Coronary angiography image tumor fistula association intelligent analysis system

Abstract

The invention discloses a coronary angiography image tumor fistula association intelligent analysis system, and relates to the technical field of medical image intelligent analysis. Comprising an image acquisition module used for acquiring a coronary artery angiography image sequence containing different time phases; and the lesion identification and correlation analysis module is used for carrying out deep processing on the coronary artery angiography image sequence and constructing a coronary artery structure analysis model based on a graph neural network so as to automatically identify coronary arteries and coronary fistula. The multi-time-sequence coronary angiography images are obtained through the image obtaining module and preprocessed, then the pathological change recognition and correlation analysis module recognizes coronary artery tumors and coronary fistula at the same time through the graph neural network, and the spatial position relation and pathophysiological connection of the coronary artery tumors and the coronary fistula are accurately captured; the problems that tumor fistula cannot be recognized simultaneously and correlation analysis is insufficient in the prior art are solved. The time sequence angiography frame analysis module tracks a flow mode of a contrast agent and distinguishes thrombus in a tumor from an active fistula orifice, so that the defect that the area is difficult to distinguish accurately in the prior art is overcome.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical image analysis technology, specifically to an intelligent analysis system for the correlation between aneurysm and fistula in coronary angiography images. Background Technology

[0002] Coronary angiography is a core imaging technique for diagnosing coronary artery disease in clinical practice. Its principle is to inject contrast agent into the coronary arteries and obtain dynamic images of the coronary arteries in real time under X-ray fluoroscopy. Doctors can clearly observe the course of the coronary arteries, the morphology of the lumen, and whether there are lesions such as stenosis, dilation, or abnormal channels through these images.

[0003] Coronary aneurysms and coronary fistulas are two types of coronary artery abnormalities that require special attention: a coronary aneurysm refers to an abnormal dilation of a local lumen in the coronary artery, with the dilated diameter typically exceeding 1.5 times that of adjacent normal vessels; a coronary fistula is an abnormal channel between the coronary artery and the heart chambers or other vessels, causing abnormal shunting of blood within the coronary artery through the fistula opening. These two types of lesions are often closely related. For example, some coronary aneurysms, due to weak local vessel wall structures, may secondary to the formation of fistulas, and the shunting at the fistula opening can further affect the blood supply to the coronary aneurysm and distal vessels, even leading to increased cardiac workload. Therefore, accurately identifying these two lesions and clarifying their relationship is a prerequisite for developing effective interventional treatment plans.

[0004] However, current technologies for diagnosis and treatment using coronary angiography images have the following problems: most technologies can only identify coronary aneurysms or coronary fistulas individually, and cannot simultaneously consider the relationship between the two; even if a few technologies attempt to identify both lesions simultaneously, they cannot accurately capture their spatial relationship within the coronary vascular network (such as the distance between the aneurysm and the fistula, and whether the fistula originates from the aneurysm) and pathophysiological connections (such as the impact of fistula shunt on the hemodynamics of the aneurysm). This is especially true for complex cases of secondary fistula formation caused by coronary aneurysms, where current technologies struggle to accurately detect the correlation between the two. Furthermore, current technologies cannot accurately distinguish between intra-aneurysmal thrombotic areas and active fistula locations by tracking the flow patterns of the contrast agent at different time phases. Intra-aneurysmal thrombotic areas have low and gradual grayscale values ​​due to slow contrast agent filling, while active fistula areas have large and rapidly decreasing grayscale values ​​due to rapid contrast agent flow. These differences in image characteristics require precise capture for differentiation, but current technologies lack effective analytical methods. This limitation directly prevents doctors from accurately calculating the impact of coronary aneurysms on distal blood supply, and from precisely assessing the shunt effect of coronary fistulas. Consequently, they are unable to comprehensively judge the combined impact of the two lesions on coronary blood flow, resulting in a lack of precise basis for the formulation of interventional treatment plans. This not only reduces the diagnostic accuracy of complex coronary aneurysm-fistula lesions, but also affects the success rate of interventional treatment, and increases the risk of postoperative recurrence and the incidence of complications. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent analysis system for the correlation between coronary angiography images and fistulas. This system solves the problems of existing technologies, such as the inability to simultaneously identify coronary aneurysms and fistulas, the difficulty in accurately capturing their spatial location and pathophysiological correlation, the inability to distinguish between intra-aneurysmal thrombi and active fistulas, and the inability to assess the combined impact of the two on coronary blood flow.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a coronary angiography image aneurysm-fistula correlation intelligent analysis system, comprising:

[0007] The image acquisition module is used to acquire coronary angiography image sequences containing different time phases;

[0008] The lesion identification and correlation analysis module is used to perform deep processing on coronary angiography image sequences, construct a coronary artery structure analysis model based on graph neural networks, automatically identify coronary aneurysms and coronary fistulas, and capture the spatial positional relationship between aneurysms and fistulas in the coronary vascular network. The spatial positional relationship includes the straight-line distance between the aneurysm and the fistula, the three-dimensional path distance, and their relative branch points on the vascular tree. It also analyzes their pathophysiological connections, specifically including the local anatomical correlation between the morphological characteristics of the aneurysm and the fistula, whether the fistula originates from the aneurysm, and the impact of fistula shunt on the hemodynamics of the aneurysm and distal vessels.

[0009] The time-series angiography frame analysis module is used to perform dynamic time-series analysis on coronary angiography image sequences, tracking the flow pattern of contrast agent in different phases, including filling rate, peak concentration, and clearance rate, thereby accurately distinguishing intratumoral thrombus areas and active fistula locations. Thrombus areas are characterized by slow contrast agent filling, low grayscale values, and gradual changes in image features, while active fistula locations are characterized by rapid contrast agent flow, large grayscale value changes, and rapid decreases in image features.

[0010] The hemodynamic quantitative assessment module is used to calculate the resistance, volume, and time effects of coronary aneurysm on distal vessel blood supply based on the lesion correlation information provided by the lesion identification and correlation analysis module and the regional blood flow characteristics distinguished by the time series angiography frame analysis module. It also assesses the blood shunting effect of coronary fistula and the degree of increase in cardiac load, so as to comprehensively judge the combined effect of coronary aneurysm fistula on the overall coronary blood flow and its severity.

[0011] The treatment planning assistance module is used to intelligently generate optimal interventional treatment path suggestions based on the lesion correlation analysis results provided by the lesion identification and correlation analysis module and the composite impact assessment results provided by the hemodynamic quantitative assessment module. The suggestions include the priority order of fistula closure and aneurysm embolization, the selection of interventional techniques, and the expected treatment effect.

[0012] Furthermore, after acquiring the coronary angiography image sequence, the image acquisition module performs preprocessing operations such as denoising, contrast enhancement, and vessel edge enhancement.

[0013] Furthermore, the process by which the lesion identification and correlation analysis module constructs a coronary artery structure analysis model based on a graph neural network includes:

[0014] Extract multi-scale, multi-modal vascular structural features and lesion features from the preprocessed images;

[0015] The vascular structural features and lesion features are mapped onto a graph structure, where coronary aneurysms and coronary fistulas are represented as specific nodes in the graph, and the vascular connection relationships, branching relationships, and spatial relationships between aneurysms and fistulas are represented as edges in the graph;

[0016] By utilizing graph convolutional networks or graph attention network models, training and inference are performed based on graph structures to identify and quantify the correlation of tumor fistula lesions.

[0017] Furthermore, the process by which the time-series contrast frame analysis module tracks the flow pattern of the contrast agent at different time phases to distinguish between intratumoral thrombus areas and active fistula locations includes:

[0018] Obtain the dynamic curve of the gray value of each pixel in the coronary angiography image sequence as a function of time;

[0019] Perform feature analysis on the dynamic curve, including the curve's rising slope, peak arrival time, peak intensity, and falling slope;

[0020] Based on the feature analysis results, the areas where the contrast agent fills slowly, the gray value is low, and the change is gradual were identified as intratumoral thrombosis areas.

[0021] Areas where contrast agent flows rapidly and grayscale values ​​change significantly and decrease rapidly are identified as active fistula locations.

[0022] Furthermore, the process by which the hemodynamic quantitative assessment module calculates the parameters of the impact of coronary aneurysm on distal vessel blood supply and the shunt effect parameters of coronary fistula includes:

[0023] Information on tumor expansion degree, tumor volume, morphological characteristics and location is obtained from the lesion identification and correlation analysis module;

[0024] The time-series angiography frame analysis module was used to obtain intratumoral blood flow velocity, tumor perfusion time, and fistula blood flow velocity and shunt volume.

[0025] Based on fluid dynamics models and hemodynamic principles, this study uses coronary aneurysm parameters and fistula parameters to calculate the effects of coronary aneurysms on distal vessel blood flow resistance, blood flow velocity, and perfusion pressure, as well as the abnormal blood shunt ratio and the degree of load on the cardiac chambers caused by coronary fistulas.

[0026] Furthermore, the process by which the hemodynamic quantitative assessment module comprehensively judges the combined impact and severity of coronary artery aneurysm fistula on overall coronary blood flow includes:

[0027] The parameters affecting blood supply and shunt effect of the fistula were weighted and fused.

[0028] Based on pre-defined clinical assessment criteria, the weighted fusion results are mapped to the severity level of the composite effect.

[0029] Furthermore, the process by which the treatment planning assistance module generates interventional treatment pathway suggestions includes:

[0030] The results of the lesion correlation analysis were analyzed, including the anatomical relationship between coronary aneurysms and fistulas, thrombosis, and fistula activity.

[0031] The results of the composite impact assessment were analyzed, including the degree of distal blood supply impairment and the impact of shunt on cardiac load.

[0032] By combining predefined clinical guidelines and expert experience knowledge bases, we can customize interventional treatment pathway recommendations for specific patient cases, including the optimal timing, technology selection, device type, and intraoperative precautions for fistula closure and aneurysm embolization.

[0033] Furthermore, during training, the graph neural network uses a labeled coronary angiography image dataset, which includes the location, type, and correlation of coronary aneurysms and fistulas labeled by experts, in order to optimize the model's recognition accuracy.

[0034] Furthermore, the system also includes:

[0035] The user interaction and visualization module is used to receive the doctor's operation instructions and present the lesion identification and correlation analysis results, time series angiography frame analysis results, hemodynamic quantitative assessment results, and interventional treatment path suggestions in a visual manner through 3D reconstruction, multi-view synchronous display, or dynamic timeline playback.

[0036] Furthermore, the system is deployed on a high-performance computing server and equipped with a graphics processing unit to support the operation of deep learning models and dynamic time series analysis algorithms.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] This invention acquires and preprocesses multi-time-series coronary angiography images through an image acquisition module. Then, a lesion identification and correlation analysis module uses a graph neural network to simultaneously identify coronary aneurysms and coronary fistulas, accurately capturing their spatial relationship and pathophysiological connections. This addresses the limitations of existing technologies in simultaneously identifying aneurysms and fistulas and insufficient correlation analysis. A time-series angiography frame analysis module tracks contrast agent flow patterns, distinguishing between intra-aneurysmal thrombi and active fistulas, overcoming the shortcomings of existing technologies in accurately differentiating this region. A hemodynamic quantitative assessment module combines lesion correlation information and blood flow characteristics to calculate the combined impact and severity of aneurysms and fistulas on coronary blood flow, providing quantitative evidence for treatment. A treatment planning assistance module generates optimal interventional recommendations based on clinical guidelines. Combined with a user interaction and visualization module, the results are presented intuitively. High-performance deployment ensures operational efficiency, comprehensively improving the accuracy and success rate of coronary aneurysm and fistula diagnosis and treatment, and reducing the risk of postoperative recurrence and complications. Attached Figure Description

[0039] Figure 1 This is a system structure diagram of the present invention;

[0040] Figure 2 This is a flowchart of the lesion identification and correlation analysis process of the present invention;

[0041] Figure 3 This is a flowchart of the time-series contrast imaging analysis of the present invention;

[0042] Figure 4 This is a flowchart of the hemodynamic assessment process of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1-4 This invention provides an intelligent analysis system for the correlation between aneurysm and fistula in coronary angiography images, comprising:

[0045] The image acquisition module is used to acquire coronary angiography image sequences containing different time phases;

[0046] The lesion identification and correlation analysis module is used to perform deep processing on coronary angiography image sequences, construct a coronary artery structure analysis model based on graph neural networks, automatically identify coronary aneurysms and coronary fistulas, and capture the spatial positional relationship between aneurysms and fistulas in the coronary vascular network. The spatial positional relationship includes the straight-line distance between the aneurysm and the fistula, the three-dimensional path distance, and their relative branch points on the vascular tree. It also analyzes their pathophysiological connections, specifically including the local anatomical correlation between the morphological characteristics of the aneurysm and the fistula, whether the fistula originates from the aneurysm, and the impact of fistula shunt on the hemodynamics of the aneurysm and distal vessels.

[0047] The time-series angiography frame analysis module is used to perform dynamic time-series analysis on coronary angiography image sequences, tracking the flow pattern of contrast agent in different phases, including filling rate, peak concentration, and clearance rate, thereby accurately distinguishing intratumoral thrombus areas and active fistula locations. Thrombus areas are characterized by slow contrast agent filling, low grayscale values, and gradual changes in image features, while active fistula locations are characterized by rapid contrast agent flow, large grayscale value changes, and rapid decreases in image features.

[0048] The hemodynamic quantitative assessment module is used to calculate the resistance, volume, and time effects of coronary aneurysm on distal vessel blood supply based on the lesion correlation information provided by the lesion identification and correlation analysis module and the regional blood flow characteristics distinguished by the time series angiography frame analysis module. It also assesses the blood shunting effect of coronary fistula and the degree of increase in cardiac load, so as to comprehensively judge the combined effect of coronary aneurysm fistula on the overall coronary blood flow and its severity.

[0049] The treatment planning assistance module is used to intelligently generate optimal interventional treatment path suggestions based on the lesion correlation analysis results provided by the lesion identification and correlation analysis module and the composite impact assessment results provided by the hemodynamic quantitative assessment module. The suggestions include the priority order of fistula closure and aneurysm embolization, the selection of interventional techniques, and the expected treatment effect.

[0050] Specifically, in practical applications, the image acquisition module utilizes clinically common coronary angiography equipment. This equipment, after injecting contrast agent into the patient's coronary arteries, continuously acquires images at preset time intervals, obtaining angiographic image sequences covering different phases such as the arterial, capillary, and venous phases. This ensures that subsequent analysis covers the complete flow of contrast agent within the coronary arteries. The lesion identification and correlation analysis module first performs preliminary preprocessing on the acquired image sequences to reduce noise interference. Then, it constructs a coronary artery structure analysis model based on a graph neural network. The model first extracts multi-scale structural features of the coronary arteries, including changes in vessel diameter, branching direction, and the morphological contours of coronary aneurysms and the channel morphology of coronary fistulas. These features are then mapped into a graph structure, where coronary aneurysms and fistulas are treated as independent nodes. Node attributes include their location coordinates and size parameters, while the connectivity of the vessels, the straight-line distance between the aneurysm and the fistula, the three-dimensional path distance, and the relative branching points of the two on the vascular tree are used as edge attributes in the graph. Through training and inference using graph neural networks, the model can not only automatically identify coronary aneurysms and coronary fistulas, but also accurately capture their spatial relationship. Simultaneously, it deeply analyzes pathophysiological connections, such as determining whether the fistula originates from the aneurysm, whether the aneurysm's morphological characteristics are related to the local anatomical structure of the fistula opening, and the impact of fistula shunt on the blood flow state within the aneurysm and the blood supply to distal vessels. This effectively solves the problem that existing technologies cannot simultaneously identify aneurysms and fistulas and clarify their relationship. The time-series angiography frame analysis module extracts the grayscale values ​​of each pixel in the image sequence at different time phases and generates dynamic change curves. By analyzing the curve's fill rate (rising slope), peak concentration (peak intensity), and clearance rate (falling slope), it distinguishes between intra-aneurysmal thrombus areas and active fistula locations. In intra-aneurysmal thrombus areas, where contrast agent penetration is difficult, the curve shows a slow rise, low peak grayscale value, and gentle fall. In contrast, in active fistula areas, where contrast agent flows rapidly, the curve shows a steep rise, high peak grayscale value, and rapid fall. This process achieves accurate differentiation between the two regions, overcoming the shortcomings of existing technologies in this regard.

[0051] The quantitative hemodynamic assessment module combines the aforementioned lesion correlation information with regional blood flow characteristics, and performs quantitative calculations using fluid dynamics models and hemodynamic principles. When calculating the abnormal blood shunt ratio caused by coronary fistula, formula (1) is used, which directly reflects the proportion of shunt blood to total coronary blood flow:

[0052] (1)

[0053] in The abnormal shunt ratio of the coronary fistula is dimensionless and ranges from 0 to 1; Qfistula is the fistula shunt volume in mL / s, calculated by the time-series angiography frame analysis module based on the fistula cross-sectional area and blood flow velocity. The calculation formula is Qfistula = Afistula × vfistula, where Afistula is the fistula cross-sectional area in mm², which needs to be converted to a unit matching the flow rate. 1 mm² corresponds to 1 × 102 -6 m², 1 mL / s corresponds to 1 × 10 -6 m³ / s, converting fistula A to m² and multiplying it by fistula v (unit: m / s) yields fistula Q; Qtotal is the total coronary blood flow, in mL / s, obtained using clinically common methods for assessing coronary flow reserve.

[0054] When assessing the degree of increased cardiac load ΔL caused by coronary fistula, the effects of abnormal shunt ratio and fistula blood flow velocity need to be considered comprehensively. Since the two have different dimensions, the fistula blood flow velocity is first normalized using formula (2), and then substituted into formula (3) for calculation:

[0055] (2)

[0056] (3)

[0057] Wherein ΔL represents the degree of increase in cardiac load caused by coronary artery fistula, dimensionless and ranging from 0 to 1, with a larger value indicating a more significant increase in load; v_fistula represents the normalized fistula blood flow velocity, dimensionless; v_fistula represents the original fistula blood flow velocity, in m / s, calculated by the time-series angiography frame analysis module based on the slope of the contrast agent flow curve; v_min represents the minimum fistula blood flow velocity as statistically observed in clinical practice, in m / s; v_max represents the maximum fistula blood flow velocity as statistically observed in clinical practice, in m / s; k1 and k2 are weighting coefficients, dimensionless and k1+k2=1, determined through the analytic hierarchy process, i.e., multiple interventional cardiovascular experts were invited to score the degree of influence of Q_fistula and v_fistula on cardiac load, and the final weights were determined after consistency testing to ensure that the assessment results are consistent with clinical understanding.

[0058] Simultaneously, this module also calculates the resistance change R, volumetric effect V, and blood flow time delay caused by the coronary aneurysm on the blood supply to distal vessels. It comprehensively assesses the combined impact of R, V, Qf, and ΔL on overall coronary blood flow and its severity, providing quantitative evidence for subsequent treatment. The treatment planning assistance module, based on the above analysis results and combined with clinical diagnosis and treatment logic, intelligently generates optimal interventional treatment pathway suggestions. For example, when the fistula originates from the aneurysm, it is recommended to prioritize occlusion of the fistula opening to reduce the continuous impact of shunt on the aneurysm, followed by aneurysm embolization. It also recommends suitable interventional techniques and device types, and explains the expected treatment effects, such as improving blood supply to distal vessels and reducing cardiac load. Overall, it achieves intelligent processing from image analysis to treatment assistance, significantly improving the accuracy and efficiency of coronary aneurysm and fistula diagnosis and treatment.

[0059] In this embodiment, after acquiring the coronary angiography image sequence, the image acquisition module performs preprocessing operations such as denoising, contrast enhancement, and vessel edge enhancement.

[0060] Specifically, after acquiring the coronary angiography image sequence, the image acquisition module immediately initiates the preprocessing process. First, a Gaussian filtering algorithm, a standard technique, is used for denoising. This algorithm uses a Gaussian kernel function to weighted smooth the image pixels, effectively filtering out interference signals caused by equipment noise, patient physiological movements, and other factors. The specific implementation process is a conventional technique in this field and will not be elaborated upon here. Next, a histogram equalization algorithm, also a standard technique, is used to enhance image contrast. This algorithm adjusts the image's grayscale distribution, amplifying the grayscale difference between the coronary arteries and surrounding tissues, making the vessel contours more prominent. Its calculation logic is well-known in this field and requires no further explanation. Finally, the Canny edge detection algorithm, a standard technique, is used to enhance the vessel edges, accurately extracting the edge contour information of the vessels and reducing the impact of edge blurring on subsequent analysis. The specific steps of this algorithm are standard operations and will not be detailed here. After the above preprocessing operations, the image quality is significantly improved, providing high-quality image data support for subsequent modules, effectively reducing recognition errors caused by image quality issues in subsequent modules, and improving the reliability of the overall system analysis results.

[0061] In this embodiment, the process of constructing a coronary artery structure analysis model based on a graph neural network by the lesion identification and correlation analysis module includes:

[0062] Extract multi-scale, multi-modal vascular structural features and lesion features from the preprocessed images;

[0063] The vascular structural features and lesion features are mapped onto a graph structure, where coronary aneurysms and coronary fistulas are represented as specific nodes in the graph, and the vascular connection relationships, branching relationships, and spatial relationships between aneurysms and fistulas are represented as edges in the graph;

[0064] By utilizing graph convolutional networks or graph attention network models, training and inference are performed based on graph structures to identify and quantify the correlation of tumor fistula lesions.

[0065] Specifically, when constructing a coronary artery structure analysis model based on graph neural networks, the lesion identification and correlation analysis module first extracts multi-scale and multi-modal vascular structure features and lesion features from the preprocessed image. The feature extraction process employs multi-scale convolution operations as used in existing technologies, extracting local and global features of the image, such as changes in vessel diameter and the morphological contour of coronary aneurysms, through convolution kernels of different sizes. The specific convolution calculation process is a well-known technique in this field and will not be elaborated here. Subsequently, the extracted features are mapped onto a graph structure. In the graph structure, coronary aneurysms and coronary fistulas are defined as specific nodes, and node attributes include key information such as the location, size, and morphology of the lesion. The vascular connections, branch hierarchy, and spatial relationships between aneurysms and fistulas of the coronary arteries are defined as edges in the graph, and edge attributes include information such as connection strength and distance parameters. Subsequently, a graph convolutional network or graph attention network model was selected, and the constructed graph structure was used as input for model training and inference. During the training process, the backpropagation algorithm in the existing technology was used to optimize the model parameters. By continuously adjusting the parameters, the model's ability to identify aneurysms and fistulas and its ability to analyze correlations were improved. Ultimately, the accurate identification of coronary aneurysms and coronary fistulas was achieved, and the correlation between the two was quantified. This process effectively improved the accuracy of lesion identification and the depth of correlation analysis, providing more comprehensive lesion information for subsequent diagnosis and treatment.

[0066] In this embodiment, the process by which the time-series contrast frame analysis module tracks the flow pattern of the contrast agent at different time phases to distinguish between intratumoral thrombus areas and active fistula locations includes:

[0067] Obtain the dynamic curve of the gray value of each pixel in the coronary angiography image sequence as a function of time;

[0068] Perform feature analysis on the dynamic curve, including the curve's rising slope, peak arrival time, peak intensity, and falling slope;

[0069] Based on the feature analysis results, the areas where the contrast agent fills slowly, the gray value is low, and the change is gradual were identified as intratumoral thrombosis areas.

[0070] Areas where contrast agent flows rapidly and grayscale values ​​change significantly and decrease rapidly are identified as active fistula locations.

[0071] Specifically, when distinguishing between intratumoral thrombus areas and active fistula locations, the time-series angiography frame analysis module first extracts the grayscale value of each pixel in the coronary angiography image sequence frame by frame, arranging them in chronological order to form a dynamic curve of the grayscale value change for that pixel. Next, feature analysis is performed on each dynamic curve, calculating the rising slope to reflect the contrast agent filling rate, peak arrival time, peak intensity to reflect the peak contrast agent concentration, and the falling slope to reflect the contrast agent clearance rate. These features are calculated using existing linear fitting and extreme value detection methods, such as calculating the rising slope through linear regression and determining the peak arrival time by traversing the curve data points. The specific calculation process is standard practice in this field and will not be elaborated upon here. Based on these characteristic analyses, the regions corresponding to curves with a small upward slope, late peak arrival time, low peak intensity, and a small downward slope are identified as intratumoral thrombus areas. These areas exhibit slow contrast agent filling, low grayscale values, and gradual changes due to the thrombus obstructing contrast agent flow. Conversely, the regions corresponding to curves with a large upward slope, early peak arrival time, high peak intensity, and a large downward slope are identified as active fistula locations. These areas exhibit rapid contrast agent flow, large grayscale value changes, and rapid declines due to the rapid passage of contrast agent through the fistula. This method allows for precise differentiation between the two regions, providing accurate regional location information for subsequent hemodynamic assessment and treatment planning, and reducing diagnostic and treatment errors caused by regional misjudgment.

[0072] In this embodiment, the process by which the hemodynamic quantitative assessment module calculates the parameters of the impact of coronary aneurysm on distal vessel blood supply and the shunt effect parameters of coronary fistula includes:

[0073] Information on tumor expansion degree, tumor volume, morphological characteristics and location is obtained from the lesion identification and correlation analysis module;

[0074] The time-series angiography frame analysis module was used to obtain intratumoral blood flow velocity, tumor perfusion time, and fistula blood flow velocity and shunt volume.

[0075] Based on fluid dynamics models and hemodynamic principles, this study uses coronary aneurysm parameters and fistula parameters to calculate the effects of coronary aneurysms on distal vessel blood flow resistance, blood flow velocity, and perfusion pressure, as well as the abnormal blood shunt ratio and the degree of load on the cardiac chambers caused by coronary fistulas.

[0076] Specifically, when calculating relevant parameters, the hemodynamic quantitative assessment module first obtains the following information from the lesion identification and correlation analysis module: the degree of coronary aneurysm expansion (i.e., the ratio of the maximum diameter of the aneurysm to the diameter of the adjacent normal vessel); the aneurysm volume calculated using three-dimensional reconstruction technology; the aneurysm morphological characteristics, such as whether it is regular and whether there are thin-walled areas; and the specific location of the aneurysm in the coronary artery. Simultaneously, it obtains the following information from the time-series angiography frame analysis module: the intra-aneurysmal blood flow velocity calculated based on the contrast agent filling time within the aneurysm and the aneurysm size; the aneurysm perfusion time (the time from the start of contrast agent entry into the aneurysm to reaching peak grayscale value); and the fistula blood flow velocity and fistula shunt volume.

[0077] Subsequently, based on fluid dynamics models such as Poiseuille's law, which are existing technologies and hemodynamic principles, the effects of coronary aneurysms on distal vessels were calculated using the acquired parameters: the changes in blood flow resistance R, blood flow velocity V, and perfusion pressure fluctuations caused by coronary aneurysms were calculated using existing formulas; the abnormal blood shunt ratio Qf caused by coronary fistula was calculated using formula (1); after normalizing the v fistula using formula (2), the increased load ΔL on the cardiac chambers caused by coronary fistula was calculated by substituting it into formula (3). Through the calculation of these parameters, the effects of coronary aneurysms and coronary fistulas on blood flow were comprehensively quantified, providing quantitative data support for a comprehensive judgment of their combined effects, and improving the scientificity and accuracy of the assessment results.

[0078] In this embodiment, the process by which the hemodynamic quantitative assessment module comprehensively judges the combined impact and severity of coronary artery aneurysm fistula on overall coronary blood flow includes:

[0079] The parameters affecting blood supply and shunt effect of the fistula were weighted and fused.

[0080] Based on pre-defined clinical assessment criteria, the weighted fusion results are mapped to the severity level of the composite effect.

[0081] Specifically, when comprehensively judging the combined effects and their severity, the hemodynamic quantitative assessment module first determines the parameters to be included in the assessment based on clinical experience and medical research conclusions: the normalized coronary aneurysm's influence on distal vessel blood flow resistance R, the normalized coronary aneurysm's influence on distal vessel blood flow velocity V, the abnormal blood shunt ratio Qf calculated by formula (1), and the degree of increase in cardiac chamber load ΔL calculated by formula (3). Since the original dimensions of each parameter are different, R and V are both processed into dimensionless parameters by the normalization method of formula (2) to ensure that the dimensions of subsequent calculations are consistent.

[0082] Next, the above parameters are weighted and fused, and the comprehensive evaluation value S is calculated using formula (4):

[0083] (4)

[0084] Where S is the comprehensive assessment value of the combined impact of coronary aneurysm fistula on the overall coronary blood flow, which is dimensionless and ranges from 0 to 1; w1, w2, w3, and w4 are the weight coefficients of each parameter, which are dimensionless and w1+w2+w3+w4=1. They are determined by the Delphi method, that is, the weights are initially set by the expert team, and then adjusted through multiple rounds of feedback until the experts' opinions tend to be consistent. The final weights can reflect the degree of contribution of each parameter to the combined impact. For example, the weights of w3 and w4 are usually higher than those of w1 and w2, because the shunt effect and cardiac load have a more significant impact on the overall blood flow.

[0085] Subsequently, by comparing with pre-defined clinical assessment criteria, S is mapped to a severity level: S < 0.3 indicates mild impact, 0.3 ≤ S < 0.6 indicates moderate impact, and S ≥ 0.6 indicates severe impact. This process clearly defines the severity of the combined impact of coronary artery aneurysm fistula on overall coronary blood flow, providing physicians with a clear basis for developing treatment strategies and ensuring the targeted and appropriate nature of treatment plans.

[0086] In this embodiment, the process by which the treatment planning assistance module generates interventional treatment path suggestions includes:

[0087] The results of the lesion correlation analysis were analyzed, including the anatomical relationship between coronary aneurysms and fistulas, thrombosis, and fistula activity.

[0088] The results of the composite impact assessment were analyzed, including the degree of distal blood supply impairment and the impact of shunt on cardiac load.

[0089] By combining predefined clinical guidelines and expert experience knowledge bases, we can customize interventional treatment pathway recommendations for specific patient cases, including the optimal timing, technology selection, device type, and intraoperative precautions for fistula closure and aneurysm embolization.

[0090] Specifically, when generating interventional treatment pathway suggestions, the treatment planning assistance module first analyzes the lesion correlation analysis results output by the lesion identification and correlation analysis module to clarify the anatomical relationship between the coronary aneurysm and the fistula, such as whether the fistula originates from the aneurysm or whether the two exist independently. It also determines whether there is thrombus in the aneurysm and the distribution range of the thrombus, and at the same time determines the activity status of the fistula, such as whether the shunt is active. Next, it analyzes the composite impact assessment results obtained by the hemodynamic quantitative assessment module, and combines the comprehensive assessment value S calculated by formula (4) and its corresponding severity level to understand the degree of distal vessel blood supply impairment, such as whether it is only slightly decreased or severely insufficient, and the degree of impact of coronary fistula shunt on cardiac load, such as whether it is only slightly increased or significantly increased.

[0091] Building upon this foundation, the system invokes predefined clinical guidelines built into its system. These guidelines cover current mainstream interventional treatment standards for coronary artery aneurysms and fistulas. Simultaneously, it incorporates extensive clinical experience from a large database of similar cases stored in an expert knowledge base. The invocation process employs a rule-matching algorithm, a technology already in use, to match the patient's lesion parameters, such as aneurysm size, fistula location, and S-value, with cases in the knowledge base, selecting suitable treatment plans. The specific matching logic is standard practice in this field and will not be elaborated upon here. Ultimately, a personalized interventional treatment pathway recommendation is tailored to the specific patient's case. This includes the optimal sequence of fistula closure and aneurysm embolization, the selection of appropriate interventional techniques such as the type of occluder or embolic material, the appropriate device model for the patient's condition, and intraoperative precautions such as avoiding damage to the vessel wall or preventing embolic agent migration. This customized recommendation fully aligns with the patient's specific condition, improving the safety and effectiveness of interventional treatment and reducing the risk of postoperative complications.

[0092] In this embodiment, the graph neural network is trained using a labeled coronary angiography image dataset, which contains the location, type, and correlation of coronary aneurysms and fistulas labeled by experts, in order to optimize the model's recognition accuracy.

[0093] Specifically, during the training of the graph neural network, a large-scale coronary angiography image dataset was first selected. This dataset covers angiography images of patients with coronary aneurysms and fistulas of different ages and disease severity, ensuring that the dataset has good representativeness and diversity. Each image was jointly annotated by multiple senior cardiovascular specialists. The annotation content included the location and extent of the coronary aneurysm, its type (e.g., cystic aneurysm, fusiform aneurysm), the location and extent of the coronary fistula, its type (e.g., coronary-left ventricular fistula, coronary-pulmonary fistula), and the relationship between the two, such as whether there is an origin correlation or a blood flow influence correlation, ensuring the accuracy and reliability of the annotation results.

[0094] The labeled dataset was divided into training, validation, and test sets using a random partitioning method common in existing technologies. The partitioning ratios were set according to standard clinical data processing ratios, with adjustments made based on the dataset size – a routine practice in this field. During training, the image data and labeled information from the training set were input into the graph neural network. The gradient descent algorithm, a standard technique, was used to optimize the model's weight and bias parameters. Iterative training continued until the model's recognition accuracy and correlation judgment accuracy on the validation set reached preset requirements. The specific optimization process is well-known in this field and will not be elaborated upon here. The graph neural network model trained in this manner can more accurately identify coronary aneurysms and coronary fistulas, and accurately analyze their correlation, effectively reducing missed diagnoses and misdiagnoses, thus improving the model's reliability and practicality in clinical applications.

[0095] In this embodiment, the system further includes:

[0096] The user interaction and visualization module is used to receive the doctor's operation instructions and present the lesion identification and correlation analysis results, time series angiography frame analysis results, hemodynamic quantitative assessment results, and interventional treatment path suggestions in a visual manner through 3D reconstruction, multi-view synchronous display, or dynamic timeline playback.

[0097] Specifically, the system's user interaction and visualization module features a simple and intuitive interface, allowing doctors to input various commands such as zooming in and out of specific image areas, switching between different result display modes, and retrieving historical analysis data. When presenting analysis results, this module employs multiple visualization methods: the 3D reconstruction function uses existing volume rendering algorithms to reconstruct 2D angiography image sequences into 3D vascular models, intuitively displaying the spatial structure of coronary arteries, coronary aneurysms, and coronary fistulas. The specific reconstruction process is a standard technique in this field and will not be elaborated upon here; the multi-view synchronous display function can simultaneously display angiography images of the same patient from different projection angles and corresponding analysis results, such as simultaneous display of anteroposterior and lateral images, helping doctors understand the lesion from multiple perspectives; the dynamic timeline playback function synchronously replays the contrast agent flow process along with lesion identification results and hemodynamic assessment results such as Qf, ΔL, and S values, clearly marking the time points of occurrence of intra-aneurysmal thrombus areas and active fistula locations, as well as changes in blood flow parameters, on the timeline. These visualization methods make complex analysis results easier to understand, enabling doctors to quickly and accurately grasp the patient's condition and assessment results, significantly improving the efficiency of diagnosis and treatment.

[0098] In this embodiment, the system is deployed on a high-performance computing server and is equipped with a graphics processing unit to support the efficient operation of deep learning models and dynamic time series analysis algorithms.

[0099] Specifically, the entire intelligent analysis system for coronary angiography image aneurysm-fistula correlation is deployed on a high-performance computing server. This server is equipped with a high-performance graphics processing unit (GPU) with powerful parallel computing capabilities. The algorithm acceleration process adopts existing parallel computing architectures, distributing large-scale data processing tasks to multiple computing cores for parallel execution. When running deep learning models, the GPU can accelerate image feature extraction, model inference, and other computational processes, significantly reducing the time required for the model to process large-scale angiography image data. When executing dynamic time series analysis algorithms, the GPU can process a large number of pixel grayscale value time series data in parallel, quickly generating dynamic change curves for each pixel and completing feature analysis. When calculating hemodynamic parameters, the GPU can simultaneously perform batch calculations of parameters such as Qf, ΔL, and S for multiple groups of patients, improving computational efficiency. Through the collaborative work of high-performance computing servers and graphics processing units, the system can efficiently process large amounts of coronary angiography image data generated in clinical practice, and quickly output results such as lesion identification, correlation analysis, hemodynamic assessment, and treatment planning suggestions. This meets the real-time requirements of clinical diagnosis and treatment, avoids doctors prolonging treatment time due to waiting for analysis results, and enhances the practical application value of the system in clinical scenarios.

[0100] In summary, this invention acquires and preprocesses multi-time-series coronary angiography images through an image acquisition module. Then, a lesion identification and correlation analysis module uses a graph neural network to simultaneously identify coronary aneurysms and fistulas, accurately capturing their spatial relationship and pathophysiological connections. This addresses the limitations of existing technologies in simultaneously identifying aneurysms and fistulas and insufficient correlation analysis. A time-series angiography frame analysis module tracks contrast agent flow patterns, distinguishing between intra-aneurysmal thrombi and active fistulas, overcoming the shortcomings of existing technologies in accurately differentiating this region. A hemodynamic quantitative assessment module combines lesion correlation information and blood flow characteristics to calculate the combined impact and severity of aneurysms and fistulas on coronary blood flow, providing quantitative evidence for treatment. A treatment planning assistance module generates optimal interventional recommendations based on clinical guidelines. Combined with a user interaction and visualization module, the results are presented intuitively. High-performance deployment ensures operational efficiency, comprehensively improving the accuracy and success rate of coronary aneurysm and fistula diagnosis and treatment, and reducing the risk of postoperative recurrence and complications.

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A coronary angiography image aneurysm-fistula correlation intelligent analysis system, characterized in that, include: The image acquisition module is used to acquire coronary angiography image sequences containing different time phases; The lesion identification and correlation analysis module is used to perform deep processing on coronary angiography image sequences, construct a coronary artery structure analysis model based on graph neural networks, automatically identify coronary aneurysms and coronary fistulas, and capture the spatial positional relationship between aneurysms and fistulas in the coronary vascular network. The spatial positional relationship includes the straight-line distance between the aneurysm and the fistula, the three-dimensional path distance, and their relative branch points on the vascular tree. It also analyzes their pathophysiological connections, specifically including the local anatomical correlation between the morphological characteristics of the aneurysm and the fistula, whether the fistula originates from the aneurysm, and the impact of fistula shunt on the hemodynamics of the aneurysm and distal vessels. The time-series angiography frame analysis module is used to perform dynamic time-series analysis on coronary angiography image sequences, tracking the flow pattern of contrast agent in different phases, including filling rate, peak concentration, and clearance rate, thereby accurately distinguishing intratumoral thrombus areas and active fistula locations. Thrombus areas are characterized by slow contrast agent filling, low grayscale values, and gradual changes in image features, while active fistula locations are characterized by rapid contrast agent flow, large grayscale value changes, and rapid decreases in image features. The hemodynamic quantitative assessment module is used to calculate the resistance, volume, and time effects of coronary aneurysm on distal vessel blood supply based on the lesion correlation information provided by the lesion identification and correlation analysis module and the regional blood flow characteristics distinguished by the time series angiography frame analysis module. It also assesses the blood shunting effect of coronary fistula and the degree of increase in cardiac load, so as to comprehensively judge the combined effect of coronary aneurysm fistula on the overall coronary blood flow and its severity. The treatment planning assistance module is used to intelligently generate optimal interventional treatment path suggestions based on the lesion correlation analysis results provided by the lesion identification and correlation analysis module and the composite impact assessment results provided by the hemodynamic quantitative assessment module. The suggestions include the priority order of fistula closure and aneurysm embolization, the selection of interventional techniques, and the expected treatment effect.

2. The intelligent analysis system for coronary angiography image aneurysm-fistula correlation according to claim 1, characterized in that, After acquiring the coronary angiography image sequence, the image acquisition module performs preprocessing operations such as denoising, contrast enhancement, and vessel edge enhancement.

3. The intelligent analysis system for coronary angiography image aneurysm-fistula correlation according to claim 1, characterized in that, The process of constructing a coronary artery structure analysis model based on a graph neural network in the lesion identification and correlation analysis module includes: Extract multi-scale, multi-modal vascular structural features and lesion features from the preprocessed images; The vascular structural features and lesion features are mapped onto a graph structure, where coronary aneurysms and coronary fistulas are represented as specific nodes in the graph, and the vascular connection relationships, branching relationships, and spatial relationships between aneurysms and fistulas are represented as edges in the graph; By utilizing graph convolutional networks or graph attention network models, training and inference are performed based on graph structures to identify and quantify the correlation of tumor fistula lesions.

4. The intelligent analysis system for coronary angiography image aneurysm-fistula correlation according to claim 1, characterized in that, The time-series contrast frame analysis module tracks the flow patterns of the contrast agent at different time phases to distinguish between intratumoral thrombus areas and active fistula locations. Obtain the dynamic curve of the gray value of each pixel in the coronary angiography image sequence as a function of time; Perform feature analysis on the dynamic curve, including the curve's rising slope, peak arrival time, peak intensity, and falling slope; Based on the feature analysis results, the areas where the contrast agent fills slowly, the gray value is low, and the change is gradual were identified as intratumoral thrombosis areas. Areas where contrast agent flows rapidly and grayscale values ​​change significantly and decrease rapidly are identified as active fistula locations.

5. The intelligent analysis system for coronary angiography image aneurysm-fistula correlation according to claim 1, characterized in that, The process by which the hemodynamic quantitative assessment module calculates parameters related to the impact of coronary aneurysm on distal vessel blood supply and parameters related to the shunt effect of coronary fistula includes: Information on tumor expansion degree, tumor volume, morphological characteristics and location is obtained from the lesion identification and correlation analysis module; The time-series angiography frame analysis module was used to obtain intratumoral blood flow velocity, tumor perfusion time, and fistula blood flow velocity and shunt volume. Based on fluid dynamics models and hemodynamic principles, this study uses coronary aneurysm parameters and fistula parameters to calculate the effects of coronary aneurysms on distal vessel blood flow resistance, blood flow velocity, and perfusion pressure, as well as the abnormal blood shunt ratio and the degree of load on the cardiac chambers caused by coronary fistulas.

6. The intelligent analysis system for coronary angiography image aneurysm-fistula correlation according to claim 1, characterized in that, The process by which the quantitative hemodynamic assessment module comprehensively judges the combined impact and severity of coronary artery aneurysm fistula on overall coronary blood flow includes: The parameters affecting blood supply and shunt effect of the fistula were weighted and fused. Based on pre-defined clinical assessment criteria, the weighted fusion results are mapped to the severity level of the composite effect.

7. The intelligent analysis system for coronary angiography image aneurysm-fistula correlation according to claim 1, characterized in that, The process by which the treatment planning assistance module generates interventional treatment pathway suggestions includes: The results of the lesion correlation analysis were analyzed, including the anatomical relationship between coronary aneurysms and fistulas, thrombosis, and fistula activity. The results of the composite impact assessment were analyzed, including the degree of distal blood supply impairment and the impact of shunt on cardiac load. By combining predefined clinical guidelines and expert experience knowledge bases, we can customize interventional treatment pathway recommendations for specific patient cases, including the optimal timing, technology selection, device type, and intraoperative precautions for fistula closure and aneurysm embolization.

8. The intelligent analysis system for coronary angiography image aneurysm-fistula correlation according to claim 3, characterized in that, During training, the graph neural network uses a labeled coronary angiography image dataset, which includes the location, type, and correlation of coronary aneurysms and fistulas labeled by experts, in order to optimize the model's recognition accuracy.

9. The intelligent analysis system for coronary angiography image aneurysm-fistula correlation according to claim 1, characterized in that, The system also includes: The user interaction and visualization module is used to receive the doctor's operation instructions and present the lesion identification and correlation analysis results, time series angiography frame analysis results, hemodynamic quantitative assessment results, and interventional treatment path suggestions in a visual manner through 3D reconstruction, multi-view synchronous display, or dynamic timeline playback.

10. The intelligent analysis system for coronary angiography image aneurysm-fistula correlation according to claim 1, characterized in that, The system is deployed on a high-performance computing server and is equipped with a graphics processing unit to support the operation of deep learning models and dynamic time series analysis algorithms.

Images