Angiography image enhancement method and system for radioscopy

Abstract

The invention discloses an angiography image enhancement method and system for radioscopy. The method comprises the steps of angiography image collection, blood vessel flow state coding, blood bone dynamic decoupling, blood vessel edge reconstruction, image enhancement and the like. The invention relates to the technical field of angiography image enhancement, in particular to an angiography image enhancement method and system for radioscopy, and provides a multiphase time sequence guided coding mechanism in the aspect of blood vessel flow state coding. A blood vessel skeleton structure dynamic decoupling method combined with confrontation feature separation is adopted to perform blood-bone dynamic decoupling, and dynamic perfusion components and static high-density skeleton background signals are effectively isolated; skeleton background interference is decoupled through dynamic information, the blood vessel edge is further reconstructed, and the image contrast and the structural integrity are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of angiographic image enhancement, and in particular to an angiographic image enhancement method and system for radiological examination. Background Art

[0002] Angiographic image enhancement methods used in radiological examinations are image processing techniques that sharpen, enhance contrast, and suppress noise in vascular images acquired during radiological examinations. This approach aims to more accurately display vascular structure and lesion details. This method significantly improves the visibility of tiny lesions, assisting physicians in making more precise diagnostic and treatment decisions. It is a commonly used adjunct technique in modern interventional radiology and cardiovascular and cerebrovascular examinations.

[0003] However, existing angiography image enhancement methods have technical problems such as blurred vascular boundaries, obvious overlap between vascular and bone or soft tissue signals, and difficulty in accurately reflecting perfusion phase details; in existing vascular coding processes, the angiography perfusion process is simply processed by time averaging in traditional technologies, which cannot truly reflect the changes in the flow state of contrast agents in blood vessels, and in particular, there is a technical problem of unclear staging boundaries; in existing blood-bone separation processes, the high-density bone structure easily interferes with vascular shape identification, especially when the blood vessels cross the pelvis, thorax or base of the skull, and traditional methods often cannot accurately distinguish between bone shadows and vascular shadows; in existing vascular edge processing processes, angiography images often have unclear boundaries of small blood vessels, jagged or discontinuous edges, making it difficult to control the enhancement accuracy of vascular edges. Summary of the Invention

[0004] In view of the above situation, in order to overcome the shortcomings of the existing technology, the present invention provides a method and system for angiographic image enhancement for radiological examination. In order to address the technical problems in existing angiographic image enhancement methods, such as blurred vascular boundaries, obvious overlap between vascular and bone or soft tissue signals, and difficulty in accurately reflecting perfusion phase details, the existing technology usually relies only on frame averaging or simple mask suppression methods, which makes it difficult to simultaneously retain dynamic features and suppress background interference. The present solution proposes for the first time a comprehensive angiographic image enhancement method that combines vascular flow pattern coding, blood-bone dynamic decoupling and separation, and vascular edge reconstruction. This method effectively suppresses the influence of background interference structures while maintaining the integrity of vascular dynamic features, enhances the display clarity of key perfusion structures, and greatly improves the visibility and diagnostic reliability of vascular lesions. In order to address the technical problems in the existing vascular coding process, the angiographic perfusion process is simply processed by time averaging in traditional technologies, which cannot truly reflect the changes in the flow state of contrast agents in blood vessels, and in particular, there is a technical problem that the stage boundaries are unclear. The present solution creatively adopts a multi-phase time-series guided perfusion flow pattern coding method to extract the perfusion process in stages and evaluate the flow velocity and perfusion separately. Intensity features can help clinically identify pathophysiological features such as arteriovenous shunts, hemangiomas, and perfusion delays. In existing blood-bone separation processes, high-density bone structures can easily interfere with vascular shape identification, especially when blood vessels cross the pelvis, thorax, or base of the skull. Traditional methods often cannot accurately distinguish bone from vascular shadows. This solution innovatively uses a dynamic decoupling method of vascular bone structure combined with adversarial feature separation to perform blood-bone dynamic decoupling. This effectively isolates dynamic perfusion components from static high-density bone background signals, improving vascular structure clarity and integrity and avoiding image artifacts or obscurations caused by misjudgment. In existing vascular edge processing processes, angiographic images often have unclear boundaries, jagged or discontinuous edges of small blood vessels, making it difficult to accurately control vascular edge enhancement. This solution innovatively uses a vascular edge-aware reconstruction method combined with an improved edge distillation network for vascular edge reconstruction. This method enhances vascular edge sharpness while suppressing over-enhancement in non-edge areas, effectively improving the display consistency of microvessels and providing more reliable imaging evidence for the accurate interpretation of arteriovenous stenosis and collateral circulation lesions.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for enhancing angiographic images for radiological examination, the method comprising the following steps:

[0006] Step S1: collecting angiographic images;

[0007] Step S2: vascular flow pattern coding;

[0008] Step S3: dynamic decoupling of blood and bones;

[0009] Step S4: vessel edge reconstruction;

[0010] Step S5: Angiography image enhancement.

[0011] Furthermore, in step S1, the angiography image collection is used to obtain an angiography process image sequence, specifically by obtaining angiography image raw data from a radiological examination image sequence library through angiography image collection, and obtaining angiography image optimized data through basic attribute enhancement;

[0012] The angiographic image collection specifically collects angiographic images of the entire process before, during, and after contrast agent injection;

[0013] The basic attribute enhancement includes image sequence preprocessing and image attribute enhancement processing.

[0014] Furthermore, in step S2, the vascular flow state coding is used to extract the dynamic characteristics of the contrast agent perfusion process. Specifically, based on the original data of the angiography image, a multi-phase time-series guided perfusion flow state coding method is used to perform vascular flow state coding to obtain vascular flow state coding data, including the following steps:

[0015] Step S21: angiography time series grouping, specifically dividing the angiography image sequence in the angiography image raw data into perfusion phase segmented data according to the time dimension, wherein the perfusion phase segmented data specifically includes early perfusion segment data, mid-term filling segment data, and late reflow segment data; the division according to the time dimension is specifically divided according to the time point of the position with the maximum grayscale change in the angiography image sequence;

[0016] Step S22: Temporal grayscale change coding enhancement, specifically by calculating each perfusion phase;

[0017] Step S23: constructing an improved blood flow path structure graph, specifically constructing blood flow path guidance structure graph data, and improving the edge weights of the blood flow path guidance structure graph by introducing a blood flow similarity term and a blood flow space guidance term, thereby obtaining the improved blood flow path structure graph data;

[0018] The construction of the blood flow path guidance structure graph data includes node construction, edge construction, and edge weight definition; the node construction specifically defines each pixel in the angiography image as a node; the edge construction specifically defines the adjacency relationship between pixels as an edge; the edge weight definition specifically improves the edge weight by introducing a blood flow similarity term and a blood flow space guidance term;

[0019] Step S24: constructing a spatially aggregated vascular flow pattern vector, specifically, constructing a spatially aggregated vascular flow pattern vector based on the improved edge weights of the improved blood flow path structure graph data and the local time embedding feature data to obtain a vascular flow pattern aggregate vector;

[0020] Step S25: Vascular flow pattern coding, specifically, performing vector splicing based on the vascular flow pattern aggregation vector and the local time embedding feature data to construct vascular flow pattern coding data.

[0021] Furthermore, in step S3, the blood-bone dynamic decoupling is used to separate bone artifacts from blood flow signals. Specifically, based on the angiography image optimization data and the vascular flow state coding data, a vascular-bone structure dynamic decoupling method combined with adversarial feature separation is used to perform blood-bone dynamic decoupling to obtain blood flow dynamic reference image and bone background reference image data, including the following steps:

[0022] Step S31: constructing a vascular skeleton classification encoder, specifically constructing a joint encoder, extracting skeleton features and blood flow features based on the angiography image optimization data and the vascular flow pattern coding data to obtain skeleton feature data and blood flow feature data; the joint encoder includes a static skeleton feature encoder and a dynamic blood flow feature encoder;

[0023] Step S32: calculating the vascular skeleton feature constraint, specifically by introducing a feature orthogonal constraint loss function, optimizing the overlapping and mixed features extracted by the joint encoder, and obtaining the vascular skeleton feature constraint loss function;

[0024] Step S33: constructing an adversarial separation subnet, specifically by constructing a separation discriminator and using the vascular skeleton classification encoder as a generator to perform generative adversarial training to obtain an adversarial separation subnet; the separation discriminator specifically uses the skeleton feature data and blood flow feature data as data input and outputs a blood-bone prediction label;

[0025] The adversarial separation subnet adopts a standard adversarial loss function as the separation discriminator loss function and a vascular skeleton feature constraint loss function as the generator loss function;

[0026] Step S34: Structural coherence guided loss improvement, specifically constructing a vascular structural coherence loss based on the spatial continuity and topological connectivity of the blood vessels, and constructing a skeletal structural correlation loss based on the local high-frequency block density characteristics of the skeletal structure, for optimizing the structural preservation of the blood vessels and bones respectively; through the structural correlation guided loss improvement, combined with the standard adversarial loss function and the vascular-skeletal feature constraint loss function, a total loss is constructed to obtain a comprehensive loss function;

[0027] The vascular structure coherence loss is calculated using a joint loss of the edge map L1 norm loss and the regional sparsity;

[0028] The coherence loss of the bone structure is calculated using the Laplacian response loss;

[0029] The comprehensive loss function is constructed based on the vascular skeleton feature constraint loss function, the standard adversarial loss function, the vascular structure coherence loss and the skeletal structure coherence loss;

[0030] Step S35: Fusion decoding blood-bone decoupling, specifically by constructing a decoder and performing model training optimization based on the comprehensive loss function to obtain a blood-bone decoupling model, and using the blood-bone decoupling model to perform blood-bone dynamic decoupling to obtain blood flow dynamic reference image and bone background reference image data.

[0031] Furthermore, in step S4, the vascular edge reconstruction is used to restore the vascular boundary and structure. Specifically, based on the blood flow dynamic reference map, a vascular edge-aware reconstruction method combined with an improved edge distillation network is used to perform vascular edge reconstruction to obtain a vascular edge reference map, including the following steps:

[0032] Step S41: edge candidate map extraction, specifically, extracting the edge candidate map by using a combined gradient edge extraction operator to obtain edge basic candidate map data; the combined gradient edge extraction operator specifically uses a multi-scale Sobel-Laplacian composite gradient operator;

[0033] Step S42: constructing an improved edge distillation network, specifically, based on the edge base candidate image data, by constructing a shallow multi-scale U-shaped network improved by combining edge fusion loss, and mapping the edge base candidate image data into refined edge response image data to obtain response edge image data;

[0034] The improved edge distillation network adopts a U-shaped network infrastructure, including an encoder module, a dilated convolution bottleneck module, and a decoder module; the encoder is used to extract multi-scale edge semantic features of the input image; the dilated convolution bottleneck module is used to expand the receptive field and connect broken edges; the decoder module is used to restore the image resolution and edge map;

[0035] The improved basic loss function of the edge distillation network is used to improve edge fusion;

[0036] Step S43: constructing a morphology-preserving constraint loss for preserving the bifurcated linear structural features of the blood vessels. Specifically, the morphology-preserving constraint loss is constructed by adopting a topology architecture extraction algorithm to obtain a morphology-preserving constraint loss function. An edge reconstruction joint loss is constructed by combining the basic loss function and the morphology-preserving constraint loss function to obtain an edge reconstruction joint loss function. Model training of an improved edge distillation network is performed based on the edge reconstruction joint loss function.

[0037] Step S44: Blood vessel edge reconstruction, specifically, generating a blood vessel edge map by fusing the response edge image data output by the improved edge distillation network and the edge basis candidate map data to obtain fused edge map reconstructed data.

[0038] Furthermore, in step S5, the angiography image enhancement is used to generate a clear image and optimize the radiological reading, specifically by combining the blood flow dynamic reference map, the bone background reference map data, the blood vessel edge reference map and the angiography image optimization data to perform comprehensive enhancement of the angiography image to obtain radiation-adapted enhanced angiography image data.

[0039] The present invention provides an angiography image enhancement system for radiological examination, comprising an angiography image collection module, a vascular flow state encoding module, a blood-bone dynamic decoupling module, a vascular edge reconstruction module, and an angiography image enhancement module;

[0040] The angiography image collection module is used to collect angiography images, obtain angiography image optimization data through angiography image collection, and send the angiography image optimization data to the vascular flow encoding module, the blood-bone dynamic decoupling module and the angiography image enhancement module;

[0041] The vascular flow state coding module is used for vascular flow state coding, obtains vascular flow state coding data through vascular flow state coding, and sends the vascular flow state coding data to the blood-bone dynamic decoupling module;

[0042] The blood-bone dynamic decoupling module is used for dynamic decoupling of blood and bone, obtains blood flow dynamic reference image and bone background reference image data through blood-bone dynamic decoupling, and sends the blood flow dynamic reference image and bone background reference image data to the blood vessel edge reconstruction module and the angiography image enhancement module;

[0043] The blood vessel edge reconstruction module is used for reconstructing blood vessel edges, obtaining a blood vessel edge reference image through blood vessel edge reconstruction, and sending the blood vessel edge reference image to the angiography image enhancement module;

[0044] The angiography image enhancement module is used for angiography image enhancement, and obtains radiation-adapted enhanced angiography image data through angiography image enhancement.

[0045] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0046] (1) In the existing angiography image enhancement methods, there are technical problems such as blurred vascular boundaries, obvious overlap of vascular and bone or soft tissue signals, and difficulty in accurately reflecting the details of the perfusion phase. Due to the rapid transfer of contrast agents between the arteriovenous system, arteries and veins overlap on the image, making it difficult for doctors to judge the degree of arterial stenosis. Existing technologies usually rely only on frame averaging or simple mask suppression methods, which are difficult to simultaneously retain dynamic features and suppress background interference. This scheme proposes for the first time a comprehensive angiography image enhancement method that combines vascular flow coding, blood-bone dynamic decoupling and separation, and vascular edge reconstruction. It can effectively suppress the influence of background interference structures while maintaining the integrity of vascular dynamic features, enhance the display clarity of key perfusion structures, and greatly improve the visibility and diagnostic credibility of vascular lesions.

[0047] (2) In the existing vascular coding process, the angiographic perfusion process is simply processed by time averaging in traditional technologies, which cannot truly reflect the changes in the flow state of contrast agents in blood vessels, especially the technical problem of unclear stage boundaries. Specifically, the arterial phase image in the early perfusion period is covered by the venous phase signal in the later period, resulting in incomplete development of arterial stenosis or dissected blood vessels; and lesions such as venous fistulas or arteriovenous malformations that rely on blood flow direction cannot reflect their true path through simple frame averaging. At the same time, in multi-phase CT perfusion images, the temporal density change trend cannot be accurately modeled, resulting in deviations in the judgment of the tumor blood supply source; and the pathological differences in perfusion speed are diluted in the average image, which will affect the judgment of early intervention. This scheme creatively adopts a multi-phase time-series guided perfusion flow coding method to realize the staged extraction of the perfusion process and the evaluation of flow velocity and perfusion intensity characteristics respectively, which is helpful for clinical identification of pathological and physiological characteristics such as arteriovenous shunts, hemangiomas, and perfusion delays.

[0048] (3) In the existing blood-bone separation process, there is a technical problem that the high-density bone structure easily interferes with the identification of blood vessel shape, especially when the blood vessels cross the pelvis, thorax or base of the skull. Traditional methods are often unable to accurately distinguish between bone shadows and blood vessel shadows. For example, conventional bone-blood separation methods are very likely to fail during patient movement, uneven dilution of contrast agents or low-dose scanning; and in the assessment of fractures and bleeding, if the fracture fragments and bleeding vessels cannot be accurately separated, it will affect the formulation of preoperative embolization or intervention plans; this solution creatively adopts a dynamic decoupling method of vascular bone structure combined with adversarial feature separation to perform dynamic decoupling of blood and bone, effectively isolating dynamic perfusion components from static high-density bone background signals, improving the clarity and structural integrity of vascular structures, and avoiding image artifacts or masking caused by misjudgment;

[0049] (4) In the existing vascular edge processing process, there is a technical problem that the boundaries of small blood vessels in angiography images are often unclear, and the edges are jagged or discontinuous, making it difficult to control the enhancement accuracy of the vascular edges. For example, when diagnosing limb peripheral artery occlusion or small blood supply arteries of liver cancer, traditional image sharpening methods are prone to mistakenly enhance small blood vessels as noise, or cause discontinuous display of blood vessels due to insufficient contrast. The image sharpening methods commonly used in traditional methods may over-enhance the edges, causing the originally smooth blood vessel contours to appear "jagged" edges, interfering with the doctor's judgment of whether there are pathological changes in the blood vessel wall; this scheme creatively adopts a vascular edge perception reconstruction method combined with an improved edge distillation network to reconstruct the vascular edges, achieving the goal of enhancing the sharpness of the vascular edges while suppressing the excessive enhancement of non-edge areas, effectively improving the display continuity of micro-vessels, and providing a more reliable image basis for the accurate interpretation of arteriovenous stenosis and collateral circulation lesions. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic flow chart of a method for enhancing angiographic images for radiological examination provided by the present invention;

[0051] Figure 2 A schematic diagram of an angiographic image enhancement system for radiological examination provided by the present invention;

[0052] Figure 3 A schematic diagram of the process of coding the vascular flow pattern in step S2;

[0053] Figure 4 This is a flow chart of the blood-bone dynamic decoupling process in step S3;

[0054] Figure 5 Schematic diagram of the process of blood vessel edge reconstruction in step S4.

[0055] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0057] Example 1, see Figure 1 The present invention provides a method for enhancing angiographic images for radiological examination, the method comprising the following steps:

[0058] Step S1: collecting angiographic images;

[0059] Step S2: vascular flow pattern coding;

[0060] Step S3: dynamic decoupling of blood and bones;

[0061] Step S4: vessel edge reconstruction;

[0062] Step S5: Angiography image enhancement.

[0063] By performing the above operations, technical problems such as blurred vascular boundaries, obvious overlap of vascular and bone or soft tissue signals, and difficulty in accurately reflecting perfusion phase details exist in existing angiography image enhancement methods. Since the contrast agent is rapidly transferred between the arteriovenous system, the arteries and veins overlap with each other in the image, making it difficult for doctors to determine the degree of arterial stenosis. Existing technologies usually only rely on frame averaging or simple mask suppression methods, which make it difficult to simultaneously retain dynamic features and suppress background interference. This scheme proposes for the first time a comprehensive angiography image enhancement method that combines vascular flow coding, blood-bone dynamic decoupling and separation, and vascular edge reconstruction. It achieves the goal of maintaining the integrity of vascular dynamic features while effectively suppressing the influence of background interference structures, enhancing the display clarity of key perfusion structures, and greatly improving the visibility and diagnostic credibility of vascular lesions.

[0064] Example 2, see Figure 1 、 Figure 2 In step S1, the angiography image collection is used to obtain an angiography process image sequence, specifically by collecting angiography images from a radiological examination image sequence library to obtain angiography image raw data, and enhancing the angiography image by basic attributes to obtain angiography image optimized data;

[0065] The angiographic image collection specifically collects angiographic images of the entire process before, during, and after contrast agent injection;

[0066] The basic attribute enhancement includes image sequence preprocessing and image attribute enhancement processing; the image sequence preprocessing includes histogram matching and normalization, time alignment, resolution normalization and Gaussian filtering time series denoising; the image attribute enhancement processing includes grayscale normalization and mild histogram equalization optimization.

[0067] Example 3, see Figure 1 、 Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the vascular flow state coding is used to extract the dynamic characteristics of the contrast agent perfusion process. Specifically, based on the original data of the angiography image, a multi-phase time-series guided perfusion flow state coding method is used to perform vascular flow state coding to obtain vascular flow state coding data, including the following steps:

[0068] Step S21: Angiography time series grouping, specifically, dividing the angiography image sequence in the angiography image raw data into perfusion phase segmented data according to the time dimension, wherein the perfusion phase segmented data specifically includes early perfusion segment data, mid-term filling segment data, and late reflow segment data; the division according to the time dimension is specifically divided according to the time point of the position with the maximum grayscale change of the angiography image sequence, and the calculation formula is:

[0069] ;

[0070] Where, P k is the perfusion phase segment data, k is the perfusion phase segment index, I(t) is the grayscale value of the angiographic image at time point t, and t is the time index.

[0071] is the time boundary of the kth perfusion phase segment;

[0072] Preferably, the time boundary , calculated by the leading frame difference method, the calculation formula is:

[0073] ;

[0074] Where, is the time boundary of the kth perfusion phase segment, R is the vascular region identifier in the angiography image, x is the horizontal index of the pixel, y is the vertical index of the pixel, and I(t,x,y) is the grayscale value of the angiography image at the corresponding pixel point (x,y) at time point t;

[0075] Step S22: Temporal grayscale change coding enhancement, specifically, by calculating the grayscale change amplitude and speed characteristics of each pixel point in each phase segment of the perfusion phase segment data, constructing local time embedded feature data, the calculation formula is:

[0076] ;

[0077] Where, f dyn (x,y) is the local time embedded feature data corresponding to the pixel point (x,y), The overall grayscale change amplitude term is used to represent the perfusion intensity. The overall grayscale change rate term is used to represent the local perfusion rate, which is calculated approximately by the first-order difference;

[0078] Step S23: constructing an improved blood flow path structure graph, specifically constructing blood flow path guidance structure graph data, and improving the edge weights of the blood flow path guidance structure graph by introducing a blood flow similarity term and a blood flow space guidance term, thereby obtaining the improved blood flow path structure graph data;

[0079] The construction of the blood flow path guidance structure graph data includes node construction, edge construction, and edge weight definition. The node construction specifically defines each pixel in the angiography image as a node. The edge construction specifically defines the adjacency relationship between pixels as an edge. The edge weight definition specifically improves the edge weight by introducing a blood flow similarity term and a blood flow space guidance term. The calculation formula is:

[0080] ;

[0081] Where w ij is the improved edge weight, f dyn (v i ) is the local time embedding feature data corresponding to the i-th node, f dyn (v j ) is the local time embedding feature data corresponding to the jth neighbor node, where i is the node index, j is the neighbor node index, is the blood flow similarity attenuation parameter, p i is the pixel coordinate corresponding to the i-th node, p j is the pixel coordinate corresponding to the jth adjacent node, is the attenuation parameter of the blood flow spatial guidance term;

[0082] Step S24: constructing a spatially aggregated vascular flow pattern vector, specifically, constructing a spatially aggregated vascular flow pattern vector based on the improved edge weights of the improved blood flow path structure graph data and the local time embedding feature data to obtain a vascular flow pattern aggregation vector. The calculation formula is:

[0083] ;

[0084] Where, F flow (v i ) is the vascular flow aggregation vector, v j is the jth neighbor node, N(v i ) is the set of neighboring nodes of the i-th node, f dyn (v j ) is the local time embedding feature data corresponding to the j-th neighbor node;

[0085] Step S25: Vascular flow pattern coding, specifically, performing vector splicing based on the vascular flow pattern aggregation vector and the local time embedded feature data to construct vascular flow pattern coding data. The calculation formula is:

[0086] ;

[0087] Where PSM(x,y) is the vascular flow state encoding data, which is used to represent the grayscale change amplitude, grayscale change speed, neighboring node aggregation dynamics and flow direction consistency of each pixel point, and f dyn (x,y) is the local time embedded feature data corresponding to the pixel point (x,y), F flow (x,y) is the vascular flow aggregation vector corresponding to the pixel point (x,y), It is a dimensional identifier of the vascular flow state coding data, and the dimensions of the vascular flow state coding data specifically include the grayscale change dimension, the grayscale change speed dimension, the neighbor node aggregation dynamic dimension and the flow direction consistency dimension; the grayscale change amplitude is specifically calculated by the blood flow similarity item, the grayscale change speed is specifically calculated by the blood flow space guidance item, the neighbor node aggregation dynamic is specifically represented by the vascular flow state aggregation vector, and the flow direction consistency is specifically represented by calculating the cosine similarity between the local time embedded feature data.

[0088] By performing the above operations, the existing vascular coding process has the problem that the angiography perfusion process is simply processed by time averaging in traditional technologies, which cannot truly reflect the changes in the flow state of the contrast agent in the blood vessels. In particular, there is a technical problem that the boundaries of the stages (such as the early arterial segment and the late venous segment) are unclear. Specifically, the arterial phase image in the early perfusion stage is covered by the late venous phase signal, resulting in incomplete visualization of arterial stenosis or dissected blood vessels; and lesions such as venous fistulas or arteriovenous malformations that rely on blood flow direction cannot reflect their true paths through simple frame averaging. In multi-phase CT perfusion images, the temporal density variation trend cannot be accurately modeled, resulting in deviations in the judgment of the source of tumor blood supply (such as multiple renal artery blood supply); and the pathological differences in perfusion speed (such as early perfusion delay in ischemic stroke) are diluted in the average image, which will affect the judgment of early intervention. This scheme creatively adopts a multi-phase time-series guided perfusion flow coding method to realize the staged extraction of the perfusion process and the separate evaluation of flow velocity and perfusion intensity characteristics, which is helpful for clinical identification of pathophysiological characteristics such as arteriovenous shunts, hemangiomas, and perfusion delays.

[0089] Example 4, see Figure 1 、 Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the blood-bone dynamic decoupling is used to separate bone artifacts from blood flow signals. Specifically, based on the angiography image optimization data and the vascular flow state coding data, a vascular-bone structure dynamic decoupling method combined with adversarial feature separation is used to perform blood-bone dynamic decoupling to obtain blood flow dynamic reference image and bone background reference image data, including the following steps:

[0090] Step S31: Constructing a vascular skeleton classification encoder, specifically constructing a joint encoder, extracting skeleton features and blood flow features based on the angiography image optimization data and the vascular flow state coding data to obtain skeleton feature data and blood flow feature data; the joint encoder includes a static skeleton feature encoder and a dynamic blood flow feature encoder, and the calculation formula is:

[0091] ;

[0092] Where Z b is the bone feature data, E b is a static skeleton encoder, X is a joint data input for representing the angiography image optimization data and the vascular flow state encoding data, and Z f is the blood flow characteristic data, E f It is a dynamic blood flow encoder;

[0093] Step S32: Calculate the vascular skeleton feature constraint. Specifically, by introducing a feature orthogonal constraint loss function, optimize the overlapping and mixed features extracted by the joint encoder, and obtain the vascular skeleton feature constraint loss function. The calculation formula is:

[0094] ;

[0095] Where, is the vascular skeleton feature constraint loss function, is the transposed matrix of bone feature data, is the Frobenius norm operator;

[0096] Step S33: constructing an adversarial separation subnet, specifically by constructing a separation discriminator and using the vascular skeleton classification encoder as a generator to perform generative adversarial training to obtain an adversarial separation subnet; the separation discriminator specifically uses the skeleton feature data and blood flow feature data as data input and outputs a blood-bone prediction label;

[0097] The adversarial separation subnet adopts a standard adversarial loss function as the separation discriminator loss function and a vascular skeleton feature constraint loss function as the generator loss function;

[0098] Step S34: Structural coherence guided loss improvement, specifically constructing a vascular structural coherence loss based on the spatial continuity and topological connectivity of the blood vessels, and constructing a skeletal structural correlation loss based on the local high-frequency block density characteristics of the skeletal structure, for optimizing the structural preservation of the blood vessels and bones respectively; through the structural correlation guided loss improvement, combined with the standard adversarial loss function and the vascular-skeletal feature constraint loss function, a total loss is constructed to obtain a comprehensive loss function;

[0099] The vascular structure coherence loss is calculated using the joint loss of the edge map L1 norm loss and the regional sparsity loss, and the calculation formula is:

[0100] ;

[0101] Where, is the loss of coherence of vascular structures, is the L1 norm loss weight, the default value is 0.6, is the gradient map of blood flow characteristic data, is the gradient map of vascular flow coding data, is the L1 norm operator, is the regional sparsity weight, the default value is 0.4, is the total variation regularization term, which is used to represent the sparsity of the image region and improve the overall coherence of the image;

[0102] The coherence loss of the bone structure is calculated using the Laplacian response loss, and the calculation formula is:

[0103] ;

[0104] Where, is the loss of coherence of the bone structure, is the second-order gradient map of bone feature data, is the second-order gradient map of the optimized data of angiographic images;

[0105] The comprehensive loss function is constructed based on the vascular skeleton feature constraint loss function, the standard adversarial loss function, the vascular structure coherence loss and the skeletal structure coherence loss. The calculation formula is:

[0106] ;

[0107] Where, is the comprehensive loss function, is the vascular skeleton feature constraint loss function, is the standard adversarial loss function, is the coherence weight of the vascular structure, is the loss of coherence of vascular structures, is the bone structure coherence weight, is the loss of coherence of the skeletal structure;

[0108] Preferably, the vascular structure coherence weight The default value of is 0.45, and the bone structure coherence weight The default value of is 0.55;

[0109] Step S35: Fusion decoding blood-bone decoupling, specifically by constructing a decoder and performing model training optimization based on the comprehensive loss function to obtain a blood-bone decoupling model, and using the blood-bone decoupling model to perform blood-bone dynamic decoupling to obtain blood flow dynamic reference image and bone background reference image data.

[0110] By performing the above operations, the technical problem that high-density bone structures easily interfere with the identification of vascular shapes in the existing blood-bone separation process, especially when the blood vessels cross the pelvis, thorax or base of the skull, is often not accurately distinguished between bone shadows and vascular shadows by traditional methods. For example, conventional bone-blood separation methods (such as those based on simple thresholds or morphological processing) are very likely to fail during patient movement, uneven dilution of contrast agents or low-dose scanning; and in the assessment of fractures and bleeding (such as pelvic trauma), if the fracture fragments and bleeding vessels cannot be accurately separated, it will affect the formulation of preoperative embolization or interventional plans; this solution creatively adopts a dynamic decoupling method of vascular bone structure combined with adversarial feature separation to perform dynamic decoupling of blood and bone, effectively isolating dynamic perfusion components from static high-density bone background signals, improving the clarity and structural integrity of vascular structures, and avoiding image artifacts or obscurations caused by misjudgment.

[0111] Example 5, see Figure 1 、 Figure 2 and Figure 5 In step S4, the vascular edge reconstruction is used to restore the vascular boundary and structure. Specifically, based on the blood flow dynamic reference map, a vascular edge perception reconstruction method combined with an improved edge distillation network is used to perform vascular edge reconstruction to obtain a vascular edge reference map, including the following steps:

[0112] Step S41: edge candidate map extraction, specifically, extracting the edge candidate map by using a combined gradient edge extraction operator to obtain edge basic candidate map data; the combined gradient edge extraction operator specifically uses a multi-scale Sobel-Laplacian composite gradient operator, and the calculation formula is:

[0113] ;

[0114] Where, E init It is the edge base candidate image data, x is the horizontal index of the pixel, y is the vertical index of the pixel, is the horizontal gradient of the blood flow dynamics reference map, is the vertical gradient of the blood flow dynamics reference map, It is the second-order gradient map of the blood flow dynamic reference map;

[0115] Step S42: constructing an improved edge distillation network, specifically, based on the edge base candidate image data, by constructing a shallow multi-scale U-shaped network improved by combining edge fusion loss, and mapping the edge base candidate image data into refined edge response image data to obtain response edge image data;

[0116] The calculation formula of the improved edge distillation network is:

[0117] ;

[0118] Where, I edge is the response edge image data output by the improved edge distillation network, EDN is the improved edge distillation network identifier, E init is the edge base candidate graph data, It is a combination of model parameters for improving the edge distillation network;

[0119] The improved edge distillation network adopts a U-shaped network infrastructure, including an encoder module, a dilated convolution bottleneck module, and a decoder module; the encoder is used to extract multi-scale edge semantic features of the input image; the dilated convolution bottleneck module is used to expand the receptive field and connect broken edges; the decoder module is used to restore the image resolution and edge map;

[0120] Preferably, Table 1 is an example parameter table of the improved edge distillation network. As shown in the table, the encoder module includes a first convolution block, a second convolution block and a maximum pooling layer;

[0121] The convolution block includes a convolution layer of size 3×3, stride 1 and padding 1, a batch normalization layer and a ReLU activation layer; the first convolution block has 2 input channels and 32 output channels, followed by a maximum pooling layer; the window size of the maximum pooling layer is set to 2×2 and the stride is 2; the second convolution block has 32 input channels and 64 output channels;

[0122] The dilated convolution bottleneck module includes three parallel convolution channels and a dimensionality reduction convolution layer;

[0123] The parallel convolution channel includes a dilated convolution layer; the convolution kernel size of the dilated convolution layer is 3×3, the stride is 1, the number of input channels is 64, and the number of output channels is 64; each parallel convolution channel layer is followed by a batch normalization layer and a ReLU activation layer;

[0124] The dilation rates of the three parallel convolution channels are set to 1, 2, and 4 respectively;

[0125] The dimensionality reduction convolution layer concatenates the outputs of each parallel convolution channel and then performs a 1×1 convolution to reduce the number of channels from 192 to 64.

[0126] The decoder module includes an upsampling layer, a first decoding convolution block, a second decoding convolution block, and an output layer; the upsampling layer uses a bilinear interpolation method for upsampling; the decoding convolution block includes two convolution layers of size 3×3, stride 1 and padding 1, a batch normalization layer, and a ReLU activation layer;

[0127] The first decoding convolution block has 64 input channels and 32 output channels; the second decoding convolution block has 32 input channels and 16 output channels;

[0128] The output layer specifically adopts 1×1 convolution, the output channel is 1, and the activation function is a sigmoid function;

[0129] Table 1 Parameter example of improved edge distillation network

[0130]

[0131] The basic loss function of the improved edge distillation network is used to improve edge fusion, and the calculation formula is:

[0132] ;

[0133] Where, L distill is the basic loss function, is the gradient map of the response edge image data, is a gradient map of a blood flow dynamics reference map, and M is a vascular region mask, which is obtained based on the blood flow dynamics reference map by using an adaptive threshold screening method;

[0134] Step S43: constructing a morphology-preserving constraint loss for preserving the bifurcated linear structural features of the blood vessels. Specifically, the morphology-preserving constraint loss is constructed by adopting a topology architecture extraction algorithm to obtain a morphology-preserving constraint loss function. An edge reconstruction joint loss is constructed by combining the basic loss function and the morphology-preserving constraint loss function to obtain an edge reconstruction joint loss function. Model training of an improved edge distillation network is performed based on the edge reconstruction joint loss function.

[0135] The calculation formula of the morphology preservation constraint loss function is:

[0136] ;

[0137] Where, L morph is the morphology preservation constraint loss function, Skel(·) is the topology architecture extraction function, I edge is the response edge image data output by the improved edge distillation network, and M is the blood vessel region mask;

[0138] Preferably, the topology architecture extraction function specifically uses the ximgproc.thinning method in OpenCV to perform topology architecture extraction;

[0139] The calculation formula of the edge reconstruction joint loss function is:

[0140] ;

[0141] Where, L total is the edge reconstruction joint loss function, e1 is the basic loss weight, the default value is 0.7, L distill is the basic loss function, e2 is the shape preservation constraint loss weight, the default value is 0.3, L morph is the morphology-preserving constraint loss function;

[0142] Step S44: Blood vessel edge reconstruction is specifically performed by fusing the response edge image data output by the improved edge distillation network and the edge basic candidate map data to generate a blood vessel edge map, thereby obtaining fused edge map reconstruction data. The calculation formula is:

[0143] ;

[0144] Where, is the fusion edge map reconstruction data, Sig is the S-type activation function, C is the edge reconstruction fusion weight, the default value is 0.7, I edge is the response edge image data output by the improved edge distillation network, E init It is the edge basic candidate graph data.

[0145] By performing the above operations, the technical problem of unclear boundaries, jagged or discontinuous edges of small blood vessels in existing blood vessel edge processing, which makes it difficult to control the enhancement accuracy of blood vessel edges, exists in angiography images. For example, when diagnosing limb peripheral artery occlusion (such as diabetic foot artery embolism) or small blood supply arteries of liver cancer, traditional image sharpening methods are prone to mistakenly enhance small blood vessels as noise, or cause intermittent display of blood vessels due to insufficient contrast. The image sharpening methods commonly used in traditional methods may over-enhance edges, causing the originally smooth blood vessel contours to appear "jagged" edges, which interferes with the doctor's judgment of whether there are pathological changes in the blood vessel wall. This solution creatively adopts a blood vessel edge-aware reconstruction method combined with an improved edge distillation network to perform blood vessel edge reconstruction. This achieves the goal of enhancing the sharpness of blood vessel edges while suppressing over-enhancement of non-edge areas, effectively improving the display consistency of microvessels, and providing more reliable image basis for the accurate interpretation of arteriovenous stenosis and collateral circulation lesions.

[0146] Example 6, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S5, the angiography image enhancement is used to generate a clear image and optimize the radiological reading. Specifically, the angiography image is comprehensively enhanced by combining the blood flow dynamic reference map, the bone background reference map data, the blood vessel edge reference map and the angiography image optimization data to obtain radiation-adapted enhanced angiography image data.

[0147] Example 7, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The present invention provides an angiography image enhancement system for radiological examination, comprising an angiography image collection module, a vascular flow state encoding module, a blood-bone dynamic decoupling module, a vascular edge reconstruction module, and an angiography image enhancement module.

[0148] The angiography image collection module is used to collect angiography images, obtain angiography image optimization data through angiography image collection, and send the angiography image optimization data to the vascular flow encoding module, the blood-bone dynamic decoupling module and the angiography image enhancement module;

[0149] The vascular flow state coding module is used for vascular flow state coding, obtains vascular flow state coding data through vascular flow state coding, and sends the vascular flow state coding data to the blood-bone dynamic decoupling module;

[0150] The blood-bone dynamic decoupling module is used for dynamic decoupling of blood and bone, obtains blood flow dynamic reference image and bone background reference image data through blood-bone dynamic decoupling, and sends the blood flow dynamic reference image and bone background reference image data to the blood vessel edge reconstruction module and the angiography image enhancement module;

[0151] The blood vessel edge reconstruction module is used for reconstructing blood vessel edges, obtaining a blood vessel edge reference image through blood vessel edge reconstruction, and sending the blood vessel edge reference image to the angiography image enhancement module;

[0152] The angiography image enhancement module is used for angiography image enhancement, and obtains radiation-adapted enhanced angiography image data through angiography image enhancement.

[0153] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0154] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0155] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A method for enhancing angiographic images for radiological examination, characterized in that: The method comprises the following steps: Step S1: collecting angiographic images, obtaining raw data of angiographic images through angiographic image collection, and obtaining optimized data of angiographic images through basic attribute enhancement; Step S2: Vascular flow pattern coding, using a multi-phase time-series guided perfusion flow pattern coding method to perform vascular flow pattern coding to obtain vascular flow pattern coding data, including the following steps: angiography time series grouping; time series grayscale change coding enhancement; constructing blood flow path guidance structure diagram data, and improving the edge weight of the blood flow path guidance structure diagram by introducing blood flow similarity terms and blood flow space guidance terms; constructing a spatially aggregated vascular flow pattern vector; and vascular flow pattern coding; Step S3: dynamic decoupling of blood and bones, using a dynamic decoupling method of blood vessel and bone structure combined with adversarial feature separation to perform dynamic decoupling of blood and bones, and obtain blood flow dynamic reference image and bone background reference image data; Step S4: Reconstructing blood vessel edges. A blood vessel edge-aware reconstruction method combined with an improved edge distillation network is used to reconstruct blood vessel edges and obtain a blood vessel edge reference map. The method includes the following steps: extracting edge candidate maps; constructing an improved edge distillation network by constructing a shallow multi-scale U-shaped network improved by combining edge fusion loss; constructing a morphology-preserving constraint loss; and reconstructing blood vessel edges. Step S5: Angiography image enhancement to obtain radiation-adapted enhanced angiography image data.

2. The method for enhancing angiographic images for radiological examination according to claim 1, wherein: In step S1, the angiography image collection is used to obtain an angiography process image sequence, specifically by collecting angiography images from a radiological examination image sequence library to obtain angiography image raw data, and enhancing the angiography image by basic attributes to obtain angiography image optimized data; The angiographic image collection specifically collects angiographic images of the entire process before, during, and after contrast agent injection; The basic attribute enhancement includes image sequence preprocessing and image attribute enhancement processing.

3. The method for enhancing angiographic images for radiological examination according to claim 2, wherein: In step S2, the vascular flow state coding is used to extract the dynamic characteristics of the contrast agent perfusion process. Specifically, based on the original data of the angiography image, a multi-phase time-series guided perfusion flow state coding method is used to perform vascular flow state coding to obtain vascular flow state coding data, including the following steps: Step S21: angiography time series grouping, specifically dividing the angiography image sequence in the angiography image raw data into perfusion phase segmented data according to the time dimension, wherein the perfusion phase segmented data specifically includes early perfusion segment data, mid-term filling segment data, and late reflow segment data; the division according to the time dimension is specifically divided according to the time point of the position with the maximum grayscale change in the angiography image sequence; Step S22: Temporal grayscale change coding enhancement, specifically, calculating the grayscale change amplitude and speed characteristics of each pixel point in each phase segment of the perfusion phase segment data to construct local time embedded feature data; Step S23: constructing an improved blood flow path structure graph, specifically constructing blood flow path guidance structure graph data, and improving the edge weights of the blood flow path guidance structure graph by introducing a blood flow similarity term and a blood flow space guidance term, thereby obtaining the improved blood flow path structure graph data; The construction of the blood flow path guidance structure graph data includes node construction, edge construction, and edge weight definition; the node construction specifically defines each pixel in the angiography image as a node; the edge construction specifically defines the adjacency relationship between pixels as an edge; the edge weight definition specifically improves the edge weight by introducing a blood flow similarity term and a blood flow space guidance term; Step S24: constructing a spatially aggregated vascular flow pattern vector, specifically, constructing a spatially aggregated vascular flow pattern vector based on the improved edge weights of the improved blood flow path structure graph data and the local time embedding feature data to obtain a vascular flow pattern aggregate vector; Step S25: Vascular flow pattern coding, specifically, performing vector splicing based on the vascular flow pattern aggregation vector and the local time embedding feature data to construct vascular flow pattern coding data.

4. The method for enhancing angiographic images for radiological examination according to claim 3, wherein: In step S3, the blood-bone dynamic decoupling is used to separate bone artifacts from blood flow signals. Specifically, based on the angiography image optimization data and the vascular flow state coding data, a vascular-bone structure dynamic decoupling method combined with adversarial feature separation is used to perform blood-bone dynamic decoupling to obtain blood flow dynamic reference image and bone background reference image data, including the following steps: Step S31: constructing a vascular skeleton classification encoder, specifically constructing a joint encoder, extracting skeleton features and blood flow features based on the angiography image optimization data and the vascular flow pattern coding data to obtain skeleton feature data and blood flow feature data; the joint encoder includes a static skeleton feature encoder and a dynamic blood flow feature encoder; Step S32: calculating the vascular skeleton feature constraint, specifically by introducing a feature orthogonal constraint loss function, optimizing the overlapping and mixed features extracted by the joint encoder, and obtaining the vascular skeleton feature constraint loss function; Step S33: constructing an adversarial separation subnet, specifically by constructing a separation discriminator and using the vascular skeleton classification encoder as a generator to perform generative adversarial training to obtain an adversarial separation subnet; the separation discriminator specifically uses the skeleton feature data and blood flow feature data as data input and outputs a blood-bone prediction label; the adversarial separation subnet uses a standard adversarial loss function as the separation discriminator loss function and uses a vascular skeleton feature constraint loss function as the generator loss function; Step S34: Structural coherence guided loss improvement, specifically constructing a vascular structural coherence loss based on the spatial continuity and topological connectivity of the blood vessels, and constructing a skeletal structural correlation loss based on the local high-frequency block density characteristics of the skeletal structure, for optimizing the structural preservation of the blood vessels and bones respectively; through the structural correlation guided loss improvement, combined with the standard adversarial loss function and the vascular-skeletal feature constraint loss function, a total loss is constructed to obtain a comprehensive loss function; Step S35: Fusion decoding blood-bone decoupling, specifically by constructing a decoder and performing model training optimization based on the comprehensive loss function to obtain a blood-bone decoupling model, and using the blood-bone decoupling model to perform blood-bone dynamic decoupling to obtain blood flow dynamic reference image and bone background reference image data.

5. The method for enhancing angiographic images for radiological examination according to claim 4, characterized in that: In step S34, the vascular structure coherence loss is calculated using the joint loss of the edge map L1 norm loss and the regional sparsity; the skeletal structure coherence loss is calculated using the Laplacian response loss; the comprehensive loss function is constructed based on the vascular skeleton feature constraint loss function, the standard adversarial loss function, the vascular structure coherence loss and the skeletal structure coherence loss.

6. The method for enhancing angiographic images for radiological examination according to claim 5, characterized in that: In step S4, the vascular edge reconstruction is used to restore the vascular boundary and structure. Specifically, based on the blood flow dynamic reference map, a vascular edge perception reconstruction method combined with an improved edge distillation network is used to perform vascular edge reconstruction to obtain a vascular edge reference map, including the following steps: Step S41: extracting edge candidate images, specifically, extracting edge candidate images by using a combined gradient edge extraction operator to obtain edge basic candidate image data; Step S42: constructing an improved edge distillation network, specifically, based on the edge basic candidate map data, by constructing a shallow multi-scale U-shaped network combined with an improved edge fusion loss, and performing an improved edge distillation network construction, mapping the edge basic candidate map data into refined edge response map data to obtain response edge image data; the improved edge distillation network adopts a U-shaped network infrastructure, including an encoder module, a hole convolution bottleneck module and a decoder module; the encoder is used to extract multi-scale edge semantic features of the input image; the hole convolution bottleneck module is used to expand the receptive field and connect broken edges; the decoder module is used to restore the image resolution and edge map; Step S43: constructing a morphology-preserving constraint loss for preserving the bifurcated linear structural features of the blood vessels. Specifically, the morphology-preserving constraint loss is constructed by adopting a topology architecture extraction algorithm to obtain a morphology-preserving constraint loss function. An edge reconstruction joint loss is constructed by combining a basic loss function and the morphology-preserving constraint loss function to obtain an edge reconstruction joint loss function. Model training of an improved edge distillation network is performed based on the edge reconstruction joint loss function. Step S44: Blood vessel edge reconstruction, specifically, generating a blood vessel edge map by fusing the response edge image data output by the improved edge distillation network and the edge basis candidate map data to obtain fused edge map reconstructed data.

7. The method for enhancing angiographic images for radiological examination according to claim 6, characterized in that: In step S5, the angiography image enhancement is used to generate a clear image and optimize the radiological reading. Specifically, the angiography image is comprehensively enhanced by combining the blood flow dynamic reference map, the bone background reference map data, the blood vessel edge reference map and the angiography image optimization data to obtain radiation-adapted enhanced angiography image data.

8. An angiographic image enhancement system for radiological examination, configured to implement the angiographic image enhancement method for radiological examination according to any one of claims 1 to 7, characterized in that: It includes angiography image collection module, vascular flow coding module, blood-bone dynamic decoupling module, vascular edge reconstruction module and angiography image enhancement module.

9. The angiographic image enhancement system for radiological examination according to claim 8, characterized in that: The angiography image collection module is used to collect angiography images, obtain angiography image optimization data through angiography image collection, and send the angiography image optimization data to the vascular flow encoding module, the blood-bone dynamic decoupling module and the angiography image enhancement module; The vascular flow state coding module is used for vascular flow state coding, obtains vascular flow state coding data through vascular flow state coding, and sends the vascular flow state coding data to the blood-bone dynamic decoupling module; The blood-bone dynamic decoupling module is used for dynamic decoupling of blood and bone, obtains blood flow dynamic reference image and bone background reference image data through blood-bone dynamic decoupling, and sends the blood flow dynamic reference image and bone background reference image data to the blood vessel edge reconstruction module and the angiography image enhancement module; The blood vessel edge reconstruction module is used for reconstructing blood vessel edges, obtaining a blood vessel edge reference image through blood vessel edge reconstruction, and sending the blood vessel edge reference image to the angiography image enhancement module; The angiography image enhancement module is used for angiography image enhancement, and obtains radiation-adapted enhanced angiography image data through angiography image enhancement.

Images