A cerebral artery dissection detection system based on image processing

Abstract

The application relates to the technical field of cerebral artery detection, and discloses a cerebral artery dissection detection system based on image processing. The system comprises the following steps: acquiring cerebral medical image data through an image data acquisition module and performing standardization and calibration to generate an original image data set; performing image enhancement and noise filtering on the original image data set through an image preprocessing module to obtain preprocessed image data; calculating blood vessel morphological features and blood flow features according to the preprocessed image data through a feature extraction module to generate blood vessel feature extraction results; analyzing the blood vessel feature extraction results through a dissection detection module adopting a pattern recognition technology to identify a cerebral artery dissection area to obtain an initial dissection detection result; quantifying dissection geometric properties and spatial distribution parameters according to the initial detection result through a parameter analysis module to generate dissection parameter analysis results; and adjusting detection threshold values and algorithm parameters based on the analysis results through a detection optimization module to generate optimized dissection detection results.

Description

Technical Field

[0001] This invention relates to the field of cerebral artery detection technology, specifically to a cerebral artery dissection detection system based on image processing. Background Technology

[0002] Cerebral artery dissection is a potentially dangerous cerebrovascular disease. Its pathogenesis is mainly related to factors such as abnormalities in the structure of the cerebral artery wall and changes in hemodynamics. If it is not detected and intervened in a timely and accurate manner, it may lead to serious complications such as cerebral infarction and cerebral hemorrhage, posing a significant threat to the patient's life and health. Currently, the clinical detection of cerebral artery dissection mainly relies on traditional medical imaging examinations, such as computed tomography angiography and magnetic resonance angiography. Although these technologies can provide imaging information of cerebral blood vessels, they have many limitations in practical applications.

[0003] Traditional imaging examinations are susceptible to interference from various factors, such as patient positioning during scanning, fluctuations in equipment imaging parameters, and artifacts in the surrounding tissues. This results in unstable image quality, making it difficult to clearly depict subtle structural changes in blood vessels. This poses a significant challenge for physicians in identifying early or minor cerebral artery dissections. Secondly, traditional methods rely heavily on manual analysis and interpretation of the acquired image data. Physicians must subjectively assess vascular morphology and blood flow in the images based on their professional experience. This approach is not only time-consuming and labor-intensive, but the results are also easily influenced by subjective factors such as individual physician experience and fatigue levels, making it difficult to guarantee consistency and accuracy.

[0004] Even when potential dissection areas are initially identified manually, traditional methods lack effective means to accurately quantify and analyze the geometric properties and spatial distribution parameters of the dissection, failing to provide detailed and reliable evidence for subsequent disease assessment and treatment planning. Furthermore, traditional testing procedures lack dynamic optimization mechanisms; once testing parameters and methods are determined, timely adjustments based on actual test results are impossible, resulting in poor adaptability and flexibility of the testing system, making it difficult to meet the complex testing needs of different patients and conditions. These problems severely impact the efficiency and quality of cerebral artery dissection detection, hindering early diagnosis and effective treatment of this disease in clinical practice. Summary of the Invention

[0005] The purpose of this invention is to provide an image processing-based cerebral artery dissection detection system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an image processing-based cerebral artery dissection detection system, the system comprising:

[0007] The image data acquisition module acquires brain medical image data, performs standardization and calibration on the image data, and generates a raw image dataset.

[0008] The image preprocessing module performs image enhancement and noise filtering operations based on the original image dataset to obtain preprocessed image data;

[0009] The feature extraction module calculates vascular morphology features and blood flow features based on the preprocessed image data, and generates vascular feature extraction results.

[0010] The dissection detection module, based on the extracted vascular features, identifies the cerebral artery dissection region using pattern recognition technology to obtain initial dissection detection results;

[0011] The parameter analysis module quantifies the geometric properties and spatial distribution parameters of the interlayer based on the initial interlayer detection results, and generates interlayer parameter analysis results.

[0012] The detection optimization module adjusts the detection threshold and algorithm parameters based on the analysis results of the interlayer parameters to generate optimized interlayer detection results.

[0013] Preferably, the original image dataset includes image resolution normalization values, image grayscale calibration sequences, and acquisition timestamp indexes; the preprocessed image data includes image enhancement intensity values, noise filtering coefficients, and image quality assessment indices; the vascular feature extraction results include vascular wall thickness measurements, blood flow velocity estimates, and vascular branch point location data; the initial dissection detection results include dissection region contour data, dissection confidence scores, and suspected region marker sequences; the dissection parameter analysis results include dissection length measurements, dissection width measurements, and dissection azimuth data; and the optimized dissection detection results include detection sensitivity adjustment values, specificity optimization coefficients, and final dissection region mapping data.

[0014] Preferably, the image data acquisition module includes:

[0015] The image acquisition submodule acquires the output data of the brain medical imaging device, records the original pixel values ​​of each image frame, and uses an image normalization algorithm to calculate the image resolution normalization value to obtain image normalized data.

[0016] The data calibration submodule obtains device calibration parameters based on the image standardization data, adjusts the image grayscale value according to the calibration parameters, and generates an image grayscale calibration sequence.

[0017] The timestamp index submodule assigns timestamp indices based on the acquisition time points, associates the timestamp indices with image frames, and obtains the original image dataset.

[0018] Preferably, the image preprocessing module includes:

[0019] The image enhancement submodule, based on the original image dataset, applies contrast stretching and histogram equalization techniques to calculate the image enhancement intensity value and obtain the enhanced image data.

[0020] The noise filtering submodule calculates the noise filtering coefficients based on the enhanced image data using a filtering algorithm, removes image noise, and generates filtered image data.

[0021] The quality assessment submodule calculates the image quality assessment index based on the filtered image data, determines whether the image quality meets the processing requirements, and obtains preprocessed image data.

[0022] Preferably, the feature extraction module includes:

[0023] The vascular morphology analysis submodule extracts the vascular edge and branch structure based on the preprocessed image data, calculates the vascular wall thickness measurement value, and obtains vascular morphology data.

[0024] The blood flow feature calculation submodule analyzes the pixel intensity change rate based on the blood vessel morphology data, estimates the blood flow velocity, and generates hemodynamic data.

[0025] The feature fusion submodule integrates the vascular morphology data and hemodynamic data, locates vascular branch points, and generates vascular feature extraction results.

[0026] Preferably, the interlayer detection module includes:

[0027] The pattern matching submodule, based on the extracted vascular features, uses template matching technology to compare vascular features with the ablation pattern, calculates the ablation confidence score, and obtains preliminary matching results.

[0028] The region segmentation submodule, based on the preliminary matching results, applies a region growing algorithm to segment the suspected interlayer region and generate interlayer region contour data;

[0029] The tagging submodule tags and serializes the contour data of the interlayer region to obtain the initial interlayer detection result.

[0030] Preferably, the parameter analysis module includes:

[0031] The geometric measurement submodule, based on the initial interlayer detection results, measures the length and width of the interlayer region, calculates the interlayer length measurement value and the interlayer width measurement value, and obtains geometric attribute data;

[0032] The spatial analysis submodule calculates the azimuth angle of the interstitial space relative to the blood vessel based on the geometric attribute data, and generates the azimuth angle data of the interstitial space.

[0033] The parameter integration submodule merges the geometric attribute data and the interlayer azimuth data to generate interlayer parameter analysis results.

[0034] Preferably, the detection optimization module includes:

[0035] The threshold adjustment submodule dynamically adjusts the detection threshold based on the interlayer parameter analysis results and interlayer characteristic data, calculates the detection sensitivity adjustment value, and obtains the threshold optimization data.

[0036] The parameter optimization submodule adjusts the algorithm parameters to optimize specificity based on the threshold optimization data, calculates the specificity optimization coefficient, and generates the parameter optimization result.

[0037] The result generation submodule, based on the parameter optimization results, remaps the interlayer region to generate final interlayer region mapping data, thus obtaining optimized interlayer detection results.

[0038] Preferably, the system further includes:

[0039] The three-dimensional reconstruction module, based on the optimized interstitial detection results, reconstructs a three-dimensional vascular model and generates three-dimensional vascular model data.

[0040] The dynamic prediction module analyzes the evolution trend of dissection based on the three-dimensional vascular model data and generates a prediction result of dissection development.

[0041] Preferably, the system further includes:

[0042] The report generation module, based on the mezzanine development prediction results and the optimized mezzanine detection results, generates structured detection report data.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] In the image data acquisition and processing stage, the system's image data acquisition module standardizes and calibrates the acquired brain medical image data to generate a raw image dataset. This processing method effectively reduces inconsistencies in image data caused by differences in equipment and scanning conditions, ensuring that the image data processed subsequently has a unified standard and a reliable quality foundation. The image preprocessing module further performs image enhancement and noise filtering on the raw image dataset. Image enhancement technology highlights the feature information of vascular regions, making vascular structures clearer and more discernible. Simultaneously, noise filtering removes various types of interference noise from the image, reducing the impact of artifacts on image quality. This provides high-quality preprocessed image data for subsequent feature extraction and layer detection, helping to improve the accuracy of subsequent detection steps.

[0045] In terms of feature extraction, the feature extraction module calculates vascular morphology and blood flow features based on the preprocessed image data and generates vascular feature extraction results. This module can accurately capture key feature information related to cerebral artery dissection from the image, including not only vascular morphological changes but also blood flow features. Compared with the traditional method of judging by visual observation of images, this feature extraction method based on quantitative calculation is more objective and comprehensive, and can uncover subtle feature changes that are difficult for humans to detect, providing strong support for the accurate identification of dissection.

[0046] The dissection detection module uses pattern recognition technology to analyze the extracted vascular features, identify the cerebral artery dissection region, and obtain initial dissection detection results. Pattern recognition technology can establish an effective dissection recognition model through learning and analysis of a large amount of feature data, automatically classifying and judging vascular features, avoiding the interference of subjective factors in manual detection. At the same time, this technology has strong information processing capabilities, can quickly process massive amounts of image feature data, significantly shorten the detection time, improve detection efficiency, and achieve rapid preliminary screening of cerebral artery dissection.

[0047] The parameter analysis module quantifies and analyzes the geometric properties and spatial distribution parameters of the dissection based on the initial dissection detection results, and generates corresponding results. This module can transform the dissection features that were originally difficult to describe precisely into specific quantitative parameters, presenting the actual condition of the dissection in detail. These quantitative parameters can clearly show doctors the size, shape, and specific distribution of the dissection in the blood vessels, helping doctors to understand the condition more deeply and providing detailed and accurate information for assessing the severity of the condition, selecting treatment plans, and monitoring treatment effects. This solves the problem that traditional methods cannot accurately quantify dissection parameters.

[0048] Based on the analysis results of interstitial parameters, the detection optimization module adjusts the detection threshold and algorithm parameters to generate optimized interstitial detection results. This dynamic optimization mechanism enables the detection system to self-adjust and improve according to the actual detection situation. When the system detects potential errors or omissions in the initial detection results, it can optimize the detection algorithm by adjusting parameters, thereby improving the accuracy and reliability of the detection. At the same time, this module also enables the system to better adapt to individual differences among different patients and the complexity of different conditions, enhancing the system's adaptability and flexibility, ensuring good detection performance in various detection scenarios, and providing more reliable detection services for clinical use. Attached Figure Description

[0049] Figure 1 This is a timing diagram of the image processing-based cerebral artery dissection detection system described in this invention;

[0050] Figure 2 A flowchart showing the association between the original image dataset and the various results;

[0051] Figure 3 This is a flowchart illustrating the working principle of the image preprocessing module. Detailed Implementation

[0052] 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.

[0053] Please see Figure 1 The present invention provides a brain artery dissection detection system based on image processing. The system includes: an environmental data acquisition module, an environmental data modeling module, a historical data analysis module, an environmental equalization and control module, an automatic adjustment module, and a remote control module.

[0054] The environmental data acquisition module deploys gridded environmental monitoring sensors based on the environmental conditions of the aquaculture shed to collect aquaculture environmental data; the environmental data modeling module acquires environmental parameter sequences and extracts feature indices, determines the initial position and frequency of key parameters, and maps them to generate feature permutation sequences to construct an original environmental sequence template; the historical data analysis module analyzes the impact of different environmental factors on each zone based on historical data and simulation experiments; the environmental equalization control module regulates environmental changes caused by environmental factors according to a regional equalization environmental regulation algorithm; the automatic adjustment module automatically adjusts the homogenized environmental parameters based on a feedback control algorithm to maintain environmental stability; and the remote control module sets up environmental control devices and establishes a remote interactive platform for recording, displaying system data, and modifying system parameters.

[0055] Example 1: See Figure 2 In constructing the original image dataset, the generation of image resolution standardization values ​​relies on the uniformization of the pixel spacing of the input images. This process uses a bicubic interpolation algorithm to adjust medical images from different sources to a nominal pixel spacing of 0.5 mm. The calculation is based on adaptive scaling according to the ratio between the device's original resolution and the target resolution. The generation of the image grayscale calibration sequence is achieved through a linear transformation model. This model uses the grayscale response curve pre-stored at the device's factory as a benchmark, combined with the brightness compensation coefficient of the current imaging environment, to perform frame-by-frame correction of the grayscale value of each pixel, ultimately forming an image sequence with consistent grayscale characteristics. The allocation of the acquisition timestamp index is accurate to the millisecond level. The system uses the clock signal of the synchronous imaging device to attach an absolute time stamp to each frame of image, and stores it together with the image data in a relational database for subsequent time-series analysis.

[0056] The image enhancement intensity value in the preprocessed image data is calculated using a contrast-limited adaptive histogram equalization algorithm. This algorithm avoids noise amplification by limiting the contrast gain of local regions, and its enhancement intensity value is dynamically adjusted according to the variance of the local gray-level distribution of the image. The noise filtering coefficients are determined using a wavelet transform-based threshold denoising method. The noise level is quantified through statistical analysis of wavelet coefficients at different scales, and adaptive filtering parameters are generated accordingly. These coefficients directly affect the size and standard deviation of the Gaussian filter kernel. The image quality assessment index is calculated by integrating multiple dimensions, including the signal-to-noise ratio of the entire image, edge sharpness, and gray-level uniformity. These indicators are weighted and fused to generate a normalized score between 0 and 1, used to objectively determine whether the image meets the minimum requirements for subsequent processing.

[0057] The vessel wall thickness measurements in the vessel feature extraction results are calculated using a morphological method based on distance transformation. First, anisotropic diffusion filtering is used to enhance the vessel edges. Then, the Canny operator is used to extract the inner and outer boundaries of the vessel. Finally, the thickness measurement sequence is obtained by calculating the Euclidean distance distribution between the boundaries. The derivation of the blood flow velocity estimate relies on temporal pixel intensity variation analysis. This system tracks the spatiotemporal changes in pixel intensity within a specific vessel region between consecutive frames, combines this with the frame rate parameters of the imaging device, and uses optical flow to calculate the vector estimate of the blood flow velocity. Vessel branch point localization data is obtained through a vessel skeletonization algorithm. This algorithm first generates a single-pixel-width vessel skeleton through morphological thinning operations, then detects the nodes and endpoints in the skeleton, and determines the 3D coordinates and connectivity of the branch points through topological analysis. The interstitial region contour data in the initial interstitial detection results is generated using a region growing algorithm. This algorithm uses seed points provided by pattern matching results as a basis, expands the region according to the similarity criteria between pixel intensity and texture features, and finally extracts the closed contour using a boundary tracking algorithm. The dissection confidence score is calculated based on a multi-feature fusion classification model. This model inputs outlier values ​​of vessel wall thickness, blood flow pattern features, and morphological parameters into a support vector machine classifier, outputting a probability score representing the likelihood of dissection. The generation of suspected region marker sequences employs a spatial clustering method. This system performs distance-based clustering analysis on all suspected regions, assigns a unique identifier to each cluster, and arranges them in descending order by region area and confidence score to form a detection sequence.

[0058] The length of the interstitial space in the parameter analysis results is obtained by calculating the actual physical length of the main axis of the interstitial region. This process first uses principal component analysis to determine the orientation of the interstitial region, then accumulates pixel distances along the main axis and multiplies them by the pixel spacing to convert them into physical dimensions. The width of the interstitial space is calculated using a cross-sectional sampling method perpendicular to the main axis. The system extracts the contour width in the vertical direction along the main axis at fixed intervals, and finally takes the average of all cross-sectional widths as the final measurement value. The azimuth angle data of the interstitial space is obtained by calculating the vector angle. The system uses the direction of the main blood vessel as the reference vector and the direction of the main axis of the interstitial space as the target vector. The relative azimuth angle of the interstitial space is determined by calculating the angle between the two vectors in three-dimensional space. The detection sensitivity adjustment value in the interstitial space detection results is optimized through a dynamic threshold adjustment mechanism. Based on the geometric features and spatial distribution characteristics in the interstitial space parameter analysis results, the system selects the corresponding sensitivity adjustment coefficient from a preset parameter table using a lookup table method. This coefficient is directly used to adjust the classification decision threshold. The generation of specificity optimization coefficients relies on confusion matrix analysis. By statistically analyzing the feature distribution of false positive cases in historical data, the feature weights and decision boundaries of the classifier are adjusted to calculate an optimization coefficient for reducing the false positive rate. Finally, the generation of the interlayer region mapping data employs a coordinate inverse transformation technique. This system transforms the optimized interlayer region coordinates from the feature space to the original image space and uses a spatial interpolation algorithm to ensure high-precision alignment between the region boundaries and the original image.

[0059] Example 2: See Figure 3 The image data acquisition module comprises three cooperating sub-modules. The image acquisition sub-module directly connects to the data output interface of the medical imaging equipment, receiving raw DICOM format image frame data in real time. This sub-module stores the pixel value matrix of each frame in a buffer memory and records the equipment parameter status at the time of acquisition. The image normalization algorithm uses bilinear interpolation to uniformly adjust input images of different resolutions to a standard size of 1024×1024 pixels. Simultaneously, it calculates the image resolution normalization value based on the physical pixel pitch parameters of the imaging equipment; this value is defined as the actual physical size represented by each pixel in the normalized image. The data calibration sub-module calls the equipment characteristic parameters pre-stored in the system calibration database. These parameters include the equipment's unique grayscale response curve and dark current compensation coefficient. The calibration process converts the raw pixel values ​​into standard grayscale values ​​through a lookup table mapping method, thereby generating an image grayscale calibration sequence with consistent grayscale characteristics. The timestamp indexing sub-module uses a high-precision clock source to mark the acquisition time of each image frame. The time information is accurate to the millisecond level and bound to the image's unique identifier. The final raw image dataset is stored in the database management system as a time-sorted multi-frame image array.

[0060] The image preprocessing module's execution flow begins with the image enhancement submodule. This submodule reads the grayscale calibration sequence from the original image dataset and first performs a global contrast stretching operation to expand the image's grayscale dynamic range. The stretching parameters are automatically calculated based on the distribution characteristics of the image histogram. Adaptive histogram equalization is then applied to local region enhancement. This technique divides the image into several sub-regions and independently calculates the histogram equalization mapping function for each region. The resulting image enhancement intensity value is recorded as the contrast gain coefficient for each region. The noise filtering submodule receives the enhanced image data and analyzes its noise characteristics. After performing multi-scale decomposition of the image using wavelet transform, the noise filtering coefficients are derived by calculating the statistical characteristics of the coefficients at each scale. These coefficients determine the kernel size and standard deviation parameters of the Gaussian filter. The filtering operation is performed jointly in the frequency and spatial domains. First, Fourier transform is used to suppress periodic noise, and then an adaptive Wiener filter is applied to process random noise. The resulting filtered image data significantly reduces noise interference while maintaining edge sharpness. The quality assessment submodule performs multi-index quantitative evaluation on the processed image. When calculating the image signal-to-noise ratio, it uses the ratio of the standard deviation of the background region to the average value of the signal region. Edge sharpness is calculated by extracting edges using the Sobel operator and then calculating the statistical characteristics of the gradient magnitude. Gray-level uniformity is calculated by dividing the image into sub-regions and calculating the inverse value of the gray-level variance. These indices are weighted and fused to generate an image quality assessment index between 0 and 1. The system sets 0.7 as the quality threshold; only images meeting this standard will proceed to subsequent processing steps.

[0061] The detailed implementation of the data calibration submodule includes two stages: device parameter loading and pixel value conversion. During system initialization, the system reads device calibration parameters from an encrypted configuration file. These parameters include the polynomial coefficients of the grayscale response curve and the dark current compensation matrix. The grayscale calibration process employs a pixel-by-pixel linear transformation model, multiplying the original pixel value by a calibration coefficient and adding a compensation value. The converted grayscale value is constrained to a standard range of 0-255. The generated image grayscale calibration sequence includes metadata recording the calibration parameter version information. The timestamp index submodule relies on a high-precision network time protocol. This submodule immediately obtains the precise system clock time upon completion of each image frame acquisition. The timestamp format adopts an international standard time format and is converted into a relative time series in milliseconds. The indexing process uses hash mapping to quickly associate the timestamp with the image storage address.

[0062] The contrast stretching algorithm in the image enhancement submodule employs a piecewise linear transformation function. This function determines the stretching interval based on the 5th and 95th quantiles of the input image histogram, mapping the original grayscale values ​​to the full dynamic range while preserving linear relationships to avoid introducing nonlinear distortion. Histogram equalization uses an adaptive method with contrast constraints. The image is divided into 8×8 sub-blocks, and the cumulative distribution function of each sub-block is calculated. A contrast constraint threshold of 3.0 is set to prevent noise amplification. Finally, bilinear interpolation smooths the grayscale transitions at the sub-block boundaries. The wavelet analysis process in the noise filtering submodule uses the Daubechies wavelet basis for three-level decomposition. The noise level is estimated by calculating the variance of the coefficients of each detail sub-band. Based on this estimate, an adaptive threshold is calculated for wavelet coefficient shrinkage. The calculation of the filtering coefficients also considers the differences in texture complexity across different regions of the image. The signal-to-noise ratio calculation of the quality assessment submodule adopts the background region method, selects the four corner regions of the image as noise reference areas and the central cardiovascular region as signal areas, and converts the logarithmic form of the ratio between the two into a decibel value. Edge sharpness assessment is achieved by measuring the edge width after Canny edge detection. Gray-level uniformity assessment is calculated by the inverse of the gray-level variance after region division. Finally, the quality index is fused into a single score value through a three-layer neural network model.

[0063] During system execution, each submodule employs a pipelined parallel processing approach. Image acquisition and standardization processing are performed synchronously in real time. Data calibration operations are batch-executed after a certain number of images have been cached. The timestamp index creation process and storage operations are completed asynchronously to improve processing efficiency. All intermediate data is labeled with version identifiers and timestamps to ensure the traceability of the data processing process. A system status monitor monitors the resource usage of each submodule in real time and dynamically adjusts processing priorities to ensure system stability. Preprocessed image data is organized in a multi-dimensional array format, including multiple data layers such as raw pixel data, calibration parameters, time series, and quality indicators. This organizational structure facilitates rapid access to required information by subsequent modules.

[0064] Example 3: The vascular morphology analysis submodule receives output data from the image preprocessing module. This submodule first uses a multi-scale vascular enhancement filter to highlight cerebral vascular structures and identifies tubular structure regions by calculating the eigenvalues ​​of the Hessian matrix. Vascular edge detection is implemented using an improved Canny operator, which employs an adaptive double-threshold algorithm. The threshold parameters are dynamically determined by analyzing the statistical characteristics of the image gradient magnitude. Vascular wall thickness measurement is achieved by calculating the distance transformation map of the binarized vascular region. First, an Euclidean distance transformation is performed on the segmented vascular region. Then, the maximum distance point is searched in each vascular cross-sectional direction, finally obtaining the thickness measurement value expressed in physical units. This process considers the spatial resolution parameters of the image.

[0065] The blood flow feature calculation submodule analyzes the pixel intensity variation patterns between consecutive image frames. This module uses a block-matching-based motion estimation algorithm to calculate the blood flow velocity vector. Specifically, the blood vessel region is divided into multiple analysis blocks. Within each analysis block, the pixel intensity cross-correlation function between adjacent frames is calculated. The displacement vector of the analysis block is estimated by finding the peak position of the cross-correlation function. The estimated blood flow velocity is obtained by dividing the displacement vector by the inter-frame time interval, while also considering the spatial scaling factor of the image, converting the pixel displacement into an actual physical velocity value. This process uses the following calculation formula:

[0066]

[0067] in: This represents the estimated blood flow velocity. This represents the number of pixels the analysis block has shifted between adjacent frames. It is the spatial resolution parameter of the image (millimeters per pixel). This indicates the time interval (in seconds) for image frame acquisition.

[0068] The feature fusion submodule integrates vascular morphology and hemodynamic data using a feature-level fusion strategy. This module first normalizes the two types of feature data to make feature values ​​of different dimensions comparable. Vessel branch point localization is achieved by analyzing the topological structure of the vascular skeleton map. A depth-first search algorithm traverses the skeleton network, identifying connection points with a degree greater than 2 as candidate branch points, and then verifying the authenticity of these points through morphological features. During the fusion process, weight coefficients for each feature term are calculated, and the weight values ​​are dynamically adjusted based on the stability and discriminative power of the feature terms, ultimately generating vascular branch point localization data that includes spatial location and feature attributes.

[0069] The pattern matching submodule in the aortic dissection detection module employs a multi-feature joint matching method. This module pre-establishes a feature template library of typical aortic dissection cases, including various feature templates such as abnormal vessel wall thickness patterns, blood flow turbulence patterns, and morphological abnormality patterns. The matching process is achieved by calculating the similarity score between the input vessel features and each template, using a weighted Euclidean distance as the similarity metric. After nonlinear transformation, the distance value yields an aortic dissection confidence score within the range of 0-1. This submodule also sets up a multi-level verification mechanism. When the matching scores of multiple feature templates all exceed a threshold, the weighted average is taken as the final confidence score, with weights allocated according to the reliability index of the templates. The region segmentation submodule extracts the aortic region based on the confidence score map, using an improved region growing algorithm combined with boundary optimization techniques. First, high-confidence score points are used as seed points, and region growing is performed according to feature similarity criteria. During the growing process, multiple constraints such as vessel wall thickness continuity and blood flow feature consistency are considered. The regional boundary is refined through an active contour model, and the gradient vector flow field is used as the external energy function to drive the contour evolution. The final output of the interlayer region contour data contains a smooth boundary coordinate sequence and regional topological connectivity information.

[0070] The labeling submodule uniformly encodes and manages all detected suspected regions. Each region is assigned a unique identifier and its spatial coordinates, confidence score, and set of feature attributes are recorded. The label sequence is arranged in descending order of confidence score, and a spatial adjacency index between regions is established for quick access and processing by subsequent modules. This module also implements region merging and splitting functions. When multiple adjacent regions have similar features, they are merged into the same detection target, while regions with high feature heterogeneity are appropriately split to ensure the internal consistency of each detection region. A modular pipeline architecture is adopted, with a data caching mechanism between the feature extraction and interlayer detection stages to ensure the continuity of the processing and data integrity. The system also integrates quality monitoring functions to monitor the processing status and data quality of each submodule in real time, automatically triggering reprocessing or alarm mechanisms when anomalies are detected. All algorithm parameters and threshold settings are managed through configuration files, supporting online adjustment and optimization to adapt to different imaging equipment characteristics and clinical diagnostic needs. Complete intermediate results and process parameters are retained during data processing, providing sufficient information support for subsequent traceability analysis and algorithm optimization.

[0071] Example 4: The geometric measurement submodule receives initial detection result data from the dissection detection module. This data includes the contour coordinates of multiple suspected dissection regions and their corresponding confidence scores. Taking a specific case of posterior traffic artery dissection detection as an example, the system first performs contour optimization processing on each detection region, using a Bezier curve fitting algorithm to smooth the region boundaries and eliminate contour irregularities caused by image noise. For each optimized region contour, its minimum bounding rectangle is calculated and the direction of the main axis is determined. The dissection length measurement is obtained by accumulating the segmented distances along the main axis, while the pixel distance is converted to millimeters using the image's spatial calibration parameters. The dissection width measurement adopts a sampling method perpendicular to the main axis. The system extracts the contour width in the vertical direction along the main axis at 0.1 mm intervals, and after excluding the influence near the region endpoints, takes a weighted average as the final width value. The weights are determined based on the contour sharpness index at that location.

[0072] The spatial analysis submodule establishes a reference coordinate system based on 3D vascular reconstruction data. This module first determines the centerline path of the main vascular artery and establishes a Frenet frame along the centerline. The azimuth data of the dissection is obtained by calculating the minimum angle between the main axis of the dissection and the tangent vector of the vascular centerline. The angle calculation uses a dot product formula and is converted to an angle value within the range of 0-180 degrees. This module also calculates the distance between the dissection region and the nearest vascular bifurcation point, as well as the radial position information of the dissection relative to the vascular lumen. These parameters collectively describe the spatial positioning characteristics of the dissection in the vascular tree. The parameter integration submodule organizes the various measurement data into a structured parameter set, using a hierarchical data model to store the characteristic parameters of each dissection region. This module performs parameter validity verification, checking the logical consistency between measured values; for example, the dissection length should be greater than the width, and the radius of curvature should be within a reasonable range. When abnormal parameter combinations are detected, a remeasurement process is automatically triggered. The integrated data includes timestamps, region identifiers, and all measurement parameters, forming a complete record of dissection parameter analysis.

[0073] The threshold adjustment submodule of the detection optimization module dynamically adjusts the detection sensitivity based on parameter analysis results. This module maintains a multi-dimensional parameter-threshold mapping relationship and selects an appropriate detection threshold according to the geometric and spatial characteristics of the dissection. For example, for small-sized dissection regions located on the outer side of a tortuous blood vessel, the system uses a lower detection threshold to improve sensitivity, while for large-sized regions in straight blood vessel segments, a stricter threshold condition is used. The sensitivity adjustment value is calculated using a multi-feature weighted decision model, comprehensively considering the contribution of multiple parameters such as dissection length, width, aspect ratio, and azimuth. The parameter optimization submodule focuses on improving detection specificity by analyzing the characteristics of false positive cases in historical detection results to establish a false positive pattern feature library. When calculating the specificity optimization coefficient, this module compares the similarity between the current dissection region features and false positive patterns. When the similarity exceeds a set threshold, the classifier parameters are automatically adjusted to reduce false alarms. The optimization process adopts a progressive adjustment strategy, re-evaluating the detection results after each parameter adjustment until the preset specificity target is reached.

[0074] The results generation submodule performs spatial remapping of the final detection results. This module transforms the optimized detection region coordinates back to the original image space, using a bicubic interpolation algorithm to ensure accurate localization of the region boundaries. The remapping process fully considers all geometric transformation operations during image preprocessing, including inverse transformation calculations for scaling, rotation, and cropping. The final output mezzanine region mapping data includes optimized region contour coordinates, confidence scores, and all feature parameters.

[0075] Table 1: Results of Sandwich Parameter Analysis

[0076]

[0077] Note: Azimuth refers to the angle between the main axis of the dissection and the centerline of the main vascular artery; radius of curvature reflects the degree of curvature of the dissection region; distance from the bifurcation point represents the minimum distance between the dissection region and the nearest vascular bifurcation point; relative vascular position describes the location of the dissection in the circumferential direction of the vascular artery. During system execution, an iterative optimization strategy is employed, forming a closed-loop feedback mechanism between the parameter analysis module and the detection optimization module. Geometric measurements and spatial analysis are performed again after each parameter adjustment until the detection results reach a stable state. All intermediate parameters and final results are recorded in the system database, including the measurement values, adjusted parameters, and optimization results of each iteration, providing complete data support for subsequent analysis and algorithm improvement. The data processing process employs a parallel computing architecture, with analysis and optimization operations for multiple dissection regions performed simultaneously, significantly improving system processing efficiency.

[0078] Example 5: The system receives the final interstitial region mapping data from the detection and optimization module. This module first reconstructs the volume data of the original medical image sequence, using an improved MarchingCubes algorithm to generate a three-dimensional vascular surface model from the two-dimensional slice data. During the reconstruction process, a region of interest is established centered on the interstitial region. Adaptive mesh subdivision technology is used to apply a higher resolution mesh near the interstitial region and a sparser mesh further away to optimize computational resources. After surface mesh generation, a Laplacian smoothing algorithm is used to eliminate mesh irregularities while preserving the geometric features of the interstitial edges. The final output three-dimensional vascular model data includes vertex coordinates, facet connectivity, and region label information. The dynamic prediction module performs hemodynamic simulation based on the three-dimensional vascular model. This module first establishes a personalized blood flow calculation model, setting biomechanical parameters such as blood viscosity and vascular wall elasticity according to the patient's imaging data. The simulation calculation uses the finite element method to solve the Navier-Stokes equations, simulating blood flow in a vessel containing an interstitial region, and calculating the pressure distribution and shear stress on the interstitial wall surface caused by periodic pulsating blood flow. The analysis of interlayer evolution trends is achieved by tracking the development and changes of stress concentration areas. The system records the maximum wall stress and its spatial distribution changes at each time step, and predicts the possible expansion direction and development speed of the interlayer based on the stress change rate.

[0079] The report generation module integrates data results from all processing stages. This module uses a templated report structure, automatically populating basic patient information, imaging parameters, test results, and predictive analyses. The structured test report comprises three main parts: text description, numerical tables, and image displays. The text description uses natural language generation technology to convert quantitative parameters into clinically compliant diagnostic expressions. The numerical tables present key measurement data and predictive indicators. The image display includes a comparative view of the original images, 3D reconstructed images, and dynamic simulation images. Report output supports multiple formats including DICOMSR, PDF, and HTML, ensuring seamless integration with hospital information systems.

[0080] The system implementation employs a multi-layered data verification mechanism. During the 3D reconstruction phase, the topological correctness of the mesh model is verified, ensuring the absence of non-manifold edges or isolated vertices. In the dynamic prediction phase, the convergence and numerical stability of the simulation calculations are monitored, and the time step is automatically adjusted and recalculated when abnormal oscillations are detected. The report generation phase performs data consistency checks, ensuring complete correspondence between textual descriptions, numerical results, and image annotations. Detailed log information is recorded for all processing steps, including processing time, parameter settings, and intermediate results, providing comprehensive data support for subsequent traceability analysis. The specific processing flow of the 3D reconstruction module includes three main stages: data preprocessing, isosurface extraction, and mesh optimization. In the data preprocessing stage, isotropic resampling of the original image is performed to ensure consistent resolution in all three directions for 3D processing. In the isosurface extraction stage, a threshold-based region growing method is used to determine the vessel wall surface, and the moving cube algorithm is used to generate the initial mesh. In the mesh optimization stage, edge folding and vertex shifting operations reduce the number of meshes while maintaining important geometric features, with particular emphasis on preserving detailed information in the interstitial region. Before the final model output, a mesh quality check is performed to ensure that all faces are manifold structures and that the normal directions are consistent.

[0081] The dynamic prediction module employs a multi-scale modeling approach, simulating overall hemodynamic parameters at the macroscopic scale and focusing on analyzing the flow field characteristics near the dissection at the microscopic scale. The simulation calculation uses periodic boundary conditions to simulate the impact of cardiac pulsation, solving for each cardiac cycle in 100 time steps, continuing for multiple cardiac cycles until a stable periodic solution is obtained. Dissection evolution prediction is based on a mechanical damage model, calculating the cumulative fatigue effect of the vessel wall under periodic stress. The prediction model considers the biomechanical processes of collagen fiber degradation and vessel wall remodeling. The report generation module adopts a component-based design, including data extraction, text generation, and format conversion components. The data extraction component retrieves all relevant results from the database; the text generation component automatically generates diagnostic descriptions based on predefined medical report templates; and the format conversion component ensures the output report conforms to the DICOM structured report standard and hospital document management requirements. The report content undergoes a multi-level review process, including automatic logical checks and human-computer interaction verification, and is finally published to the medical information system after being electronically signed by an authorized physician.

[0082] The system operates on a high-performance computing cluster, with 3D reconstruction and dynamic prediction tasks distributed across multiple computing nodes for parallel execution. A load balancing strategy optimizes resource utilization. The database system uses a time-series database to store dynamic prediction results, a relational database to store structured measurement data, and a document database to store report documents and image data. All data transmission employs encryption protocols, and patient data is anonymized to ensure the security and privacy of medical data. The system provides a complete application programming interface (API) to support data exchange and functional integration with existing hospital PACS, HIS, and RIS systems.

[0083] 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.

[0084] 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 cerebral artery dissection detection system based on image processing, characterized in that, The system includes: The image data acquisition module acquires brain medical image data, performs standardization and calibration on the image data, and generates a raw image dataset. The image preprocessing module performs image enhancement and noise filtering operations based on the original image dataset to obtain preprocessed image data; The feature extraction module calculates vascular morphology features and blood flow features based on the preprocessed image data, and generates vascular feature extraction results. The dissection detection module, based on the extracted vascular features, identifies the cerebral artery dissection region using pattern recognition technology to obtain initial dissection detection results; The parameter analysis module quantifies the geometric properties and spatial distribution parameters of the interlayer based on the initial interlayer detection results, and generates interlayer parameter analysis results. The detection optimization module adjusts the detection threshold and algorithm parameters based on the analysis results of the interlayer parameters to generate optimized interlayer detection results; The original image dataset includes image resolution normalization values, image grayscale calibration sequences, and acquisition timestamp indexes. The preprocessed image data includes image enhancement intensity values, noise filtering coefficients, and image quality assessment indices. The vascular feature extraction results include vascular wall thickness measurements, blood flow velocity estimates, and vascular branch point location data. The initial dissection detection results include dissection region contour data, dissection confidence scores, and suspected region marker sequences. The dissection parameter analysis results include dissection length measurements, dissection width measurements, and dissection azimuth data. The optimized dissection detection results include detection sensitivity adjustment values, specificity optimization coefficients, and final dissection region mapping data. The interlayer detection module includes: The pattern matching submodule, based on the extracted vascular features, uses template matching technology to compare vascular features with the ablation pattern, calculates the ablation confidence score, and obtains preliminary matching results. The region segmentation submodule, based on the preliminary matching results, applies a region growing algorithm to segment the suspected interlayer region and generate interlayer region contour data; The marking submodule marks and serializes the contour data of the interlayer region to obtain the initial interlayer detection result; The parameter analysis module includes: The geometric measurement submodule, based on the initial interlayer detection results, measures the length and width of the interlayer region, calculates the interlayer length measurement value and the interlayer width measurement value, and obtains geometric attribute data; The spatial analysis submodule calculates the azimuth angle of the interstitial space relative to the blood vessel based on the geometric attribute data, and generates the azimuth angle data of the interstitial space. The parameter integration submodule merges the geometric attribute data and the interlayer azimuth data to generate interlayer parameter analysis results.

2. The image processing-based cerebral artery dissection detection system according to claim 1, characterized in that, The image data acquisition module includes: The image acquisition submodule acquires the output data of the brain medical imaging device, records the original pixel values ​​of each image frame, and uses an image normalization algorithm to calculate the image resolution normalization value to obtain image normalized data. The data calibration submodule obtains device calibration parameters based on the image standardization data, adjusts the image grayscale value according to the calibration parameters, and generates an image grayscale calibration sequence. The timestamp index submodule assigns timestamp indices based on the acquisition time points, associates the timestamp indices with image frames, and obtains the original image dataset.

3. The image processing-based cerebral artery dissection detection system according to claim 2, characterized in that, The image preprocessing module includes: The image enhancement submodule, based on the original image dataset, applies contrast stretching and histogram equalization techniques to calculate the image enhancement intensity value and obtain the enhanced image data. The noise filtering submodule calculates the noise filtering coefficients based on the enhanced image data using a filtering algorithm, removes image noise, and generates filtered image data. The quality assessment submodule calculates the image quality assessment index based on the filtered image data, determines whether the image quality meets the processing requirements, and obtains preprocessed image data.

4. The image processing-based cerebral artery dissection detection system according to claim 3, characterized in that, The feature extraction module includes: The vascular morphology analysis submodule extracts the vascular edge and branch structure based on the preprocessed image data, calculates the vascular wall thickness measurement value, and obtains vascular morphology data. The blood flow feature calculation submodule analyzes the pixel intensity change rate based on the blood vessel morphology data, estimates the blood flow velocity, and generates hemodynamic data. The feature fusion submodule integrates the vascular morphology data and hemodynamic data, locates vascular branch points, and generates vascular feature extraction results.

5. The image processing-based cerebral artery dissection detection system according to claim 4, characterized in that, The detection optimization module includes: The threshold adjustment submodule dynamically adjusts the detection threshold based on the interlayer parameter analysis results and interlayer characteristic data, calculates the detection sensitivity adjustment value, and obtains the threshold optimization data. The parameter optimization submodule adjusts the algorithm parameters to optimize specificity based on the threshold optimization data, calculates the specificity optimization coefficient, and generates the parameter optimization result. The result generation submodule, based on the parameter optimization results, remaps the interlayer region to generate final interlayer region mapping data, thus obtaining optimized interlayer detection results.

6. The image processing-based cerebral artery dissection detection system according to claim 5, characterized in that, The system also includes: The three-dimensional reconstruction module, based on the optimized interstitial detection results, reconstructs a three-dimensional vascular model and generates three-dimensional vascular model data. The dynamic prediction module analyzes the evolution trend of dissection based on the three-dimensional vascular model data and generates a prediction result of dissection development.

7. The image processing-based cerebral artery dissection detection system according to claim 6, characterized in that, The system also includes: The report generation module, based on the mezzanine development prediction results and the optimized mezzanine detection results, generates structured detection report data.

Images