A method and system for identifying intracranial aneurysms based on brain CT

Abstract

The application relates to the technical field of medical image processing, and discloses an intracranial aneurysm recognition method and system based on a brain CT. The method obtains a brain CT image sequence, adopts an anisotropic diffusion filtering algorithm for pretreatment to suppress noise and retain blood vessel edge details. Subsequently, a region growing algorithm is used to segment an intracranial blood vessel region, and a skeletonization algorithm is used to extract a blood vessel center line to calculate curvature and diameter changes. Finally, based on curvature anomaly and diameter ratio analysis, combined with a multi-scale sliding window and a blood vessel topological structure verification, automatic recognition of intracranial aneurysms is realized, and the accuracy and reliability of diagnosis are improved.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a method and system for identifying intracranial aneurysms based on cranial CT. Background Technology

[0002] In the field of medical image processing, accurate identification of intracranial aneurysms is crucial for preventing serious diseases such as cerebral hemorrhage. Traditional methods rely on doctors manually interpreting cranial CT images, which is not only time-consuming and labor-intensive but also susceptible to subjective factors, leading to missed or misdiagnosed cases. While existing automatic identification technologies can improve efficiency, they still fall short in handling complex vascular structures, noise interference, and detecting small aneurysms. In particular, there is a lack of effective automated solutions to address the problems of blurred vessel edges, low contrast, and high false positives at vessel bifurcation in cranial CT images. Therefore, developing a technology capable of automatically and accurately identifying intracranial aneurysms has become an urgent need in the field of medical image processing. Summary of the Invention

[0003] To address the aforementioned technical shortcomings, the purpose of this invention is to provide a method and system for identifying intracranial aneurysms based on cranial CT scans, thereby solving the problems of low accuracy and insufficient automation in existing intracranial aneurysm identification methods.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for identifying intracranial aneurysms based on cranial CT scans, the method comprising: Acquire brain CT image sequences; The cranial CT image is preprocessed using an anisotropic diffusion filtering algorithm, which adaptively adjusts the diffusion intensity based on the local gray-level gradient of the cranial CT image to suppress noise while preserving vascular edge details. Based on preprocessed cranial CT images, intracranial vascular regions are segmented using a region growing algorithm. Seed points for region growing are automatically selected based on the enhancement features of vascular structures, and the growth criteria combine pixel gray-level similarity and spatial continuity. The vascular centerline is extracted from the segmented vascular region. The vascular structure is iteratively refined using a skeletonization algorithm to obtain the centerline. The curvature and diameter changes along the centerline are then calculated. Intracranial aneurysms were identified based on curvature anomaly and diameter ratio analysis. This involved detecting local curvature maxima and diameter change rates using a multi-scale sliding window, and verifying candidate intracranial aneurysm sites by combining vascular topology.

[0005] Preferably, in one possible implementation of the first aspect, the anisotropic diffusion filtering algorithm includes: Construct a diffusion tensor based on the local structure of the image. This diffusion tensor adaptively adjusts the diffusion direction and intensity according to the gray-level gradient of the pixel and the eigenvalues ​​of the Hessian matrix. Multi-scale spatial analysis is introduced, and local gradient magnitudes are calculated at different scales using Gaussian scale spatial theory. Multi-scale information is then integrated to optimize the diffusion process. The diffusion coefficient is expressed as an adaptive nonlinear function, and its expression is:

[0006] in, Represents the magnitude of the image gradient. The scale parameter is calculated based on the gray-level variance within a local window centered on the pixel. This is a contrast-sensitive parameter.

[0007] Preferably, in one possible implementation of the first aspect, the region growing algorithm includes: Based on the preprocessed cranial CT images, seed point selection is based on the vascular enhancement filter response, tubular structures are identified through Hessian matrix feature analysis, and vascular features are enhanced using multi-scale filtering. The growth criteria combine pixel grayscale similarity and spatial continuity. Grayscale similarity uses a dynamic threshold, which is calculated based on the statistical characteristics of the seed point's neighborhood and is adaptively updated during the growth process. Spatial continuity ensures the topological integrity of the vascular region by checking the connectivity and geometric constraints of pixels. The growth process is iterative, and a queue is used to manage the growth frontier.

[0008] Preferably, in one possible implementation of the first aspect, the dynamic threshold employs a weighted function based on local grayscale statistics and spatial distance, the expression of which is:

[0009] in, The threshold for the current position. This is the weighted average gray level of the neighboring pixels. For weighted standard deviation, To adjust the parameters; The weighted gray mean and weighted standard deviation The calculation formula is:

[0010]

[0011] in, The number of neighboring pixels. For the neighboring region grayscale value of each pixel. For neighboring pixel index, These are the weighting coefficients. , Let be the Euclidean distance from the pixel to the seed point. This is the distance attenuation parameter.

[0012] Preferably, in one possible implementation of the first aspect, the skeletonization algorithm employs a level set method based on curve evolution, iteratively refining the vascular structure by solving partial differential equations, the equations being defined as:

[0013] in, For level set functions, The gradient of the level set function. The edge stopping function is defined as follows: , Let be the gradient magnitude of the level set function. For divergence operators, These are the parameters of the balloon force. During the evolution process, the centerline is extracted through the zero level set, and the property of the signed distance function is preserved by re-initialization, finally obtaining the blood vessel centerline with a single pixel width.

[0014] Preferably, in one possible implementation of the first aspect, the calculation of the curvature change and diameter change along the centerline includes: First, parameterize the centerline as a curve. ,in For arc length parameters; curvature Calculated using the following formula:

[0015] in and curves The first and second derivatives; The diameter variation is obtained by calculating the diameter of the blood vessel cross-section at each center point. The cross-section is obtained by intersecting the center point plane with the segmented blood vessel region. The diameter is defined as the diameter of the largest inscribed circle of the intersecting region and is calculated as a local maximum of the distance transformation.

[0016] Preferably, in one possible implementation of the first aspect, the identification of intracranial aneurysms includes: Based on curvature anomaly and diameter ratio analysis, a multi-scale sliding window scan of the centerline is used, with the window size adaptively adjusted according to the vessel diameter. Within each window, calculate the local curvature maxima and the diameter change rate, where the diameter change rate is defined as the ratio of the current point's diameter to the average diameter upstream and downstream. The candidate point selection criteria are that the curvature exceeds a preset curvature threshold and the diameter ratio is greater than a preset diameter ratio threshold. Candidate points are validated by combining vascular topology. The vascular network is analyzed using graph theory methods, and candidate points located near vascular bifurcation nodes are further discriminated.

[0017] Preferably, in one possible implementation of the first aspect, the secondary discrimination includes: For candidate points located near the bifurcation node of a blood vessel, calculate the branching angle and the ratio of the branching diameter of the blood vessel branches at the candidate point; The branch angle is compared with the preset normal bifurcation angle threshold, and the branch diameter ratio is compared with the preset normal bifurcation diameter ratio threshold. If the branch angle is within the normal bifurcation angle range and the branch diameter ratio conforms to the normal bifurcation pattern, then the candidate point is determined to be a false positive and is excluded. Otherwise, the candidate site is confirmed as a candidate site for intracranial aneurysm.

[0018] Preferably, in one possible implementation of the first aspect, the diameter ratio calculation employs a multi-scale fusion method, defining the comprehensive diameter ratio as:

[0019] in, The diameter of the current point. The reference diameter is determined in the following way: Select a stable vessel segment at a location at least 3 times the vessel diameter upstream of the current point, and calculate the average diameter of this segment as the reference diameter. If the upstream reference diameter cannot be obtained, use the median of the global vessel diameter distribution as the reference.

[0020] In a second aspect, the present invention provides an intracranial aneurysm identification system based on cranial CT, the system being used to implement the intracranial aneurysm identification method based on cranial CT as described in the first aspect, comprising: The image acquisition module acquires sequences of cranial CT images; The diffusion filtering module preprocesses the cranial CT image. The preprocessing uses an anisotropic diffusion filtering algorithm, which adaptively adjusts the diffusion intensity according to the local gray-level gradient of the cranial CT image to suppress noise while preserving the details of the blood vessel edges. The blood vessel segmentation module, based on preprocessed cranial CT images, segments intracranial blood vessel regions using a region growing algorithm. Seed points for region growing are automatically selected based on the enhancement features of blood vessel structures, and the growth criteria combine pixel gray-level similarity and spatial continuity. The feature extraction module extracts the vascular centerline from the segmented vascular region, uses a skeletonization algorithm to iteratively refine the vascular structure to obtain the centerline, and calculates the curvature and diameter changes along the centerline. The lesion identification module identifies intracranial aneurysms based on curvature abnormalities and diameter ratio analysis. It detects local curvature maxima and diameter change rates through multi-scale sliding windows and verifies intracranial aneurysm candidate points by combining vascular topology.

[0021] The beneficial effects of this invention are as follows: This invention employs image recognition technology to intelligently identify intracranial aneurysms. Through an anisotropic diffusion filtering algorithm, noise in cranial CT images is effectively suppressed while preserving vascular edge details. A region growing algorithm, combining pixel grayscale similarity and spatial continuity, achieves accurate segmentation of intracranial vascular regions. A skeletonization algorithm iteratively refines the vascular structure, accurately extracting the vascular centerline and calculating curvature and diameter changes, providing key features for aneurysm identification. Based on curvature anomaly and diameter ratio analysis, combined with multi-scale sliding window and vascular topology verification, the false positive rate is effectively reduced, and the identification accuracy is improved.

[0022] This system not only reduces the workload of doctors, but also improves the accuracy and reliability of diagnosis, and has important value for the clinical treatment and prevention of serious diseases such as cerebral hemorrhage. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This application provides a flowchart of a method for identifying intracranial aneurysms based on cranial CT.

[0025] Figure 2 This application provides a structural diagram of an intracranial aneurysm identification system based on cranial CT.

[0026] Figure labels: 1-Image acquisition module, 2-Diffusion filtering module, 3-Vascular segmentation module, 4-Feature extraction module, 5-Lesion recognition module. Detailed Implementation

[0027] 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.

[0028] Example 1: As Figure 1 As shown, the present invention provides a method for intracranial aneurysm identification based on cranial CT, comprising: Obtain CT image sequences of the brain.

[0029] In this embodiment, the cranial CT image sequence is sourced from routine or contrast-enhanced clinical CT scanning equipment. The image sequence covers the entire cranial region, including continuous cross-sectional images from the base to the top of the skull. First, the metadata of the image sequence is verified, including pixel spacing, slice spacing, and scanning parameters, to ensure compliance with the basic requirements for intracranial vascular analysis. For multi-phase scanning data, arterial phase image sequences are preferentially selected to optimize the visualization of vascular structures. The system records the original information of the image sequence, including patient identification, scan date, and equipment model, for traceability.

[0030] The cranial CT image is preprocessed using an anisotropic diffusion filtering algorithm, which adaptively adjusts the diffusion intensity based on the local gray-level gradient of the cranial CT image to suppress noise while preserving the details of the blood vessel edges.

[0031] In this embodiment, the preprocessing of cranial CT images employs an anisotropic diffusion filtering algorithm to suppress image noise while preserving vascular edge details. This algorithm is based on partial differential equation theory, constructing a diffusion tensor based on the local structure of the image to achieve an adaptive diffusion process. The diffusion tensor depends on the gray-level gradient of each pixel and the eigenvalues ​​of the Hessian matrix, thereby dynamically adjusting the diffusion direction and intensity. Specifically, for each pixel, its gray-level gradient vector and Hessian matrix are calculated. The Hessian matrix is ​​composed of the second derivative of the image. By analyzing its eigenvalues, local structural features can be identified: for example, in edge regions, the eigenvalues ​​are large and directional, so the diffusion tensor is adjusted to enhance diffusion along the edge tangent direction to smooth noise, while weakening diffusion along the normal direction to preserve edge details; in flat regions, the eigenvalues ​​are small, and the diffusion tensor tends to be isotropic to promote uniform smoothing.

[0032] Furthermore, multi-scale spatial analysis is introduced to optimize the diffusion process. Gaussian scale space theory is employed to calculate local gradient magnitudes at different scales, thereby fusing multi-scale information to improve the algorithm's robustness. Specifically, a multi-scale image pyramid is generated by convolving the image with a series of Gaussian kernel functions, each scale corresponding to a different spatial resolution. Scale parameters... The gray-level variance within a local window centered on each pixel is calculated, reflecting local contrast variations. At multiple scales, the gradient magnitude of each pixel is calculated, and a weighted fusion strategy is employed to integrate gradient information from different scales, such as through linear interpolation or maximum response selection, ensuring both large-scale noise suppression and preservation of subtle vascular structures. This multi-scale approach enhances the algorithm's adaptability to local image features, avoiding edge blurring or noise residue issues that may result from single-scale analysis.

[0033] The diffusion coefficient is expressed as an adaptive nonlinear function, and its expression is:

[0034] in, Represents the magnitude of the image gradient. The scale parameter is calculated based on the gray-level variance within a local window centered on the pixel. This is a contrast-sensitive parameter. This function ensures dynamic adjustment of the diffusion coefficient: when the gradient magnitude is small, it indicates a flat region or noise dominance, and the diffusion coefficient approaches 1, promoting strong diffusion and smoothing noise; when the gradient magnitude is large, it indicates the presence of edges or details, and the diffusion coefficient decays exponentially, suppressing diffusion and protecting edges. Parameters The value used to adjust the algorithm's sensitivity to contrast was set experimentally and optimized based on the grayscale distribution range of typical cranial CT images, balancing noise suppression and edge preservation. Scale parameter The calculation involves selecting a local window, the size of which is set according to the image resolution, using a 3×3 or 5×5 pixel area. Local contrast is estimated by calculating the variance of pixel grayscale within the window, thereby... It can adapt to image content.

[0035] In the algorithm implementation, an iterative process is used to gradually optimize the image quality. In each iteration, the diffusion tensor and diffusion coefficient are updated according to the current image state, and the image evolution is solved by discretized partial differential equations. The number of iterations is preset according to the initial noise level of the image and application requirements; in this embodiment, it is 10 iterations to ensure convergence without excessive smoothing.

[0036] Based on preprocessed cranial CT images, intracranial vascular regions are segmented using a region growing algorithm. Seed points for region growing are automatically selected based on the enhancement features of vascular structures, and the growth criteria combine pixel gray-level similarity and spatial continuity.

[0037] In this embodiment, a region growing algorithm is used to segment intracranial vascular regions, automatically selecting seed points and accurately extracting vascular structures based on intelligent growth criteria. Automatic seed point selection is based on vascular enhancement filtering responses, identifying tubular structures through Hessian matrix feature analysis. Specifically, the preprocessed cranial CT image is first enhanced with multi-scale filtering to improve vascular features. A Gaussian kernel function is used to construct a scale space, and the Hessian matrix of the image is calculated at different scales. For each pixel, eigenvalue analysis of the Hessian matrix is ​​used to distinguish between tubular structures, plaque structures, and background noise. Candidate vascular points are selected by calculating the ratio and magnitude of eigenvalues ​​and setting a threshold. The eigenvalues ​​corresponding to tubular structures meet specific conditions: one eigenvalue is significantly smaller than the other two and has a negative sign. Multi-scale filtering enhances the continuity of vascular structures and suppresses noise interference by integrating response maps at different scales. Finally, the point with the highest response value is selected as the initial seed point for region growing.

[0038] The growth criterion combines pixel grayscale similarity and spatial continuity to ensure the accuracy and topological integrity of the segmented regions. Grayscale similarity employs a dynamic thresholding mechanism, which is calculated based on the statistical characteristics of the seed point's neighborhood and adaptively updated during the growth process. The expression for the dynamic threshold is:

[0039] in, The threshold for the current position. This is the weighted average gray level of the neighboring pixels. For weighted standard deviation, To adjust the parameters. The weighted grayscale mean. and weighted standard deviation The calculation formula is:

[0040]

[0041] in, The number of neighboring pixels. For the neighboring region grayscale value of each pixel. For neighboring pixel index, These are the weighting coefficients. , Let be the Euclidean distance from the pixel to the seed point. This is the distance decay parameter. This weighting function assigns higher weights to pixels closer to the seed point, thus more accurately reflecting the local grayscale distribution. The dynamic threshold is updated as the growth front moves, avoiding oversegmentation or undersegmentation problems caused by the global threshold.

[0042] Spatial continuity is maintained by checking pixel connectivity and geometric constraints to preserve the topology of the vascular region. During growth, each candidate pixel must satisfy either 8-connectivity or 26-connectivity conditions to ensure the continuity of the vascular segment without breaks. Simultaneously, geometric constraints such as the vascular diameter range and smoothness priors are introduced to eliminate isolated noise points. The growth process is iterative, using a queue data structure to manage the growth front, prioritizing pixels with high gray-level similarity to the current region, i.e., points with small dynamic threshold differences. Specifically, the queue is initialized with seed points. Each time, a pixel is taken from the head of the queue, and its neighboring pixels are checked. If they meet the gray-level similarity condition and spatial connectivity requirements, they are added to the tail of the queue and marked as a vascular region. Queue priority is based on similarity scores, with highly similar pixels processed first to improve algorithm efficiency and reduce error accumulation. The iteration continues until the queue is empty, completing the segmentation of the entire vascular network and finally outputting a binarized image of the vascular region.

[0043] The vascular centerline is extracted from the segmented vascular region. The vascular structure is iteratively refined using a skeletonization algorithm to obtain the centerline, and the curvature and diameter changes along the centerline are calculated.

[0044] In this embodiment, the skeletonization algorithm employs a level set method based on curve evolution. It iteratively refines the vascular structure by solving partial differential equations, extracting the centerline of a single pixel width. The level set method introduces a level set function. The boundary of the blood vessel is represented by a partial differential equation, the evolution of which is governed by the equation defined as:

[0045] in, For level set functions, The gradient of the level set function. The edge stopping function is defined as follows: , Let be the gradient magnitude of the level set function. For divergence operators, For balloon force parameters. Edge stopping function. Its function is to stop evolution at the edge of the blood vessel, preserving detailed structure, while the balloon force parameter The evolution speed is controlled to ensure the centerline converges towards the interior of the blood vessel. The evolution process begins with the initial blood vessel segmentation region and is solved iteratively using discretized equations, updating the level set function value in each iteration. To maintain numerical stability, a re-initialization technique is used to preserve the level set function as a signed distance function, avoiding distortion during the evolution process. The evolution termination condition is based on the centerline width converging to a single pixel, ultimately achieved by extracting the zero level set. The isosurface is used to obtain the centerline of the blood vessel. This centerline is a continuous curve, representing the geometric framework of the blood vessel.

[0046] After obtaining the vascular centerline, the centerline is first parameterized and represented as a parametric curve. ,in The arc length parameter ensures uniform and continuous parameterization. The calculation starts from the centerline and is obtained by accumulating Euclidean distances, thus ensuring that the parameterization is independent of the curve geometry. Based on the parameterized curve, the curvature change along the centerline is calculated, and the curvature... The degree of local curvature of the centerline is characterized by the following formula:

[0047] in and curves The first and second derivatives of represent the tangent vector and the curvature vector, respectively. First derivative The second derivative is calculated using the central difference method. A second-order central difference approximation is used to ensure numerical stability. Curvature calculations are performed at each center point to obtain a continuous curvature distribution.

[0048] The diameter variation is obtained by calculating the diameter of the blood vessel cross-section at each center point. For each center point, its normal plane is first determined, which is perpendicular to the tangent vector at that point. The intersection of the normal plane and the segmented blood vessel region yields a cross-sectional area, which is represented by a binary image of the blood vessel at that point. The diameter is defined as the diameter of the largest inscribed circle in the cross-sectional area, calculated using a distance transform method: a Euclidean distance transform is performed on the cross-sectional area to obtain the shortest distance from each pixel to the boundary; then, a local maximum of the distance transform is found, which is the radius of the largest inscribed circle, twice the diameter. This method accurately reflects local variations in blood vessel diameter, and is particularly suitable for non-circular cross-sections.

[0049] Intracranial aneurysms were identified based on curvature anomaly and diameter ratio analysis. This involved detecting local curvature maxima and diameter change rates using a multi-scale sliding window, and verifying candidate intracranial aneurysm sites by combining vascular topology.

[0050] In this embodiment, firstly, a multi-scale sliding window is used to scan the vessel centerline, with the window size adaptively adjusted according to the local vessel diameter. The window size is set to an integer multiple of the current point vessel diameter to ensure that the window can cover sufficient vessel segments to reflect local feature changes. For each window, the local curvature maxima and diameter change rate within the window are calculated. The local curvature maxima are determined by detecting the curvature function. The local peak value is used to determine the peak value. The peak value must satisfy the condition that the first derivative is zero and the second derivative is negative to ensure that it is a local maximum. The diameter change rate is defined as the ratio of the diameter at the current point to the average diameter upstream and downstream. The average diameter upstream and downstream is calculated by taking the average diameter of a segment upstream and downstream of the current point. The segment length is usually half the window size to smooth out local fluctuations.

[0051] The diameter change rate was calculated using a multi-scale fusion method, and the overall diameter ratio was defined as:

[0052] in, The diameter of the current point. The reference diameter is determined as follows: a stable vessel segment is selected at least three times the vessel diameter upstream of the current point. This segment should avoid bifurcation or bends. The average diameter of this segment is calculated as the reference diameter. For cases where an upstream reference diameter cannot be obtained, such as at the vessel initiation point or image boundary, the median of the global vessel diameter distribution is used as the reference diameter to ensure robustness. Diameter variation rate. It reflects the degree of local vascular dilation. A value greater than 1 indicates the possible presence of aneurysmal dilatation.

[0053] Candidate point selection criteria are based on curvature anomalies and diameter ratio analysis. A curvature threshold is set. and diameter ratio threshold These thresholds are optimized through training with clinical data. Based on the percentile of the normal blood vessel curvature distribution, Determined based on the range of normal blood vessel diameter variation. Candidate points must simultaneously satisfy the curvature. Exceed and diameter ratio Greater than The initial candidate points may include bends or bifurcation points in blood vessels, so further verification based on blood vessel topology is required.

[0054] The vascular topology verification method analyzes the vascular network using graph theory. The vascular centerline is abstracted as a graph structure, where nodes represent center points, edges represent vascular segments, and branching nodes are defined as nodes connected by three or more edges. For each candidate point, its Euclidean distance to the nearest branching node is calculated. If the distance is less than a preset value, it is marked as being near a branching node and requires secondary discrimination.

[0055] The secondary discrimination aims to eliminate false positives caused by normal bifurcation. For candidate points located near the bifurcation node, the branching angle and branch diameter ratio of the vessel branches at the candidate point are calculated. The branching angle is defined as the angle between the direction vectors of each branch at the bifurcation point, calculated by the dot product of the branch tangent vectors. The branch diameter ratio is defined as the ratio of the branch diameter to the diameter of the parent vessel, where the parent vessel diameter is the average diameter of the upstream segment of the vessel at the bifurcation point.

[0056] Branch angle Compared with the preset normal bifurcation angle threshold and In comparison, the normal bifurcation angle range is defined based on anatomical knowledge. Branch diameter ratio. The ratio threshold of the normal bifurcation diameter and In contrast, the normal branching pattern requires the branch diameter ratio to conform to Murray's law.

[0057] If branch angle In to Within the range and the branch diameter ratio Conforms to a normal fork pattern (i.e.) If the result is negative, the candidate point is considered a false positive and excluded. Otherwise, the candidate point is confirmed as an intracranial aneurysm candidate point, and its location and characteristic parameters are recorded.

[0058] The entire identification process is iterative until all centerline points have been scanned. The final candidate point list is post-processed, such as by non-maximum suppression and merging adjacent candidate points, to generate the final intracranial aneurysm detection result. This method effectively improves the accuracy and reliability of detection and reduces the false alarm rate through multi-scale analysis and topological validation.

[0059] Example 2: Figure 2 As shown, the present invention provides an intracranial aneurysm identification system based on cranial CT, comprising: Image acquisition module 1 acquires a sequence of cranial CT images; The diffusion filtering module 2 preprocesses the cranial CT image. The preprocessing uses an anisotropic diffusion filtering algorithm, which adaptively adjusts the diffusion intensity according to the local gray-level gradient of the cranial CT image to suppress noise while preserving the details of the blood vessel edges. The vascular segmentation module 3, based on the preprocessed cranial CT image, segments the intracranial vascular region using a region growing algorithm. The seed points for region growing are automatically selected based on the enhancement features of the vascular structure, and the growth criteria combine pixel gray-level similarity and spatial continuity. Feature extraction module 4 extracts the vascular centerline from the segmented vascular region, uses a skeletonization algorithm to iteratively refine the vascular structure to obtain the centerline, and calculates the curvature change and diameter change along the centerline; The lesion identification module 5 identifies intracranial aneurysms based on curvature abnormality and diameter ratio analysis. It detects local curvature maxima and diameter change rates through multi-scale sliding windows and verifies intracranial aneurysm candidate points by combining vascular topology.

[0060] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for identifying intracranial aneurysms based on cranial CT, characterized in that, The method includes: Acquire brain CT image sequences; The cranial CT image is preprocessed using an anisotropic diffusion filtering algorithm, which adaptively adjusts the diffusion intensity based on the local gray-level gradient of the cranial CT image to suppress noise while preserving vascular edge details. Based on preprocessed cranial CT images, intracranial vascular regions are segmented using a region growing algorithm. Seed points for region growing are automatically selected based on the enhancement features of vascular structures, and the growth criteria combine pixel gray-level similarity and spatial continuity. The vascular centerline is extracted from the segmented vascular region. The vascular structure is iteratively refined using a skeletonization algorithm to obtain the centerline. The curvature and diameter changes along the centerline are then calculated. Intracranial aneurysms were identified based on curvature anomaly and diameter ratio analysis. This involved detecting local curvature maxima and diameter change rates using a multi-scale sliding window, and verifying candidate intracranial aneurysm sites by combining vascular topology.

2. The method for identifying intracranial aneurysms based on cranial CT as described in claim 1, characterized in that, The anisotropic diffusion filtering algorithm includes: Construct a diffusion tensor based on the local structure of the image. This diffusion tensor adaptively adjusts the diffusion direction and intensity according to the gray-level gradient of the pixel and the eigenvalues ​​of the Hessian matrix. Multi-scale spatial analysis is introduced, and local gradient magnitudes are calculated at different scales using Gaussian scale spatial theory. Multi-scale information is then integrated to optimize the diffusion process. The diffusion coefficient is expressed as an adaptive nonlinear function, and its expression is: in, Represents the magnitude of the image gradient. The scale parameter is calculated based on the gray-level variance within a local window centered on the pixel. This is a contrast-sensitive parameter.

3. The method for identifying intracranial aneurysms based on cranial CT as described in claim 1, characterized in that, The region growing algorithm includes: Based on the preprocessed cranial CT images, seed point selection is based on the vascular enhancement filter response, tubular structures are identified through Hessian matrix feature analysis, and vascular features are enhanced using multi-scale filtering. The growth criteria combine pixel grayscale similarity and spatial continuity. Grayscale similarity uses a dynamic threshold, which is calculated based on the statistical characteristics of the seed point's neighborhood and is adaptively updated during the growth process. Spatial continuity ensures the topological integrity of the vascular region by checking the connectivity and geometric constraints of pixels. The growth process is iterative, and a queue is used to manage the growth frontier.

4. The method for identifying intracranial aneurysms based on cranial CT as described in claim 3, characterized in that, The dynamic threshold adopts a weighted function based on local gray-level statistics and spatial distance, and its expression is: in, The threshold for the current position. This is the weighted average gray level of the neighboring pixels. For weighted standard deviation, To adjust the parameters; The weighted gray mean and weighted standard deviation The calculation formula is: in, The number of neighboring pixels. For the neighboring region grayscale value of each pixel. For neighboring pixel index, These are the weighting coefficients. , Let be the Euclidean distance from the pixel to the seed point. This is the distance attenuation parameter.

5. The method for identifying intracranial aneurysms based on cranial CT as described in claim 1, characterized in that, The skeletonization algorithm employs a level set method based on curve evolution, iteratively refining the vascular structure by solving partial differential equations, defined as follows: in, For level set functions, The gradient of the level set function. The edge stopping function is defined as follows: , Let be the gradient magnitude of the level set function. For divergence operators, These are the parameters of the balloon force. During the evolution process, the centerline is extracted through the zero level set, and the property of the signed distance function is preserved by re-initialization, finally obtaining the blood vessel centerline with a single pixel width.

6. The method for identifying intracranial aneurysms based on cranial CT as described in claim 5, characterized in that, The calculation of the curvature and diameter changes along the centerline includes: First, parameterize the centerline as a curve. ,in For arc length parameters; curvature Calculated using the following formula: in and curves The first and second derivatives; The diameter variation is obtained by calculating the diameter of the blood vessel cross-section at each center point. The cross-section is obtained by intersecting the center point plane with the segmented blood vessel region. The diameter is defined as the diameter of the largest inscribed circle of the intersecting region and is calculated as a local maximum of the distance transformation.

7. The method for intracranial aneurysm identification based on cranial CT as described in claim 6, characterized in that, The identification of intracranial aneurysms includes: Based on curvature anomaly and diameter ratio analysis, a multi-scale sliding window scan of the centerline is used, with the window size adaptively adjusted according to the vessel diameter. Within each window, calculate the local curvature maxima and the diameter change rate, where the diameter change rate is defined as the ratio of the current point's diameter to the average diameter upstream and downstream. The candidate point selection criteria are that the curvature exceeds a preset curvature threshold and the diameter ratio is greater than a preset diameter ratio threshold. Candidate points are validated by combining vascular topology. The vascular network is analyzed using graph theory methods, and candidate points located near vascular bifurcation nodes are further discriminated.

8. The method for identifying intracranial aneurysms based on cranial CT as described in claim 7, characterized in that, The secondary discrimination includes: For candidate points located near the bifurcation node of a blood vessel, calculate the branching angle and the ratio of the branching diameter of the blood vessel branches at the candidate point; The branch angle is compared with the preset normal bifurcation angle threshold, and the branch diameter ratio is compared with the preset normal bifurcation diameter ratio threshold. If the branch angle is within the normal bifurcation angle range and the branch diameter ratio conforms to the normal bifurcation pattern, then the candidate point is determined to be a false positive and is excluded. Otherwise, the candidate site is confirmed as a candidate site for intracranial aneurysm.

9. The method for identifying intracranial aneurysms based on cranial CT as described in claim 8, characterized in that, The diameter ratio calculation employs a multi-scale fusion method, defining the comprehensive diameter ratio as: in, The diameter of the current point. The reference diameter is determined in the following way: Select a stable vessel segment at a location at least 3 times the vessel diameter upstream of the current point, and calculate the average diameter of this segment as the reference diameter. If the upstream reference diameter cannot be obtained, use the median of the global vessel diameter distribution as the reference.

10. A system for identifying intracranial aneurysms based on cranial CT scans, characterized in that, The system is used to implement a method for identifying intracranial aneurysms based on cranial CT as described in any one of claims 1 to 9, comprising: The image acquisition module acquires sequences of cranial CT images; The diffusion filtering module preprocesses the cranial CT image. The preprocessing uses an anisotropic diffusion filtering algorithm, which adaptively adjusts the diffusion intensity according to the local gray-level gradient of the cranial CT image to suppress noise while preserving the details of the blood vessel edges. The blood vessel segmentation module, based on preprocessed cranial CT images, segments intracranial blood vessel regions using a region growing algorithm. Seed points for region growing are automatically selected based on the enhancement features of blood vessel structures, and the growth criteria combine pixel gray-level similarity and spatial continuity. The feature extraction module extracts the vascular centerline from the segmented vascular region, uses a skeletonization algorithm to iteratively refine the vascular structure to obtain the centerline, and calculates the curvature and diameter changes along the centerline. The lesion identification module identifies intracranial aneurysms based on curvature abnormalities and diameter ratio analysis. It detects local curvature maxima and diameter change rates through multi-scale sliding windows and verifies intracranial aneurysm candidate points by combining vascular topology.

Images