​​of all damaged fiber bundles affecting the node and inputting them into a pre-stored regression model to obtain a percentage value, such as 25% and 18%, which represents the degree of expected performance decline of the functional node.

[0072] Example 2: See Figure 2The image segmentation and topology construction module performs multi-scale segmentation on the 3D images of the patient's brain, identifying the core trauma region, edema region, and surrounding normal brain tissue region, and constructs a spatial topological connectivity map between these regions. Specifically, this involves loading the 3D medical image sequence of the brain and using adaptive threshold segmentation and region growing algorithms to separate the voxel set of the core trauma region. Around the voxel set of the core trauma region, a level set evolution contour based on signal intensity is applied to delineate the continuous boundary of the edema region. Using the continuous boundary of the edema region as the boundary, the area extending outwards to tissues with image grayscale values ​​within the normal physiological range is defined as the surrounding normal brain tissue region. The spatial positional relationships between the core trauma region, edema region, and surrounding normal brain tissue region are calculated. Adjacency relationships are established using the region centroid as a reference point, and the boundary contact area is calculated using shared voxel surfaces to generate a topological connectivity map describing the spatial relationships between regions. Specifically, loading the 3D medical image sequence of the brain and using adaptive threshold segmentation and region growing algorithms to separate the voxel set of the core trauma region includes reading the grayscale value data of each voxel in the 3D medical image sequence of the brain. A grayscale histogram is generated based on the grayscale value distribution of the entire image sequence. An adaptive segmentation threshold is dynamically calculated based on the peak and trough values ​​of the histogram. Voxels with grayscale values ​​higher than the adaptive segmentation threshold are used as initial seed points. A region growing algorithm is used to iteratively check the neighboring voxels of each seed point. If the difference between the grayscale value of a neighboring voxel and the grayscale value of the seed point is less than a preset similarity tolerance, the neighboring voxels are merged into the current region. The iterative process is repeated until no new voxels are added, resulting in a set of connected voxels in the core trauma region.

[0073] In practical implementation, the image segmentation and topology construction module performs multi-scale segmentation on the three-dimensional image of the patient's brain, identifies the core trauma region, edema region, and surrounding normal brain tissue region, and constructs a spatial topological connection map between each region. Specifically, this includes loading the three-dimensional medical image sequence of the brain, using adaptive threshold segmentation and region growing algorithms to separate the voxel set of the core trauma region, applying a level set evolution contour based on signal intensity to delineate the continuous boundary of the edema region around the voxel set of the core trauma region, using the continuous boundary of the edema region as the boundary, extending outward to tissues with image gray values ​​within the normal physiological range, defining them as the surrounding normal brain tissue region, calculating the spatial positional relationship between the core trauma region, edema region, and surrounding normal brain tissue region, establishing regional adjacency relationships with the regional centroid as the reference point, calculating the boundary contact area using shared voxel surfaces, and generating a topological connection map describing the spatial relationship between regions. The process involves loading a three-dimensional medical image sequence of the brain and using adaptive threshold segmentation and region growing algorithms to separate the voxel set of the trauma core region. Specifically, this includes reading the grayscale value data of each voxel in the three-dimensional medical image sequence of the brain, generating a grayscale histogram based on the grayscale value distribution of the entire image sequence, dynamically calculating an adaptive segmentation threshold based on the peak and valley values ​​of the histogram, using voxels with grayscale values ​​higher than the adaptive segmentation threshold as initial seed points, and using a region growing algorithm to iteratively check the adjacent voxels of each seed point. If the difference between the grayscale value of an adjacent voxel and the grayscale value of the seed point is less than a preset similarity tolerance, the adjacent voxels are merged into the current region. The iterative process is repeated until no new voxels are added, thus obtaining the connected voxel set of the trauma core region.

[0074] In practical implementation, the segmentation and construction process is illustrated using a computed tomography (CT) scan of a patient with acute subdural hematoma as an example. The system loads the patient's three-dimensional medical image sequence of the brain, which is composed of consecutive stacked two-dimensional slices. The system reads the grayscale value data of each voxel in the sequence. The system generates a global grayscale histogram based on the grayscale values ​​of all voxels. The adaptive segmentation threshold is dynamically determined based on the valley position between the peak of the normal grayscale distribution of brain tissue and the peak of the high grayscale distribution of the hemorrhage or contusion area in the histogram. The preset similarity tolerance can be understood as a fixed parameter set according to the image noise level, used to control the inclusiveness of region growth. After obtaining the initial seed point, the region growth algorithm starts running. The algorithm checks the six neighboring voxels of the seed point. If the difference between the grayscale value of a neighboring voxel and the grayscale value of the current seed point is less than the preset similarity tolerance, the neighboring voxel is merged. The newly merged voxel becomes the new seed point. The iterative process continues until no new voxels meet the merging condition, ultimately outputting a connected set of voxels representing the core region of the trauma. In some embodiments, the voxel set of the trauma core region identified by the system appears as an irregular mass in three-dimensional space. Around the voxel set of the trauma core region, the system applies a level set evolution profile based on signal intensity to delineate the edema region. The level set function is initialized at the boundary of the trauma core region, and its evolution velocity field is driven by the local signal intensity gradient of the image, stopping at the edge between low-intensity edema regions and relatively normal white matter regions, thus delineating a continuous, smooth three-dimensional boundary of the edema region. Using the continuous boundary of the edema region as the outer boundary, the system performs voxel navigation outwards until it encounters tissue whose grayscale value falls within the statistical range of healthy brain tissue; this tissue is then defined as the surrounding normal brain tissue region.

[0075] In some embodiments, the construction of the spatial topological connectivity map is based on the three segmented regions described above. Calculating the spatial positional relationship between the trauma core region, the edema region, and the surrounding normal brain tissue region involves quantifying the geometric center and contact relationship. The system calculates the arithmetic mean of the coordinates of all voxels in the trauma core region to obtain the centroid coordinates of the trauma core region. The centroid coordinates of the edema region and the surrounding normal brain tissue region are calculated using the same method. Using the region centroid as a reference point, the system determines the spatial adjacency relationship between regions. If two regions share at least one voxel surface in three-dimensional space, they are determined to be adjacent and marked as 1 in the region adjacency relationship matrix; otherwise, they are marked as 0. The boundary contact area is calculated by traversing all voxels. For each pair of adjacent regions, the total area of ​​the shared voxel surfaces between them is counted; this area value is the boundary contact area. The centroid distance vector is obtained by subtracting the centroid coordinates of another adjacent region from the centroid coordinates of one region; this vector contains direction and distance information. For example, in a segmentation result, the trauma core region and the edema region are determined to be adjacent, with a boundary contact area of ​​150 square millimeters. The vector from the centroid of the trauma core region to the centroid of the edema region is (5.2, 3.1, 0.8) millimeters. The edema region is adjacent to some surrounding normal brain tissue regions, with a boundary contact area of ​​300 square millimeters. The trauma core region does not directly contact the surrounding normal brain tissue regions, so its corresponding value in the region adjacency matrix is ​​0, and its boundary contact area is 0. However, the system still calculates the Euclidean distance between their centroids as a spatial relationship reference. Optionally, the adaptive segmentation threshold can be calculated using a formula based on a histogram bimodal model:

[0076]

[0077] in: This represents the calculated adaptive segmentation threshold. and These represent the mean and standard deviation of the gray level of the first main peak in the gray-level histogram, respectively. and These represent the mean and standard deviation of the grayscale values ​​of the second primary peak, respectively. The driving velocity function of the level set evolution profile is inversely proportional to the grayscale gradient magnitude of the image, ensuring that evolution stops at the tissue boundary. Optionally, when calculating the boundary contact area, the system traverses each boundary voxel of the edema region, checking whether its six neighboring voxels belong to the surrounding normal brain tissue region. If so, the surface accumulation of this shared surface is added to the total boundary contact area between the edema region and the surrounding normal brain tissue region. Finally, the system integrates the region adjacency matrix, a series of boundary contact area values, and multiple sets of centroid distance vectors to generate a complete and quantifiable spatial topological connectivity graph. This graph is stored and transmitted in system memory in the form of data structures and associative arrays.

[0078] Example 3: See Figure 3 The trauma evolution analysis module, based on a spatial topology connectivity map, tracks the volume, shape parameters, and diffusion trends of the trauma core region at different time points, calculating the morphological evolution rate of the trauma region over consecutive time intervals. Specifically, this includes calculating the total volume, surface area, and sphericity of the trauma core region at multiple consecutive acquisition time points. Based on the total volume, surface area, and sphericity values ​​of the trauma core region at different time points, the changes in each parameter between adjacent time points are calculated. Combining the adjacency relationships between the trauma core region and surrounding areas in the spatial topology connectivity map, the main spatial orientation of the trauma region's volume growth is analyzed, and the diffusion trend vector is determined. Using the changes in each parameter and the diffusion trend vector, divided by the corresponding time interval, a set of morphological evolution rates reflecting the dynamic changes of the trauma region is obtained. Specifically, calculating the changes in each parameter between adjacent time points based on the total volume, surface area, and sphericity values ​​of the trauma core region at different time points includes, for each acquisition time point, multiplying the total number of all voxels in the trauma core region by the volume of a single voxel to obtain the total volume. The sum of the outer surface areas of the boundary voxels of the trauma core region is calculated as the surface area. The sphericity index is calculated based on the volume and surface area of ​​the trauma core region, and is expressed as the ratio of the region's volume to the surface area of ​​a sphere of the same volume. The difference in total volume between adjacent time points is extracted as the volume change, the difference in surface area is extracted as the surface area change, and the difference in sphericity is extracted as the sphericity change.

[0079] In practical implementation, the trauma evolution analysis module, based on a spatial topology connectivity map, tracks the volume, shape parameters, and diffusion trends of the trauma core region at different time points, and calculates the morphological evolution rate of the trauma region within continuous time intervals. Specifically, this includes calculating the total volume, surface area, and sphericity of the trauma core region at multiple consecutive acquisition time points. Based on the total volume, surface area, and sphericity values ​​of the trauma core region at different time points, the changes in each parameter between adjacent time points are calculated. Combining the adjacency relationship between the trauma core region and surrounding areas in the spatial topology connectivity map, the main spatial orientation of the trauma region's volume growth is analyzed, and the diffusion trend vector is determined. By dividing the changes in each parameter and the diffusion trend vector by the corresponding time interval, a set of morphological evolution rates reflecting the dynamic changes of the trauma region is obtained. Specifically, based on the total volume, surface area, and sphericity values ​​of the trauma core region at different time points, the changes in each parameter between adjacent time points are calculated. This includes, for each acquisition time point, multiplying the total number of all voxels in the trauma core region by the volume of a single voxel to obtain the total volume, calculating the sum of the outer surface areas of the boundary voxels of the trauma core region as the surface area, calculating the sphericity index based on the volume and surface area of ​​the trauma core region, which is expressed as the ratio of the region's volume to the surface area of ​​a sphere of the same volume, extracting the difference in total volume between adjacent time points as the volume change, the difference in surface area as the surface area change, and the difference in sphericity as the sphericity change.

[0080] In practical implementation, the workflow of the trauma evolution analysis module is illustrated using a series of MRI images of a patient with cerebral contusion and laceration. The system acquired three-dimensional images of the brain on days 1, 3, and 7 after the injury. On day 1, the system identified the core trauma region from the segmented images. The system counted the total number of voxels contained in the core trauma region, for example, N1 equals 15,000 voxels. Each voxel has a physical size of 1 mm x 1 mm x 1 mm, meaning a single voxel volume is 1 cubic millimeter. The total volume of the core trauma region is calculated by multiplying the total number of voxels by the volume of a single voxel, i.e., the total volume V1 equals 15,000 cubic millimeters. The system identified all boundary voxels located on the surface of the core trauma region and calculated the sum of the exposed areas of these boundary voxels as the surface area S1 of the core trauma region. It can be understood that the sphericity index is used to quantify how close the region's shape is to a perfect sphere. The system calculates the total volume of the core trauma region based on the total volume on day 1. and surface area Calculate sphericity Sphericity The calculation formula is:

[0081]

[0082] in: Indicates sphericity, This represents the total volume of the core region of the trauma. This represents the surface area of ​​the core region of the trauma. For a perfect sphere, sphericity... Equals 1; for other shapes, sphericity is 1. Less than 1. On day 3, the system repeats the above calculation process to obtain the total volume of the trauma core region. Surface area and sphericity Volume change Through calculation minus The change in surface area was obtained. Through calculation minus The change in sphericity was obtained. Through calculation minus The system calculates the total volume on day 7. Surface area and sphericity And then calculate the volume change. Surface area change and sphericity change .

[0083] In some embodiments, the determination of the diffusion trend vector relies on adjacency information provided by a spatial topological connectivity map. The system reads the spatial topological connectivity maps constructed at time points 1 and 3. The system analyzes which surrounding regions the trauma core region significantly increased in boundary contact area with during the volume growth from day 1 to day 3. For example, by comparing the spatial topological connectivity maps at time points 1 and 3, the system finds that the boundary contact area between the trauma core region and a normal brain tissue region called "temporal lobe white matter" increased from 0 square millimeters to 80 square millimeters, while the boundary contact area with the other side of the edema region did not change much. The system calculates the centroid displacement vector of the trauma core region from time point 1 to time point 3. The direction of the centroid displacement vector is consistent with the direction of the significant increase in boundary contact area, jointly indicating the main spatial orientation of volume growth. This comprehensively determined spatial orientation is quantified as a unit direction vector in three-dimensional space and, combined with the magnitude of volume growth, constitutes the diffusion trend vector. The calculation of the morphological evolution rate set is based on the aforementioned changes and diffusion trend vectors. Volume expansion rate. From volume change The rate of change of surface area is obtained by dividing by the time interval. Change in surface area Divide by the same time interval. The growth rate of surface irregularity can reflect the rate of evolution of shape complexity by differentiating the change in sphericity over time or by calculating its rate of change. Diffusion trend vector It already contains directional information, and its modulus reflects the rate of space intrusion. The volume expansion rate... Surface area change rate Surface irregularity growth rate and diffusion trend vector The set of morphological evolution rates constitutes the set of morphological evolution rates over the time interval from day 1 to day 3. For the interval from day 3 to day 7, the system performs the exact same calculation process to obtain the set of morphological evolution rates. , and .

[0084] Optionally, the rate of increase in surface irregularity can be quantified using the time derivative of sphericity. The system calculates the change in sphericity at adjacent time points. and divide it by the corresponding time interval. The value obtained / This represents the average surface irregularity growth rate. A negative surface irregularity growth rate indicates that the region's shape is becoming more irregular, while a positive rate indicates that the shape is becoming more regular. It is understandable that the calculation of the diffusion trend vector can be further refined. The system not only considers the centroid displacement but also analyzes the normal vector direction and distance of the outward expansion of various local regions on the three-dimensional surface of the trauma core region. The system meshes the surface of the trauma core region, compares the surface models at time point 1 and time point 3, and calculates the displacement of each surface vertex along its normal direction. The system performs vector synthesis on all vertex displacements to obtain a more accurate diffusion trend vector that reflects the overall expansion direction. In some embodiments, the data comparison reflects the differences in evolution rates at different time intervals. The system records and compares the set of morphological evolution rates. , , and , , For example, the rate of volume expansion. The value is 750 cubic millimeters of expansion per day, while the volume expansion rate is... The value decreased to 200 cubic millimeters of expansion per day, indicating that the rate of expansion of the trauma core slowed down in the later stages. (Diffusion trend vector) The direction may point towards the temporal lobe, while the diffusion trend vector The direction points towards the parietal lobe, suggesting that the dominant direction of trauma spread may have changed over time. The rate of increase in surface irregularity in the early intervals... / A negative value indicates a later interval. / Approaching zero indicates that the shape of the trauma core has transitioned from rapidly becoming irregular to a relatively stable stage. All these calculated morphological evolution rate parameters are stored by the system and transmitted to the signal event recognition module for subsequent matching analysis. Optionally, when calculating the sum of the outer surface areas of boundary voxels, the system examines the six adjacent faces of each voxel identified as a boundary voxel, counts the number of adjacent faces that do not belong to the trauma core region, multiplies this number by the area of ​​a single face to obtain the surface area contributed by that boundary voxel, and sums the contribution values ​​of all boundary voxels to obtain the total surface area of ​​the trauma core region.

[0085] Example 4: The signal event recognition module identifies inflection points in edema signal changes based on morphological evolution rates and the signal intensity change curves of the edema region within the same time interval. It marks signal events matching abrupt changes in the core trauma evolution rate. Specifically, this includes extracting the average signal intensity value of the edema region at continuous time points to form a time-varying edema signal intensity curve. The edema signal intensity curve is smoothed and differentiated to locate the zero-crossing point of the first derivative of the signal intensity, identifying it as an inflection point. The time point of the signal change inflection point is compared with the moment of abrupt change in the rate value in the morphological evolution rate set. Inflection points with time deviations less than a preset tolerance are matched with the abrupt change moments, and the matching pairs are marked as signal events with spatiotemporal correlation. The structural connectivity analysis module extracts diffusion tensor imaging data of neural fiber bundles associated with normal brain tissue regions at the time of the signal event, quantifies the degree of deviation in fiber bundle integrity, and generates damage propagation paths based on structural connectivity. Specifically, this includes acquiring diffusion tensor imaging data of normal brain tissue regions spatially adjacent to the edema region and the core trauma region at the time point marked by the signal event. The main neural fiber bundles traversing normal brain tissue regions were reconstructed from diffusion tensor imaging (DTI) data, and the anisotropy fraction and mean diffusion rate of each bundle were calculated. The calculated anisotropy fractions and mean diffusion rates were compared with corresponding values ​​in a standard template database of healthy individuals to calculate the deviation. The bundles were sorted according to the magnitude of the deviation, and bundles with deviations exceeding a threshold were selected and connected in series according to their spatial connectivity to form a lesion propagation path describing the potential spread of damage along white matter fiber pathways. Specifically, the reconstruction of the main neural fiber bundles traversing normal brain tissue regions from DTI data and the calculation of the anisotropy fraction and mean diffusion rate of each bundle included preprocessing operations on the DTI data, including eddy current distortion correction, head motion artifact correction, and brain tissue extraction. Within the normal brain tissue region, the principal diffusion direction was determined based on the eigenvector of the diffusion tensor matrix for each voxel, and a fiber tract tracing algorithm was used to progressively connect adjacent voxels along the principal direction to generate a continuous fiber tract path. For each reconstructed nerve fiber bundle, the eigenvalues ​​of the diffusion tensor are extracted to calculate the anisotropy fraction, reflecting the direction dependence of water molecule diffusion. The arithmetic mean of the three eigenvalues ​​of the diffusion tensor is calculated as the average diffusion rate of the nerve fiber bundle.

[0086] In practical implementation, the signal event recognition module identifies inflection points of edema signal change based on the morphological evolution rate and the signal intensity change curve of the edema region within the same time interval. It then marks signal events that match the abrupt change points of the trauma core evolution rate. Specifically, this includes extracting the average signal intensity value of the edema region at continuous time points to form an edema signal intensity curve that changes over time, smoothing and differentiating the edema signal intensity curve, locating the zero-crossing point of the first derivative of the signal intensity, identifying it as an inflection point of signal change, comparing the time point of the inflection point with the moment when the rate value in the morphological evolution rate set changes abruptly, matching inflection points with time deviations less than a preset tolerance with the abrupt change time, and marking the matching pairs as signal events with spatiotemporal correlation. In practice, the structural connectivity analysis module extracts diffusion tensor imaging data of nerve fiber bundles associated with normal brain tissue regions at the time of the signal event, quantifies the degree of deviation of fiber bundle integrity, and generates a damage propagation path based on structural connectivity. Specifically, this includes acquiring diffusion tensor imaging data of normal brain tissue regions spatially adjacent to the edema region and the core of the injury at the time point marked by the signal event, reconstructing the main nerve fiber bundles passing through the normal brain tissue regions from the diffusion tensor imaging data, calculating the anisotropy fraction and average diffusion rate of each fiber bundle, comparing the calculated anisotropy fraction and average diffusion rate with the corresponding values ​​in the standard template database of healthy individuals, calculating the deviation, sorting the fiber bundles according to the magnitude of the deviation, filtering out fiber bundles with deviation exceeding the threshold, and connecting them in series according to their spatial connectivity to form a damage propagation path describing the possible extension of damage along the white matter fiber pathway. The process involves reconstructing the main neural fiber bundles traversing normal brain tissue regions from diffusion tensor imaging data, calculating the anisotropy fraction and average diffusion rate of each fiber bundle, and specifically applying preprocessing operations to the diffusion tensor imaging data, including eddy current distortion correction, head motion artifact correction, and brain tissue extraction. Within the normal brain tissue region, the main diffusion direction is determined based on the eigenvector of the diffusion tensor matrix for each voxel. A fiber bundle tracing algorithm is used to progressively connect adjacent voxels along the main direction to generate continuous fiber bundle paths. For each reconstructed neural fiber bundle, the eigenvalues ​​of the diffusion tensor are extracted to calculate the anisotropy fraction, reflecting the direction dependence of water molecule diffusion. The arithmetic mean of the three eigenvalues ​​of the diffusion tensor is calculated as the average diffusion rate of the neural fiber bundle.

[0087] In practical implementation, taking continuous magnetic resonance imaging data of a trauma patient as an example, the workflow of the signal event recognition module is as follows: The system extracts the average signal intensity value of the edema region from the edema region segmentation results, based on images acquired on days 1, 2, 3, 5, 7, and 10 post-injury, forming a time-varying sequence of data. The system uses Gaussian filtering to smooth the edema signal intensity sequence to reduce noise and calculates the first derivative of the smoothed sequence. The system locates the zero-crossing points on the first derivative curve, i.e., the points where the derivative changes from positive to negative or vice versa. Points where the derivative changes from positive to negative correspond to peak signal intensity times, while points where the derivative changes from negative to positive may correspond to the start of signal decay or the start of a plateau phase. The system records the identified inflection point times, for example, identifying a peak signal intensity time on day 3 post-injury and a signal decay start time on day 7 post-injury. Simultaneously, the system receives a set of morphological evolution rates from the trauma evolution analysis module, which includes changes in parameters such as volume expansion acceleration and surface irregularity growth rate over time. The system detects moments when parameter values ​​in the morphological evolution rate set undergo significant abrupt changes, such as a sharp increase in volume expansion acceleration around day 3 post-injury. The system compares the peak signal intensity time with the abrupt change in volume expansion acceleration. A preset tolerance is set to, for example, 12 hours. Since the time deviation is within the tolerance range, the system marks the peak signal intensity on day 3 and the abrupt change in volume expansion acceleration on day 3 as a matching signal event. Refer to Table 1; the system records the matching results.

[0088] Table 1: Matching Table of Signaling Events and Abrupt Changes in Morphological Evolution Rate

[0089] In some embodiments, the structural connectivity analysis module initiates analysis at the time point marked by the signal event. The module acquires diffusion tensor imaging (DTI) data collected on day 3 post-injury. Based on the spatial topological connectivity map, the module determines the spatial coordinate range of normal brain tissue regions spatially adjacent to the edema region and the trauma core region. The module applies preprocessing operations to the DTI data from day 3 post-injury, including eddy current distortion correction using a field map-based correction method, head motion artifact correction using a rigid body registration algorithm, and scalp and skull stripping using a brain tissue extraction algorithm. In the preprocessed DTI data, the module performs fiber tract reconstruction within the coordinate range of the target normal brain tissue region. The module calculates the diffusion tensor matrix at each voxel location and performs eigenvalue decomposition on the matrix to obtain the principal eigenvector as the principal direction of water molecule diffusion at that voxel. The fiber tract tracing algorithm starts from a set of predefined seed points and extends progressively along the principal diffusion direction, connecting adjacent voxels to generate continuous fiber tract paths traversing the target region, such as reconstructing the arcuate fasciculus and inferior longitudinal fasciculus fibers traversing the temporal lobe target region. For each reconstructed neural fiber bundle, the module extracts the diffusion tensor eigenvalues ​​of all voxels along the fiber path. The anisotropy fraction is calculated based on these eigenvalues, using the following formula:

[0090]

[0091] in: Represents the fractions of anisotropy. , , These represent the three eigenvalues ​​of the diffusion tensor. The formula for calculating the mean diffusion rate (MD) is ( + + ) / 3. The module calculates the average anisotropy fraction and average diffusion rate for each fiber bundle.

[0092] Understandably, the calculation of fiber tract integrity deviation is achieved through comparison with a standard template. The module accesses a stored database of standard templates for healthy individuals, which contains normal reference values ​​and standard deviations for the mean anisotropy fraction and mean diffusion rate of each major fiber tract in different brain regions. The module compares the mean anisotropy fraction of the target fiber tract in the current patient with the mean anisotropy fraction of the corresponding fiber tract in the standard template. Fiber tract integrity deviation. The calculation can employ a standardized distance method, such as calculating the difference between the patient's value and the template mean, and then dividing by the template's standard deviation. Optionally, the module can calculate the deviation of the anisotropy score separately. Deviation from average dispersion The system allows setting a deviation threshold, such as 2.0. The module then filters out... or Fiber bundles exceeding 2.0 are included in the list of damaged fiber bundles. Fiber bundles are sorted according to the degree of deviation, for example, the left arcuate bundle... It is 2.5, the lower longitudinal beam If the score is 1.8, the left arcuate fasciculus ranks higher. The module, based on the white matter fiber connectivity map, strings together the selected, anatomically connected damaged fiber bundles according to their spatial connectivity. For example, damaged uncinate fibers and damaged subfronto-occipital fasciculus fibers are anatomically connected anteriorly and posteriorly; the system strings them together to form a potential damage propagation path from the temporal pole to the occipital lobe. This path includes a list of damaged fiber bundle numbers, the integrity deviation value of each fiber, and the number of path jumps required from the injury area to the distal node, collectively constituting a damage propagation path based on structural connectivity. In some embodiments, data comparison is reflected in the difference between the patient's fiber bundle indices and the standard template values. For example, if the patient's mean anisotropy score for the left arcuate fasciculus is 0.45, while the mean anisotropy score for the corresponding fiber bundle in the standard template is 0.55 with a standard deviation of 0.04, then the deviation... The calculation is (0.45-0.55) / 0.04=-2.5. The average diffusion rate per patient is 0.90x10⁻¹⁰. - ³mm² / s, template value 0.80x10 - ³mm² / s, with a standard deviation of 0.05 x 10³ mm² / s. - ³mm² / s, deviation The deviation is 2.0. Since either of the two absolute deviation values ​​exceeds the threshold of 2.0, the left arcuate tract is marked as damaged, with an integrity deviation value of -2.5. Optionally, when reconstructing the neural fiber tract path, the fiber tract tracking algorithm employs a deterministic tracking method. Starting from the seed point, it tracks one step forward along the principal diffusion direction of the current voxel. At the next voxel, it determines the next direction based on its principal diffusion direction. Tracking termination conditions include an anisotropy fraction below a preset lower limit or a fiber turning angle exceeding a preset upper limit.

[0093] See Figure 4 This is a line graph showing the evolution rate of traumatic brain injury, illustrating the changes in morphological parameters of the core traumatic region over the number of days post-injury. The volume expansion acceleration rapidly rises to its peak 1-3 days post-injury, then declines continuously. The surface irregularity rate increases continuously from 1-8 days post-injury, then slightly decreases after 8 days. Rate abrupt changes occur at 3 days (peak volume expansion acceleration) and 8 days (peak surface irregularity rate). This type of graph is a core visualization tool in traumatic brain injury prognostic analysis, used to track the dynamic evolution of the traumatic region; identify abrupt changes in morphological parameters to provide a basis for subsequent "signal event matching"; and assist physicians in determining the stage of traumatic development.

[0094] Example 5: The functional network mapping module maps damage propagation paths based on structural connectivity to functional network nodes in a standard brain atlas. It identifies the set of functional network nodes affected by the damage propagation path and generates key prognostic brain network identifiers. Specifically, this includes spatially registering the fiber bundles included in the structural connectivity-based damage propagation path with the standard white matter fiber atlas. It determines the cortical or deep nucleus regions connected by the fiber bundles, maps these regions to the standard brain functional network atlas, and obtains the corresponding functional network node numbers. It counts all affected functional network node numbers, removes duplicates, and forms a functional network node set. It queries the preset functional network type to which each node in the functional network node set belongs, estimates the proportion of connection strength loss based on historical data, and generates key prognostic brain network identifiers.

[0095] In practical implementation, the functional network mapping module maps damage propagation paths based on structural connectivity to functional network nodes in a standard brain atlas, identifies the set of functional network nodes affected by the damage propagation paths, and generates key prognostic brain network identifiers. Specifically, this includes spatially registering the fiber bundles included in the damage propagation paths based on structural connectivity with standard white matter fiber maps, determining the cortical or deep nuclei regions connected by the fiber bundles, mapping the cortical or deep nuclei regions to standard brain functional network maps to obtain the corresponding functional network node numbers, counting all affected functional network node numbers, deduplicating them to form a set of functional network nodes, querying the preset functional network type to which each node in the set belongs, estimating the proportion of connection strength loss based on historical data, and generating key prognostic brain network identifiers.

[0096] In practical implementation, taking the analysis results of a trauma patient as an example, the damage propagation path based on structural connectivity includes a list of damaged fiber tract numbers, such as "AF_L" and "UF_L". The functional network mapping module loads a standard white matter fiber atlas, which is a digital template annotated with the three-dimensional spatial trajectories of major white matter fiber tracts. The module registers the patient's diffusion tensor imaging space with the space of the standard white matter fiber atlas, aligning the spatial coordinates of the patient's image to the standard atlas space using a nonlinear transformation algorithm. In the registered space, the module compares the spatial position of the reconstructed "AF_L" fiber tract in the damage propagation path with the trajectory of the "arctic fasciculus" defined in the standard atlas to confirm their consistency. Similarly, it compares the "UF_L" fiber tract with the trajectory of the "uncinate fasciculus" in the standard atlas. The cortical or deep nucleus regions connected by the fiber tracts are determined by querying the anatomical annotation information of the standard white matter fiber atlas. For example, the standard white matter fiber atlas records that the anterior end of the "arctic fasciculus" connects to "Brodman's area 44" and the posterior end connects to the "posterior part of the superior temporal gyrus." The "uncinate fasciculus" connects the "orbitofrontal cortex" and the "temporal pole." The module records that the cortical regions connected by the "AF_L" fiber tracts are "Brodman's area 44" and the "posterior part of the superior temporal gyrus," and that the cortical regions connected by the "UF_L" fiber tracts are "orbitofrontal cortex" and the "temporal pole."

[0097] In some embodiments, mapping the cerebral cortex or deep nuclei to a standard brain functional network atlas is a subsequent step. The module loads a standard brain functional network atlas, which divides the whole brain into multiple nodes, each with a unique number and corresponding to an anatomical brain region, and predefines the large-scale functional network type to which each node belongs. The module maps the anatomical region “Brodman 44” to the node number “Node_101” in the functional network atlas, “posterior superior temporal gyrus” to “Node_205”, “orbitofrontal cortex” to “Node_033”, and “temporal pole” to “Node_188” using coordinate mapping. The module counts all affected functional network node numbers, forming a list [Node_101, Node_205, Node_033, Node_188]. The module removes duplicates from this list; since there are no duplicates in this example, a set of functional network nodes {Node_101, Node_205, Node_033, Node_188} is formed. The module queries the preset functional network type to which each node in the set of functional network nodes belongs. It accesses the metadata of the standard brain functional network map and finds that Node_101 belongs to "language network", Node_205 belongs to "default network", Node_033 belongs to "salience network", and Node_188 belongs to "default network".

[0098] It is understandable that estimating the proportion of connection strength loss based on historical data is a crucial step in generating key prognostic brain network identifiers. Historical data refers to a pre-established database containing a large number of trauma cases, which records statistical models of the relationship between fiber tract integrity deviations and corresponding functional connection strength loss proportions. For each node in the functional network node set, such as Node_101, the module retrieves the integrity deviations of all damaged fiber tracts terminating at or passing through that node. Assuming only the “AF_L” fiber tract connects to Node_101, its integrity deviation is -2.5. The module inputs this deviation value into a specific statistical model for the “language network,” which can be a sigmoid function, outputting a value between 0 and 1 representing the estimated proportion P of connection strength loss. The formula for calculating the connection strength loss proportion L can be expressed as:

[0099]

[0100] in: Represents the estimated percentage of connection strength loss for a specific function network node; This represents the deviation of the integrity of the damaged fiber bundle at the input node; and These are model parameters trained based on historical data, used to describe the S-shaped curve relationship between the deviation value and the connection loss probability; It is a natural constant. The module performs this estimation process for each node in the node set, for example, obtaining a predicted loss L of 30% for Node_101 and a predicted loss L of 30% for Node_205. The estimated loss L for Node_033 is 15%, for Node_188 it is 20%, and for Node_188 it is 10%. Ultimately, the prognostic key brain network identifiers generated by the module contain the following structured information: a list of affected node numbers {Node_101, Node_205, Node_033, Node_188}, the functional network type of each node [language network, default network, salience network, default network], and the estimated intra-network connectivity strength loss for each node [30%, 15%, 20%, 10%]. In some embodiments, the data comparison reflects the different estimated losses at different nodes due to the different damaged fiber bundles in the connections. Node_101 is connected by only one high-deviation fiber bundle "AF_L", and its estimated loss proportion is relatively high. Node_205 and Node_188 both belong to the default network, but Node_205 is connected by "AF_L" while Node_188 is connected by "UF_L". Therefore, even though they belong to the same functional network type, their estimated connection strength loss is different, at 15% and 10% respectively. This reflects the impact of the damage propagation path on the heterogeneity of different nodes within the network. Optionally, the spatial registration process uses a nonlinear registration algorithm to better handle differences in individual brain anatomy and accurately align the fiber tract trajectories of individual patients to the standard white matter fiber atlas space. Optionally, when multiple damaged fiber tracts converge at the same functional network node, the module integrates the integrity deviation values ​​of all relevant fiber tracts when estimating the connection strength loss ratio L, for example, taking the deviation with the largest absolute value or the weighted average deviation as the model input value. .

[0101] See Figure 5 This is a trauma evolution analysis - morphological rate change graph. The graph marks two key time points: 5 days (red dashed line): the abrupt change in volume expansion acceleration, after which the rate of decrease slows; 7 days (blue dashed line): the abrupt change in the rate of increase in surface irregularity, after which the rate of increase turns to decrease. In the initial stage, the object is dominated by "rapid volume expansion," with slow changes in surface morphology; in the middle stage (5-7 days), the rate of volume expansion slows, and the surface morphology begins to become drastically irregular; in the later stage, volume expansion tends to stagnate (or even shrink), and the degree of surface irregularity gradually stabilizes. The purpose of this type of analysis is to help researchers determine the dynamic evolution stage of an object or optimize related regulatory strategies.

[0102] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional model-based traumatic brain injury prognosis prediction analysis system, characterized by, Includes the following modules: The image segmentation and topology construction module is used to perform multi-scale segmentation of the three-dimensional images of the patient's brain, identify the core trauma region, edema region and surrounding normal brain tissue region, and construct a spatial topology connection map between each region. The trauma evolution analysis module is used to track the volume, shape parameters and diffusion trend of the trauma core region at different time points based on the spatial topology connection map, and to calculate the morphological evolution rate of the trauma region in continuous time intervals. The signal event recognition module is used to identify the inflection point of edema signal change based on the morphological evolution rate and the signal intensity change curve of the edema region within the same time interval, and to mark the signal event that matches the mutation point of the trauma core evolution rate. The structural connectivity analysis module is used to extract diffusion tensor imaging data of nerve fiber bundles associated with normal brain tissue regions when the signal event occurs, quantify the degree of deviation of fiber bundle integrity, and generate damage propagation paths based on structural connectivity. The functional network mapping module is used to map the damage propagation path based on structural connectivity to functional network nodes in a standard brain atlas, identify the set of functional network nodes affected by the damage propagation path, and generate key prognostic brain network identifiers. The spatial topology connectivity graph includes a region adjacency matrix, boundary contact area, and centroid distance vector. The morphological evolution rate includes volume expansion acceleration, surface irregularity growth rate, and vector direction of invasion into surrounding regions. The signal events include signal intensity peak time, signal attenuation start time, and signal plateau duration. The damage propagation path based on structural connectivity includes a list of damaged fiber bundle numbers, integrity deviation value, and path hop count. The prognostic key brain network identifier includes the affected node number, the functional network type to which the node belongs, and the estimated loss of connectivity strength within the network.

2. The three-dimensional model-based cranial brain trauma prognosis prediction analysis system according to claim 1, characterized in that, The image segmentation and topology construction module performs multi-scale segmentation on the three-dimensional image of the patient's brain, identifies the core trauma region, edema region, and surrounding normal brain tissue region, and constructs a spatial topological connectivity map between the regions, specifically including: Load the three-dimensional medical image sequence of the brain and use adaptive threshold segmentation and region growing algorithm to separate the voxel set of the core trauma region; Around the voxel set in the core region of the trauma, a level set evolution profile based on signal intensity is applied to delineate the continuous boundary of the edema region. The area of ​​normal brain tissue surrounding the edema region is defined as the continuous boundary of the edema region, extending outward to the tissue in which the image gray value is within the normal physiological range. The spatial relationship between the trauma core region, the edema region and the surrounding normal brain tissue region is calculated. The region adjacency relationship is established with the region centroid as the reference point. The boundary contact area is calculated with the shared voxel surface, and a topological connection map describing the spatial relationship between the regions is generated.

3. The three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury according to claim 2, characterized in that, The trauma evolution analysis module, based on the spatial topology map, tracks the volume, shape parameters, and diffusion trend of the trauma core region at different time points, and calculates the morphological evolution rate of the trauma region over continuous time intervals, specifically including: The total volume, surface area, and sphericity of the trauma core region were calculated at multiple consecutive acquisition time points. Based on the total volume, surface area and sphericity values ​​of the trauma core region at different time points, calculate the changes in each parameter between adjacent time points; Based on the adjacency relationship between the trauma core region and the surrounding region in the spatial topology connection diagram, the main spatial orientation of the trauma region volume growth is analyzed to determine the diffusion trend vector; By dividing the changes and diffusion trend vectors of the aforementioned parameters by the corresponding time interval, a set of morphological evolution rates reflecting the dynamic changes of the trauma area is obtained.

4. The three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury according to claim 3, characterized in that, The signal event recognition module, based on the morphological evolution rate and combined with the signal intensity change curve of the edema region within the same time interval, identifies inflection points in edema signal changes and marks signal events that match the abrupt change points in the evolution rate of the trauma core. Specifically, this includes: The average signal intensity value of the edema region at continuous time points is extracted to form an edema signal intensity curve that changes with time. The edema signal intensity curve is smoothed and differentiated to locate the zero-crossing point of the first derivative of the signal intensity, which is identified as the inflection point of the signal change. Compare the time point at which the inflection point of the signal change occurs with the time when the rate value in the morphological evolution rate set undergoes a sudden change; The inflection point with a time deviation less than the preset tolerance is matched with the moment of sudden change, and the matching pair is marked as a signal event with spatiotemporal correlation.

5. The three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury according to claim 4, characterized in that, The structural connectivity analysis module extracts diffusion tensor imaging data of nerve fiber tracts associated with normal brain tissue regions when the signal event occurs, quantifies the degree of deviation from fiber tract integrity, and generates a damage propagation path based on structural connectivity, specifically including: At the time point marked by the signal event, diffusion tensor imaging data of normal brain tissue regions that are spatially adjacent to the edema region and the core trauma region are acquired; The main nerve fiber bundles passing through the normal brain tissue region are reconstructed from the diffusion tensor imaging data, and the anisotropy fraction and mean diffusion rate of each fiber bundle are calculated. The calculated anisotropy score and average diffusion rate are compared with the corresponding values ​​in the standard template database of healthy individuals to calculate the deviation. The fiber bundles are sorted according to the degree of deviation, and fiber bundles with deviation exceeding the threshold are selected and connected in series according to their spatial connection relationship to form a damage propagation path that describes the possible spread of damage along the white matter fiber pathway.

6. The three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury according to claim 5, characterized in that, The functional network mapping module maps the damage propagation path based on structural connectivity to functional network nodes in a standard brain atlas, identifies the set of functional network nodes affected by the damage propagation path, and generates key prognostic brain network identifiers, specifically including: The fiber bundles contained in the damage propagation path based on structural connectivity are spatially registered with the standard white matter fiber atlas. Identify the cerebral cortex or deep nucleus regions connected by the fiber bundles, map the cerebral cortex or deep nucleus regions to a standard brain functional network atlas, and obtain the corresponding functional network node numbers; Collect all affected functional network node numbers, and form a set of functional network nodes after deduplication; Query the preset functional network type to which each node in the set of functional network nodes belongs, estimate the connection strength loss ratio based on historical data, and generate a prognostic key brain network identifier.

7. The three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury according to claim 6, characterized in that, The loaded three-dimensional medical image sequence of the cranium is used to separate the voxel set of the core trauma region using adaptive threshold segmentation and region growing algorithms, specifically including: Read the grayscale value data of each voxel in a three-dimensional medical image sequence of the brain; A gray-level histogram is generated based on the gray-level value distribution of the entire image sequence, and an adaptive segmentation threshold is dynamically calculated based on the peak and valley values ​​of the histogram. Using voxels with gray values ​​higher than the adaptive segmentation threshold as initial seed points, the region growing algorithm iteratively checks the adjacent voxels of each seed point. If the difference between the gray value of an adjacent voxel and the gray value of the seed point is less than a preset similarity tolerance, the adjacent voxels are merged into the current region. Repeat the iterative process until no new voxels are added, and obtain the set of connected voxels in the core region of the trauma.

8. The three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury according to claim 7, characterized in that, The step of calculating the changes in each parameter between adjacent time points based on the total volume, surface area, and sphericity values ​​of the trauma core region at different time points specifically includes: For each acquisition time point, the total volume is obtained by multiplying the total number of all voxels in the core trauma region by the volume of a single voxel. The sum of the outer surface areas of the boundary voxels of the trauma core region is calculated as the surface area; The sphericity index is calculated based on the volume and surface area of ​​the core trauma region, and is expressed as the ratio of the region's volume to the surface area of ​​a sphere of the same volume. The difference in total volume between adjacent time points is extracted as the volume change, the difference in surface area is extracted as the surface area change, and the difference in sphericity is extracted as the sphericity change.

9. The three-dimensional model-based prognostic prediction and analysis system for traumatic brain injury according to claim 8, characterized in that, The process of reconstructing the main nerve fiber bundles passing through the normal brain tissue region from the diffusion tensor imaging data, and calculating the anisotropy fraction and mean diffusion rate of each fiber bundle, specifically includes: Preprocessing operations are applied to diffusion tensor imaging data, including eddy current distortion correction, head motion artifact correction, and brain tissue extraction. Within the normal brain tissue region, the main diffusion direction is determined based on the eigenvector of the diffusion tensor matrix of each voxel, and a fiber tract tracing algorithm is used to gradually connect adjacent voxels along the main direction to generate a continuous fiber tract path. For each reconstructed nerve fiber bundle, the eigenvalues ​​of the diffusion tensor are extracted to calculate the anisotropy fraction, which reflects the direction dependence of water molecule diffusion. The arithmetic mean of the three eigenvalues ​​of the diffusion tensor is calculated as the average diffusion rate of the nerve fiber bundle.

Images