DSA image-based blood vessel stenosis mode identification and automatic quantitative analysis method

Abstract

The invention relates to the technical field of image recognition and analysis, in particular to a DSA image-based blood vessel stenosis mode recognition and automatic quantitative analysis method, which comprises the following steps: S1, acquiring a DSA image sequence, positioning a blood vessel region, generating a blood vessel mask, dynamically adjusting an image enhancement parameter, performing time consistency calibration, and performing cross calibration on a calibrated image; generating a standardized DSA image sequence; s2, constructing a blood vessel topological structure chart, identifying blood vessel bifurcation nodes, analyzing a change curve of the blood vessel diameter along a center line, and identifying narrow candidate points at the bifurcation; performing mode classification, and outputting classified narrow position information; s3, calculating contrast agent passing time, and estimating blood flow velocity; measuring the diameter of the blood vessel and calculating fractional flow reserve; and calculating a comprehensive score of stenosis severity, and generating a quantitative analysis report of vascular stenosis. According to the method, a stenosis severity scoring system reflecting two dimensions of functions and anatomy is constructed and can be directly used for generating a structured analysis report.

Description

Technical Field

[0001] This invention relates to the field of image recognition and analysis technology, and in particular to a method for pattern recognition and automated quantitative analysis of vascular stenosis based on DSA images. Background Technology

[0002] In interventional diagnosis and treatment, digital subtraction angiography (DSA) is an important imaging tool for clinical assessment of vascular stenosis and guidance of intraoperative intervention, with advantages such as high temporal resolution and clear spatial structure. However, existing DSA image processing workflows often rely on human experience for image enhancement and stenosis identification, which is highly subjective and lacks a consistent image quality evaluation mechanism, making it difficult to effectively adapt to image changes in different patients, different anatomical regions, or different angiographic conditions. In addition, conventional image enhancement methods often ignore inter-frame consistency, resulting in discontinuous grayscale flickering or vascular edge drift in the image sequence, which in turn affects the stability of subsequent vascular structure extraction and quantitative analysis.

[0003] On the other hand, current quantitative analysis of vascular stenosis is mostly based on static measurement of vascular diameter, without fully considering hemodynamic changes. Especially in complex structural areas such as vascular bifurcation and tortuosity, it is easy to miss or misjudge the degree of stenosis. At the same time, due to the lack of accurate estimation of blood flow velocity and pressure drop, existing methods are difficult to truly reflect the impact of stenosis on blood supply function, which limits their application value in functional evaluation and intervention decision-making. Summary of the Invention

[0004] This invention provides a method for pattern recognition and automatic quantitative analysis of vascular stenosis based on DSA images. It can realize DSA sequence standardization processing, automatic identification of vascular structures, and quantitative analysis of blood flow reserve by fusing anatomical and functional parameters, so as to improve the accuracy of stenosis identification and clinical interpretability.

[0005] A method for pattern recognition and automated quantitative analysis of vascular stenosis based on DSA images includes the following steps: S1: Acquire DSA image sequence, locate blood vessel regions in each frame of image, and generate a blood vessel mask; calculate the blood vessel imaging quality index based on the blood vessel mask, and dynamically adjust the image enhancement parameters according to the blood vessel imaging quality index; perform temporal consistency calibration on the enhanced image sequence to eliminate inter-frame grayscale fluctuations; perform cross-calibration on the calibrated images based on the standard angiography parameter library to generate a standardized DSA image sequence. S2: Extract the vessel centerline from the standardized DSA image sequence, and construct a vessel topology map based on the vessel centerline; identify vessel bifurcation nodes in the vessel topology map, and delineate a region of interest for bifurcation with each bifurcation node as the center; within each region of interest for bifurcation, analyze the change curve of vessel diameter along the centerline, and identify candidate stenosis points at the bifurcation; classify the candidate stenosis points at the bifurcation into patterns, distinguish between bifurcation-type stenosis and non-bifurcation-type stenosis, and output the classified stenosis location information; S3: Calculate the contrast agent transit time based on the standardized DSA image sequence, and estimate the blood flow velocity based on the contrast agent transit time; combine the classified stenosis location information, measure the vessel diameter at the identified stenosis location, and calculate the fractional flow reserve based on the blood flow velocity and vessel diameter measurement results; calculate the comprehensive stenosis severity score based on the fractional flow reserve and vessel diameter stenosis rate; generate a quantitative analysis report of vascular stenosis based on the comprehensive stenosis severity score.

[0006] Optionally, S1 specifically includes: S11: Obtain the DSA angiography sequence, and use a deep learning segmentation model based on the U-Net architecture to segment the blood vessel region of each frame of the sequence to obtain a binarized blood vessel mask image. S12: Based on the vascular mask image, calculate the peak signal-to-noise ratio of the image within the coverage area and the average gradient amplitude of the vascular edge to construct a vascular imaging quality index; dynamically select and apply the corresponding combination of image enhancement algorithm parameters, which includes window width, window level and contrast stretching coefficient. S13: For the enhanced image sequence, using the frame with the most sufficient contrast agent filling as the reference benchmark, inter-frame registration and gray-level histogram matching techniques based on optical flow are used to perform temporal consistency calibration. S14: Call the pre-built standard contrast parameter library, cross-compare and non-linearly map the calibrated images with the corresponding standards in the parameter library, and generate a DSA image sequence with grayscale and contrast normalization.

[0007] Optionally, the optical flow method includes: The frame image with the most complete contrast agent filling is used as the reference frame, and the enhanced current frame image and the reference frame image are used as input; Based on the constraints of brightness consistency and spatial smoothness, the displacement direction and displacement amplitude of each pixel in the current frame relative to the reference frame are calculated to generate a pixel-level displacement vector field. The current frame image is reverse-mapped and resampled according to the displacement vector field, so that the blood vessel structure in the current frame image is aligned with the reference frame in spatial position, thereby completing the inter-frame spatial registration.

[0008] Optionally, S14 specifically includes: S141: Call the pre-built standard angiography parameter library, which includes ideal blood vessel grayscale distribution functions and contrast curves constructed under different standard projection angles and anatomical locations, and retrieve the corresponding standard parameter items from the standard angiography parameter library; S142: Cross-compare the grayscale and contrast features of the calibrated image with the retrieved standard parameter items. Based on the comparison results, use a nonlinear mapping strategy to redistribute the grayscale and enhance the local contrast of the image to generate a DSA image sequence with normalized grayscale and contrast.

[0009] Optionally, S2 specifically includes: S21: For the frame with the clearest blood vessel visualization in the standardized DSA image sequence, apply a morphological thinning algorithm to iteratively process the blood vessel mask to obtain a single-pixel-wide blood vessel centerline, and record the spatial coordinates and branch connection relationships of each centerline pixel. S22: Based on the connection relationship of the central line pixels, construct a blood vessel topology diagram with pixels as vertices and connection relationships as edges; S23: Taking each candidate bifurcation node as the center, extend a preset distance distally along the center lines of each connected blood vessel branch to define the enclosed spatial region as the bifurcation region of interest. S24: In each bifurcation region of interest, calculate the local diameter of the blood vessel perpendicular to the centerline along the centerline of each branch, and generate a curve showing the change in blood vessel diameter along the centerline; identify local minimum points in the curve where the diameter value is lower than the preset stenosis identification ratio threshold of the proximal reference diameter, and mark them as stenosis candidate points at the bifurcation. S25: For each narrow candidate point, extract the geometric features of its location relative to the bifurcation node, based on the geometric features.

[0010] Optionally, the geometric features include the candidate point being located on the trunk before the fork, the trunk after the fork, or a branch, and the distance between the candidate point and the fork node.

[0011] Optionally, S25 further includes using a pre-trained classifier to classify narrow candidate points into bifurcated narrow points or non-bifurcated narrow points, and outputting the spatial location information of all classified narrow points.

[0012] Optionally, S3 specifically includes: S31: In the standardized DSA image sequence, a first and a second region of interest are defined at the beginning and end of the target blood vessel, respectively; the curve of the average pixel intensity changing with time in each region is extracted to obtain a time-density curve; the peak time difference between the time-density curve of the second region of interest and the time-density curve of the first region of interest is calculated as the contrast agent transit time. S32: Estimate the average blood flow velocity based on the known physical length of the target blood vessel or the three-dimensional length of the blood vessel segment obtained through three-dimensional calibration technology, and the contrast agent transit time; S33: For each of the aforementioned narrow locations, perform grayscale profile analysis on the blood vessel centerline normal upward within a preset range from near to far at the narrow location, determine the local blood vessel diameter by fitting the blood vessel edge points, and simultaneously measure the normal blood vessel diameter in the proximal reference area at the narrow location. S34: Input the average blood flow velocity, the local vessel diameter at the stenosis location, and the diameter of the normal vessel at the proximal end into a pre-established simplified computational fluid dynamics model to simulate and calculate the pressure difference between the distal and proximal ends of the stenosis, and calculate the fractional blood flow reserve at the stenosis location. S35: The blood flow reserve fraction is linearly weighted and fused with the vascular diameter stenosis rate calculated based on the local vessel diameter and the proximal normal vessel diameter according to a preset weighting coefficient to calculate a comprehensive score of stenosis severity. S36: Generate a structured report based on the comprehensive score of all narrow locations.

[0013] Optionally, the simplified computational fluid dynamics model is based on the functional relationship between the local diameter of the stenotic segment, the normal diameter of the proximal end, and the average blood flow velocity. It uses a nonlinear formula to estimate the pressure difference before and after the stenosis and calculates the fractional flow reserve through linear transformation. The simplified computational fluid dynamics model comprehensively considers the coupled influence of blood flow velocity and changes in vessel radius on pressure drop, thereby achieving an approximate estimation of the fractional flow reserve.

[0014] Optionally, the structured report includes a schematic diagram of the stenosis location, a stenosis pattern classification, a vessel diameter stenosis rate, a fractional flow reserve, a comprehensive score, and clinical significance hints based on the score interval.

[0015] The beneficial effects of this invention are: This invention constructs a vascular imaging quality index and dynamically adjusts enhancement parameters such as window width, window level, and contrast stretching to enable the image enhancement process to adaptively address grayscale fluctuations between different individuals and time frames. Combined with optical flow-based temporal consistency calibration and grayscale histogram matching technology, it improves the continuity and structural fidelity of DSA sequences in the temporal dimension. Furthermore, it incorporates a standard angiography parameter library for nonlinear grayscale mapping, unifying image grayscale and contrast distributions, ultimately outputting standardized DSA sequences with strong anatomical site comparability and high boundary clarity, providing stable input for subsequent stenosis detection and blood flow analysis.

[0016] This invention uses vascular centerline extraction and topological modeling to accurately locate stenosis candidate points in bifurcation regions. By combining their spatial relationships within the bifurcation structure, it distinguishes between bifurcation and non-bifurcation stenosis, improving the classification ability for stenosis identification in complex vascular regions. Furthermore, it uses time-density curves to estimate mean blood flow velocity and combines local diameter changes to construct a simplified CFD model to calculate the fractional flow reserve (FFR) of the stenotic segment. Finally, it integrates FFR and vascular stenosis rate with weighted coefficients to construct a stenosis severity scoring system reflecting both functional and anatomical dimensions, which can be directly used to generate structured analysis reports. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this 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 for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a standardized flowchart of DSA according to an embodiment of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0020] like Figures 1-2 As shown, the method for pattern recognition and automated quantitative analysis of vascular stenosis based on DSA images includes the following steps: S1: Acquire DSA image sequences, locate blood vessel regions in each frame, and generate a blood vessel mask; calculate the blood vessel imaging quality index based on the blood vessel mask, and dynamically adjust the image enhancement parameters according to the blood vessel imaging quality index; perform temporal consistency calibration on the enhanced image sequence to eliminate inter-frame grayscale fluctuations; perform cross-calibration on the calibrated images based on the standard angiography parameter library to generate a standardized DSA image sequence.

[0021] S1 specifically includes: S11, Vascular region segmentation and mask generation: Obtaining DSA angiography sequences For each frame of image Application of a deep segmentation model based on the U-Net architecture The corresponding binary blood vessel mask image is obtained. , represented as: ; in, For the first The original DSA image of the frame, This is the mask image for the corresponding blood vessel region, with values ​​of 0 or 1. This represents the total number of frames.

[0022] DSA image sequences are a medical imaging technique used for vascular imaging, and their acquisition process includes the following steps: Contrast agent injection: Before image acquisition, doctors inject iodine contrast agent into the target blood vessel area through a catheter to enhance the contrast of the blood vessel in the X-ray image.

[0023] Continuous X-ray acquisition: During the contrast agent injection process, the X-ray machine continuously captures several frames of images, forming a time-series image. These images capture the dynamic process of the contrast agent flowing in the blood vessels, typically at a frame rate of 7.5 to 30 frames per second, with the entire sequence consisting of dozens to hundreds of frames.

[0024] Foreground and background subtraction processing: Using DSA subtraction technology, the initial image without contrast agent is acquired, and then subsequent frames are processed frame by frame to eliminate background structures such as bones and tissues, making blood vessels clearer.

[0025] Image sequence export: The final set of DSA image sequences is usually in DICOM format, including metadata such as time, equipment parameters, and anatomical locations, and can be exported through a medical workstation for subsequent processing.

[0026] U-Net is a classic medical image segmentation network, particularly suitable for pixel-level segmentation tasks with small sample sizes, and it performs exceptionally well in blood vessel segmentation. It consists of an encoder and a decoder, with skip connections in between to preserve spatial details, as detailed below: Input module: Receives single-frame DSA grayscale images, uniformly sized at 512×512 pixels, and performs image normalization processing to distribute pixel values ​​within a certain range. If the original image size is not 512×512, then bilinear interpolation is used for uniform scaling.

[0027] The encoder section includes multiple convolutional and pooling layers; each layer extracts increasingly abstract features while reducing image resolution; it is used to capture global contextual information about blood vessels.

[0028] Decoder section: Includes four upsampling modules, each layer including: A 2×2 deconvolutional layer is used for upsampling; the encoder features of the corresponding layer are spliced ​​through skip connections to fuse shallow edge details; after splicing, they are then passed through two 3×3 convolutional layers and ReLU activation; the number of channels in each layer decreases in reverse to 512, 256, 128, and 64.

[0029] Output layer: A 1×1 convolutional kernel is used to compress the feature map of the last layer into a single-channel output; a binarized probability map is generated by the Sigmoid activation function; a 0.5 threshold is applied to the output map for binarization to obtain the blood vessel mask image.

[0030] Model training configuration: During training, the U-Net model employs a composite loss function to optimize blood vessel region segmentation performance. This loss function is composed of a weighted average of the Dice loss and the binary cross-entropy loss, where: The Dice loss is primarily used to improve the model's sensitivity to small blood vessel regions and pixel-level overlap, ensuring high segmentation accuracy when processing small, blurred-boundary blood vessel regions. The binary cross-entropy loss is used to improve overall classification stability and convergence efficiency, suppressing the risk of missegmentation of background regions. This weighted combination can improve the model's segmentation performance at blood vessel edges and bifurcation details, making it suitable for scenarios with complex blood vessel structures and significant contrast changes in DSA images.

[0031] In terms of optimizer configuration, the Adam optimizer is used to iteratively update the model parameters. The initial learning rate is set to 0.0001 and dynamically adjusted according to the performance of the validation set to improve the model training stability and generalization ability.

[0032] In the training data preparation phase, real DSA image sequences were used, and artificially labeled vascular masks were used as supervision labels. The training data covered typical anatomical sites such as coronary arteries and cerebral arteries, and model training and validation were performed using a five-fold cross-validation method to reduce the impact of sample bias.

[0033] To improve the model's robustness and adaptability to different shooting conditions, various image enhancement strategies were introduced during training, including image rotation, scaling, horizontal flipping, and Gamma perturbation, to simulate image deformation and brightness differences in actual clinical settings.

[0034] The model performance evaluation employs a combination of metrics, including Dice coefficient, crossover ratio, sensitivity, and specificity, to comprehensively measure the model's accuracy and stability in vessel recognition.

[0035] S12, Construction of vascular imaging quality index and adaptive adjustment of enhancement parameters: S121: Apply blood vessel masking to each frame of image Within the covered area, two indicators, peak signal-to-noise ratio and average gradient amplitude at the vessel edge, are calculated to construct a vessel imaging quality index, expressed as: ; ; ; in, Peak signal-to-noise ratio, This represents the average gradient amplitude at the edge of the blood vessel. The maximum grayscale value of an 8-bit image. The value is set to 255 to ensure consistent PSNR calculation. This represents the average grayscale value of the blood vessel mask area, ranging from 0 to 255. It is the average pixel value within the blood vessel mask area, reflecting the overall brightness and serving as a basis for measuring local contrast. This is the vascular mask image of frame n, a binary image of 0 / 1, used to calculate the area of ​​the vascular imaging region. This represents the number of pixels in the blood vessel region, calculated from the regions with a pixel count of 1 in the blood vessel mask image. The region representing the edge of blood vessels was obtained through mask edge extraction and used to calculate the edge sharpness index. For the nth frame image at pixel point The grayscale value at that location ranges from 0 to 255. The gradient at a pixel represents the magnitude of the grayscale change at that pixel, reflecting the sharpness of the edge. This is a vascular imaging quality index used to comprehensively evaluate the vascular imaging quality of a single frame and provide a basis for enhancement strategies. These are weighting coefficients. , ,control and Weight in the vascular imaging quality index.

[0036] S122: According to The interval Select the corresponding image enhancement parameter combination and the original image Apply enhancement functions , represented as: ; in, For the enhanced image result, This represents the development quality threshold range. This indicates the combination of enhancement parameters, namely window width, window level, and contrast stretch factor. The value ranges from 50 to 400, controlling the compression or expansion of the image's dynamic range to enhance detail contrast. The value ranges from 50 to 200, controlling the center position of the image's grayscale and determining which segment of grayscale is emphasized. The value ranges from 0.5 to 2.5, adjusting the overall brightness and contrast in the low to medium grayscale areas to improve the visibility of faint blood vessels. This is an image enhancement function used to perform targeted enhancement processing on low-quality frames. It specifically includes the following three sub-strategies, all of which are executed based on the selected combination of enhancement parameters for the current frame: Window width - window position adjustment: Adjusted according to the set window width. and window position Linear remapping is performed on the grayscale image to map the key grayscale range to... The method can improve the dynamic range of blood vessel grayscale, enhance structural contrast, and suppress background interference.

[0037] Contrast stretching: This involves applying an exponential mapping to the image, normalizing the original grayscale values, and then applying an exponential function to stretch the low and medium grayscale areas to highlight the outline of blood vessels. The larger the contrast stretching value, the more obvious the enhancement, but excessive values ​​can cause overexposure or loss of detail.

[0038] Multi-scale histogram equalization: Local histogram equalization is performed on image blocks, and then the blocks are stitched together to enhance details and maintain local structural consistency; it is suitable for areas with uneven contrast agent or low dose, and improves the visibility of dark and weak blood vessel segments.

[0039] The output image of the entire enhancement function is not only clearer in terms of brightness and contrast, but also maintains the coherence of the blood vessel edges and grayscale structure, providing higher quality input data for subsequent segmentation, topology reconstruction and stenosis analysis.

[0040] S13, Timing Consistency Calibration: Frames with the most adequate contrast agent filling As a reference frame, an optical flow-based method is used. Inter-frame registration and grayscale histogram matching The technique involves performing temporal consistency calibration on each frame of the enhanced image, as shown below: ; ; in, For images after time consistency calibration, For the first Frame-enhanced DSA image, As the reference frame image, This is the vascular mask image for frame n. This represents the area of ​​the blood vessel mask in frame n, measuring the degree of contrast agent filling in the frame, and serving as a benchmark for selection. This represents the optical flow registration function, which registers the current frame image to the coordinate system of the reference frame. This represents a histogram matching function that adjusts the grayscale distribution of the current image to be close to that of the reference frame. The frame that indicates the most complete filling shows the largest blood vessel area.

[0041] This step performs temporal consistency calibration on the enhanced DSA image sequence, eliminating differences in image displacement and grayscale instability caused by patient breathing, heartbeat, or minor equipment movements between frames, ensuring a unified visual reference benchmark throughout the sequence. First, the area of ​​the vascular region in each frame is calculated using a vascular mask map, and the frame with the most complete contrast agent filling is selected as the reference frame—that is, the frame with the clearest and most complete vascular structure. Each other frame is then aligned to the spatial position and grayscale distribution of this reference frame. This includes estimating the pixel-level displacement vector field between adjacent frames using an optical flow-based image registration technique to achieve spatial registration of each frame, and mapping the grayscale distribution of the registered image to the grayscale distribution of the reference frame using a histogram matching algorithm, ensuring consistent image brightness and contrast. Temporal consistency calibration, ensuring the spatial and grayscale continuity of the vascular structure in the image sequence, is a necessary prerequisite for subsequent processing.

[0042] Optical flow registration is an inter-frame registration method based on dense optical flow estimation. It aligns the spatial structure of the current frame image to the coordinate system of the reference frame image to eliminate minor structural misalignments caused by respiratory motion, equipment vibration, or physiological fluctuations. The specific execution process includes: Image input: Receive the current frame image and the reference frame image as input, requiring that both have undergone preliminary enhancement and size unification processing; Optical flow estimation: The Farneback optical flow algorithm based on a multi-scale pyramid is used to estimate the pixel-level displacement vector field of the current frame relative to the reference frame at multiple resolution levels of the image. The algorithm fits a quadratic polynomial brightness distribution in the local neighborhood at each pixel and calculates the displacement vector with the smallest brightness change, thereby obtaining a dense optical flow map; Edge protection strategy: Introduce gradient smoothing and orientation constraint terms at the edge of the blood vessel region to prevent optical flow estimation errors at strong boundaries from spreading to the background region; Pixel resampling: Based on the displacement vector in the optical flow field, reverse mapping interpolation is performed on the current frame image to move each pixel to the corresponding position in the target reference frame, thereby achieving image registration; Output: Output the current frame image that is spatially consistent with the reference frame, for use in subsequent grayscale consistency matching processing.

[0043] Histogram matching functions are used to unify the grayscale distribution between the current frame image and the reference frame image, eliminating differences in brightness and contrast caused by exposure time, contrast agent concentration, X-ray penetration angle, etc., so as to maintain visual consistency in the sequence of images. The steps include: Statistical grayscale distribution: Calculate the grayscale histograms of the current frame image and the reference frame image respectively, and calculate the corresponding cumulative grayscale distribution based on the cumulative pixel frequency; Constructing a mapping relationship: Based on the cumulative grayscale distribution of the reference frame, find the corresponding cumulative probability value at each grayscale level, and find the grayscale value that is closest to the probability value in the cumulative grayscale distribution of the current frame, thereby establishing a grayscale mapping table from the current frame to the reference frame. Grayscale remapping: For each pixel in the current frame image, find the corresponding grayscale value in the mapping table based on its original grayscale value, and complete the grayscale value replacement operation; Edge preservation processing: To prevent loss of local high-frequency details, a local histogram matching strategy or a guided filtering mechanism can be used during grayscale remapping to maintain the contrast gradient of the blood vessel edges. Output image: Generates the current frame image with the same grayscale distribution as the reference frame, improving brightness uniformity between sequences and optimizing visual coherence.

[0044] S14, Grayscale and Contrast Normalization: Calling the standard imaging parameter library Each of the standard items Including the ideal vascular grayscale distribution function at specific angles and anatomical locations. With contrast curve Calibrate the temporal consistency of each frame of the image. Find the corresponding standard item Nonlinear mapping is performed to generate standardized DSA images, represented as: ; in, For the final standardized image, For the first The standard imaging parameters are those that match the shooting angle and body part of the current frame. The ideal vascular grayscale distribution function is a monotonically increasing curve with grayscale values ​​ranging from 0 to 255. It describes the typical grayscale distribution contour of a vascular structure under a certain anatomical view and is used to guide grayscale mapping. Specifically, it includes the following: Gray-level hierarchical structure model: The vascular region is divided into three gray-level regions: main trunk, bifurcation, and distal end. The gray-level mean, gray-level variance, and gray-level gradient distribution range are established for each region. The gray-level value of the main trunk should be the highest and most uniform. The gray-level value of the bifurcation region is transitional and has a slight gradient decrease. The gray-level value of the distal vascular region is lower and the edges are blurred.

[0045] Region-based grayscale distribution template: Extract the grayscale histogram of each region in the standard image and construct a standardized grayscale probability density curve as a matching benchmark; the standard grayscale of the LAD backbone region is concentrated in 200-240; the branch regions are concentrated in 150-200; the background region should be below 80.

[0046] Adaptive adjustment mechanism: Based on external conditions such as image resolution and noise level, the standard function is locally adjusted through moving average to enhance its adaptability and generalization ability.

[0047] The contrast curve function describes the local or overall contrast distribution characteristics of the standard image, specifically including the following elements: Sigmoid edge enhancement template: Defines the ideal gray-level transition range that the blood vessel edge should present. It uses the Sigmoid function to describe the transition process of gray level from the background to the blood vessel wall, and has two key parameters: center gray value and steepness. Among them, the center point reflects the edge position, and the steepness controls the edge sharpness. The larger the steepness, the sharper the edge. Local contrast target distribution: In standard images, the contrast index of different anatomical regions is measured, and target distribution curves are generated; Contrast mapping strategy: During the normalization process, dynamically select based on the difference between the contrast curve of the current image and the target image. Localized contrast stretching: Enhances detail in low-contrast areas; Edge sharpening: Enhances edge response intensity; Overexposure compression: Contrast compression is applied to overexposed areas to prevent loss of vascular texture.

[0048] The standardization mapping function is completed based on the grayscale distribution and contrast curve, and consists of the following three processing steps: Region structure matching analysis: Based on the image capture angle and vascular anatomy location information, the current frame image is divided into anatomical partitions and aligned with the structure distribution template in the standard parameter items to ensure the consistency of region correspondence during the grayscale mapping process; Gray-level distribution fitting and nonlinear mapping: Extract the actual gray-level distribution histogram of the current frame image within the corresponding region and compare it with the standard gray-level distribution function. Based on the difference between the two, select an appropriate nonlinear transformation function to remap the gray-level dynamic range of the current image to make it closer to the ideal gray-level shape of the standard model; Contrast Detail Enhancement: Based on consistent grayscale structure, the Sigmoid edge stretching strategy defined in the standard contrast curve is further applied to edge regions or local tissues to improve the contrast level and texture clarity of key blood vessel segments, ensuring that the image achieves both structural standard and clear detail.

[0049] S2: Extract the vascular centerline from the standardized DSA image sequence and construct a vascular topology map based on the vascular centerline; identify vascular bifurcation nodes in the vascular topology map and delineate a region of interest for bifurcation with each bifurcation node as the center; analyze the change curve of vascular diameter along the centerline within each bifurcation region of interest to identify candidate stenosis points at the bifurcation; classify the candidate stenosis points at the bifurcation into patterns, distinguish between bifurcation-type stenosis and non-bifurcation-type stenosis, and output the classified stenosis location information.

[0050] S2 specifically includes: S21, Vessel Centerline Extraction and Pixel Topological Recording: The frame with the clearest vessel visualization in the standardized DSA image sequence is set as... The corresponding vascular mask is Morphological thinning algorithm right After iterative processing, a single-pixel-width image of the blood vessel centerline is obtained, represented as: Each point is denoted as ; For images Each centerline pixel records its spatial coordinates in the image, and pixel-level connectivity is established based on the adjacency relationships of the centerline pixels within their 8-neighborhood, represented as: ; in, To enhance and normalize the vascular mask of the selected frame, The extracted centerline image, For the pixels on the center line, Indicates and Adjacent centerline pixel sets, The Euclidean distance between pixels is defined as the distance between pixels. If this distance is no greater than one pixel, it is determined that there is a connection between the pixels. The morphological thinning function's specific processing steps include: (1) Input object: The input is a segmented binary mask image of blood vessels, where the pixel value of the blood vessel region is 1 and the pixel value of the background region is 0. At this time, the blood vessels usually have a width of several to tens of pixels.

[0051] (2) Iterative boundary stripping idea: adopt the layer-by-layer stripping method, starting from the boundary of the blood vessel mask, and try to delete some boundary pixels in each iteration, so that the blood vessel area gradually shrinks towards the center.

[0052] (3) Deletion Condition Constraints: In each iteration, not all boundary pixels are deleted, but only those pixels that meet the following conditions are deleted: This pixel is located at the outer edge of the blood vessel region; Deleting this pixel will not cause the blood vessel structure to break. Deleting this pixel will not damage the endpoints or bifurcation structure of the blood vessel.

[0053] These constraints ensure that the connectivity and topology of blood vessels remain unchanged.

[0054] (4) Multiple iterations until convergence: The stripping process will be repeated continuously. After multiple iterations, the blood vessel mask will gradually become thinner. The algorithm stops when further deletion of any pixel will destroy the connectivity or structure.

[0055] (5) Output result: The final result is a single-pixel wide blood vessel centerline. The center line is located at the geometric midline of the original vascular region, preserving the direction, branches, and connections of the blood vessels.

[0056] S22, Construction of blood vessel topology map and identification of bifurcation nodes: using the set of centerline pixels Using vertices as vertices and the connections between adjacent pixels as edges, construct an undirected graph, represented as: ; Define the connectivity of a node: ; According to the definition of connectivity: like Then it is marked as Endpoint; like Then mark the candidate fork node as ; in, These are pixels on the blood vessel centerline, representing the blood vessel centerline pixels retained in the thinning result; they are nodes in the topology graph. For pixels Adjacent pixels, Is with The set of adjacent pixels is determined based on the 8-neighborhood connection rule, which includes adjacent pixels belonging to the center line in the top, bottom, left, right, and four diagonal directions. Let be the set of vertices in the graph structure. Let the set of edges in the graph structure be . This is a topological diagram of the blood vessel centerline. For pixels The connectivity, ranging from 1 to 8, indicates how many neighboring pixels a pixel is connected to, reflecting its position type in the topology.

[0057] S23, Delineation of Region of Interest for Forking: For each candidate forking node... Extending the length to the distal end along the centerline direction of all its connected branches And form a region of interest by extending the region. , represented as: ; in, A bifurcation node represents a candidate bifurcation node identified in the vascular topology map. This node corresponds to a pixel with a connectivity of 3 or greater in the vascular centerline pixels, reflecting whether the vascular branch or merges at that location. Indicates from the branch node Depart, along the first The centerline direction of the connected blood vessel branches, at a distance of from the bifurcation node. The centerline pixel at the position. This is the length of the fork extension, ranging from 15 to 30 pixels. Indicated by The region of interest is the bifurcation centered on the region.

[0058] S24, Generation of vessel diameter variation curves and identification of stenosis candidate points: in each In the middle, along the trajectory of the center line of each branch, calculate the local blood vessel diameter in the vertical direction. This forms a diameter variation curve, represented as: ; Define the narrowing recognition threshold: if at a certain location Local minima exist satisfy Then this point Marked as a narrow candidate point; in, The distance from the fork node is The centerline pixel, The position parameters are along the centerline. The vertical diameter at that location represents the width of the blood vessel at each location along the centerline. This is a normal vector perpendicular to the centerline, used to determine the measurement direction orthogonal to the vessel centerline, thus achieving perpendicular measurement of the vessel diameter and avoiding diameter deviation caused by vessel curvature. To detect the maximum width of the blood vessel boundary in the normal direction, The reference diameter is located at a certain distance from the proximal end of the bifurcation. It represents the average diameter of the relatively healthy segment of the blood vessel within a certain distance from the bifurcation node and is used as a relative baseline for the local diameter. The threshold value for narrow recognition ratio is 0.6-0.8. This represents the location of a local minimum on the diameter curve. The location of the narrow candidate point corresponds to the coordinates of the curve's minimum point, and is used to mark the actual location of the potential narrow region in the image.

[0059] Within the bifurcation region of interest, to determine whether structural stenosis exists in the blood vessel, this step unfolds around the vessel's centerline, performing a continuous analysis of the vessel width. Specifically, for each pixel along the centerline trajectory, its local normal direction is determined, i.e., the direction perpendicular to the current centerline direction. Along this direction, the boundary positions on the left and right sides of the point are detected, thereby measuring the transverse diameter of the blood vessel at that point. By repeating the above operation throughout the entire region of interest, a sequence of blood vessel diameters unfolding along the centerline direction can be obtained, thus forming a continuous diameter variation curve. This curve accurately reflects the spatial expansion and contraction of the blood vessel and is the basic data for identifying stenosis regions. After obtaining the diameter variation curve, local minima within the curve are further identified and compared with the proximal reference diameter. If the diameter decrease at a certain location exceeds a preset stenosis identification ratio threshold, that point is determined to be a potential stenosis candidate.

[0060] S25, Geometric Feature Extraction and Narrow Point Classification: For each narrow candidate point Extracting geometric feature vectors Input to the pre-trained classifier Perform pattern classification and output the classified narrow location information, represented as: ; ; like : bifurcation narrow; like Non-bifusible narrow type; in, For the location of narrow candidate points, To describe the narrow point The structural relationship between the centerline segment and the bifurcation node. This is the distance from the point to the nearest fork node. The angle between the centerline of the branch where the narrow point is located and the direction of the main trunk. Narrow candidate points The geometric eigenvectors, combined with multiple eigenvalues, For a narrow classifier, the input feature vector Output narrow type labels. This is the output of the classifier.

[0061] S3: Calculate the contrast agent transit time based on standardized DSA image sequences and estimate blood flow velocity based on the contrast agent transit time; combine the classified stenosis location information to measure the vessel diameter at the identified stenosis location, and calculate the fractional flow reserve based on the blood flow velocity and vessel diameter measurement results; calculate the comprehensive score of stenosis severity based on the fractional flow reserve and vessel diameter stenosis rate; generate a quantitative analysis report of vascular stenosis based on the comprehensive score of stenosis severity.

[0062] S3 specifically includes: S31, Contrast agent transit time calculation: In a standardized DSA image sequence, let... The measurement area is at the origin of the blood vessel. For the measurement area at the vessel termination end, the curves of the change in average gray value over time in the two areas are extracted and represented as follows: ; ; The time difference between the peak values ​​of the two curves is taken as the contrast agent transit time. , represented as: ; in, , The number of pixels in the region. , The recommended area is 20-100 pixels. If the area is too small, it is easily affected by noise; if it is too large, it may include non-target structures. A medium-sized area can stably reflect the average grayscale trend. Used to capture the dynamic process of angiography downstream of blood vessels. For pixels At any moment The grayscale value ranges from 0 to 255. The time-density curve for the first ROI reflects the dynamic changes during the contrast agent filling process at the beginning. The time-density curve for the second ROI reflects the contrast agent filling process at the termination point. The peak time of the first ROI. This is the peak time of the second ROI.

[0063] S32, Mean Blood Flow Velocity Estimation: Assume the physical length of the target vessel segment is... then mean blood flow velocity Represented as: ; Where DSA is a 2D sequence, Based on the known anatomical length, if 3D calibration technology is used, then... This is the result of three-dimensional reconstruction; S33, Blood vessel diameter measurement: for each stenosis point Gray-scale profile sampling is performed along its normal direction. Extract the edge points of the blood vessel, calculate the local diameter, and represent it as follows: ; The normal diameter measured in the proximal reference region is represented as: ; in, The extreme point of the first derivative of the gray level of the profile is symmetrically located. This is the point where the derivative is at its minimum on the left edge. This is the point where the derivative reaches its maximum value on the right edge. For narrow candidate points, The local diameter represents the minimum vascular diameter at the lesion site. For the proximal reference diameter, This means averaging the diameters of multiple normal blood vessel segments within the region. This represents the set of centerline pixels within the near-end reference region.

[0064] S34, Calculation of fractional flow reserve: Estimation of the pressure difference between the proximal and distal ends of the stenosis based on a simplified computational fluid dynamics model. Calculate the FFR value, expressed as: ; in, To simplify the drag coefficient in the computational fluid dynamics model, a value ranging from 0.2 to 1.0 was selected, taking into account empirical conditions such as blood viscosity, blood vessel wall compliance, and approximate Poiseuille flow. This refers to the local blood vessel diameter at the location of the stenosis. This is the reference diameter of a normal blood vessel proximal to the stenosis. The theoretical blood pressure value at the distal end of the stenosis is estimated using a simplified computational fluid dynamics model. The theoretical blood pressure value for the proximal end of the stenosis is set to 100; In narrowed segments, blood flow accelerates locally, accompanied by a pressure drop and energy loss. Vascular stenosis causes a decrease in local pressure, expressed using pressure difference, defined as... The more severe the narrowing, The flow rate tends to decrease, and the FFR tends to 0; this is achieved by utilizing flow resistance in fluid mechanics. And consider pressure drop ,get This indicates that, while maintaining a constant flow velocity, the more severe the local narrowing, the greater the... The smaller the value, the greater the pressure drop; to adapt to different types of blood vessels and imaging systems, an empirical adjustment coefficient is added. This forms the final expression.

[0065] S35, Comprehensive Stenosis Severity Score: Based on a linear fusion of fractional flow reserve and stenosis rate Sr, the score is expressed as follows: ; ; in, For the narrowing rate, The weighting coefficient for fractional blood flow reserve. The narrowing rate weighting coefficient, , .

[0066] S36, Generate a structured analysis report: The final output includes the following structured content: Diagram of the narrow location; Narrow pattern classification; Blood vessel diameter stenosis rate ; Fractional Flow Reserve (FFR) Overall score ; Clinical significance of the score range, such as whether interventional treatment is recommended.

[0067] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0068] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for pattern recognition and automated quantitative analysis of vascular stenosis based on DSA images, characterized in that, Includes the following steps: S1: Acquire DSA image sequence, locate blood vessel regions in each frame of image, and generate a blood vessel mask; calculate the blood vessel imaging quality index based on the blood vessel mask, and dynamically adjust the image enhancement parameters according to the blood vessel imaging quality index; perform temporal consistency calibration on the enhanced image sequence to eliminate inter-frame grayscale fluctuations; perform cross-calibration on the calibrated images based on the standard angiography parameter library to generate a standardized DSA image sequence. S2: Extract the vessel centerline from the standardized DSA image sequence, and construct a vessel topology map based on the vessel centerline; identify vessel bifurcation nodes in the vessel topology map, and delineate a region of interest for bifurcation with each bifurcation node as the center; within each region of interest for bifurcation, analyze the change curve of vessel diameter along the centerline, and identify candidate stenosis points at the bifurcation; classify the candidate stenosis points at the bifurcation into patterns, distinguish between bifurcation-type stenosis and non-bifurcation-type stenosis, and output the classified stenosis location information; S3: Calculate the contrast agent transit time based on the standardized DSA image sequence, and estimate the blood flow velocity based on the contrast agent transit time; combine the classified stenosis location information, measure the vessel diameter at the identified stenosis location, and calculate the fractional flow reserve based on the blood flow velocity and vessel diameter measurement results; calculate the comprehensive stenosis severity score based on the fractional flow reserve and vessel diameter stenosis rate; generate a quantitative analysis report of vascular stenosis based on the comprehensive stenosis severity score.

2. The method for vascular stenosis pattern recognition and automatic quantitative analysis based on DSA images according to claim 1, characterized in that, S1 specifically includes: S11: Obtain the DSA angiography sequence, and use a deep learning segmentation model based on the U-Net architecture to segment the blood vessel region of each frame of the sequence to obtain a binarized blood vessel mask image. S12: Based on the vascular mask image, calculate the peak signal-to-noise ratio of the image within the coverage area and the average gradient amplitude of the vascular edge to construct a vascular imaging quality index; dynamically select and apply the corresponding combination of image enhancement algorithm parameters, which includes window width, window level and contrast stretching coefficient. S13: For the enhanced image sequence, using the frame with the most sufficient contrast agent filling as the reference benchmark, inter-frame registration and gray-level histogram matching techniques based on optical flow are used to perform temporal consistency calibration. S14: Call the pre-built standard contrast parameter library, cross-compare and non-linearly map the calibrated images with the corresponding standards in the parameter library, and generate a DSA image sequence with grayscale and contrast normalization.

3. The method for vascular stenosis pattern recognition and automatic quantitative analysis based on DSA images according to claim 2, characterized in that, The optical flow method includes: The frame image with the most complete contrast agent filling is used as the reference frame, and the enhanced current frame image and the reference frame image are used as input; Based on the constraints of brightness consistency and spatial smoothness, the displacement direction and displacement amplitude of each pixel in the current frame relative to the reference frame are calculated to generate a pixel-level displacement vector field. The current frame image is reverse-mapped and resampled according to the displacement vector field, so that the blood vessel structure in the current frame image is aligned with the reference frame in spatial position, thereby completing the inter-frame spatial registration.

4. The method for vascular stenosis pattern recognition and automatic quantitative analysis based on DSA images according to claim 2, characterized in that, S14 specifically includes: S141: Call the pre-built standard angiography parameter library, which includes ideal blood vessel grayscale distribution functions and contrast curves constructed under different standard projection angles and anatomical locations, and retrieve the corresponding standard parameter items from the standard angiography parameter library; S142: Cross-compare the grayscale and contrast features of the calibrated image with the retrieved standard parameter items. Based on the comparison results, use a nonlinear mapping strategy to redistribute the grayscale and enhance the local contrast of the image to generate a DSA image sequence with normalized grayscale and contrast.

5. The method for vascular stenosis pattern recognition and automatic quantitative analysis based on DSA images according to claim 1, characterized in that, S2 specifically includes: S21: For the frame with the clearest blood vessel visualization in the standardized DSA image sequence, apply a morphological thinning algorithm to iteratively process the blood vessel mask to obtain a single-pixel-wide blood vessel centerline, and record the spatial coordinates and branch connection relationships of each centerline pixel. S22: Based on the connection relationship of the central line pixels, construct a blood vessel topology diagram with pixels as vertices and connection relationships as edges; S23: Taking each candidate bifurcation node as the center, extend a preset distance distally along the center lines of each connected blood vessel branch to define the enclosed spatial region as the bifurcation region of interest. S24: In each bifurcation region of interest, calculate the local diameter of the blood vessel perpendicular to the centerline along the centerline of each branch, and generate a curve showing the change in blood vessel diameter along the centerline; identify local minimum points in the curve where the diameter value is lower than the preset stenosis identification ratio threshold of the proximal reference diameter, and mark them as stenosis candidate points at the bifurcation. S25: For each narrow candidate point, extract the geometric features of its location relative to the bifurcation node, based on the geometric features.

6. The method for vascular stenosis pattern recognition and automatic quantitative analysis based on DSA images according to claim 5, characterized in that, The geometric features include the candidate point being located on the trunk before the fork, the trunk or branch after the fork, and the distance between the candidate point and the fork node.

7. The method for vascular stenosis pattern recognition and automatic quantitative analysis based on DSA images according to claim 5, characterized in that, S25 further includes using a pre-trained classifier to classify narrow candidate points into bifurcated narrow points or non-bifurcated narrow points, and outputting the spatial location information of all classified narrow points.

8. The method for vascular stenosis pattern recognition and automatic quantitative analysis based on DSA images according to claim 1, characterized in that, S3 specifically includes: S31: In the standardized DSA image sequence, a first and a second region of interest are defined at the beginning and end of the target blood vessel, respectively; the curve of the average pixel intensity changing with time in each region is extracted to obtain a time-density curve; the peak time difference between the time-density curve of the second region of interest and the time-density curve of the first region of interest is calculated as the contrast agent transit time. S32: Estimate the average blood flow velocity based on the known physical length of the target blood vessel or the three-dimensional length of the blood vessel segment obtained through three-dimensional calibration technology, and the contrast agent transit time; S33: For each of the aforementioned narrow locations, perform grayscale profile analysis on the blood vessel centerline normal upward within a preset range from near to far at the narrow location, determine the local blood vessel diameter by fitting the blood vessel edge points, and simultaneously measure the normal blood vessel diameter in the proximal reference area at the narrow location. S34: Input the average blood flow velocity, the local vessel diameter at the stenosis location, and the diameter of the normal vessel at the proximal end into a pre-established simplified computational fluid dynamics model to simulate and calculate the pressure difference between the distal and proximal ends of the stenosis, and calculate the fractional blood flow reserve at the stenosis location. S35: The blood flow reserve fraction is linearly weighted and fused with the vascular diameter stenosis rate calculated based on the local vessel diameter and the proximal normal vessel diameter according to a preset weighting coefficient to calculate a comprehensive score of stenosis severity. S36: Generate a structured report based on the comprehensive score of all narrow locations.

9. The method for pattern recognition and automatic quantitative analysis of vascular stenosis based on DSA images according to claim 8, characterized in that, The simplified computational fluid dynamics model is based on the functional relationship between the local diameter of the stenotic segment, the normal diameter of the proximal end, and the average blood flow velocity. It uses a nonlinear formula to estimate the pressure difference before and after the stenosis and calculates the fractional flow reserve through linear transformation. The simplified computational fluid dynamics model comprehensively considers the coupled influence of blood flow velocity and changes in vessel radius on pressure drop, thereby achieving an approximate estimate of the fractional flow reserve.

10. The method for pattern recognition and automatic quantitative analysis of vascular stenosis based on DSA images according to claim 8, characterized in that, The structured report includes a schematic diagram of the stenosis location, a classification of stenosis patterns, a stenosis rate of vessel diameter, a fractional flow reserve, a comprehensive score, and clinical significance notes based on the score interval.

Images