Image analysis-based cerebral angiography thrombectomy data prediction method and system

Abstract

This invention discloses a method and system for predicting thrombectomy data in cerebral angiography based on image analysis, belonging to the field of medical image processing technology. The method acquires a continuous temporal sequence of angiographic images and event time window parameters; obtains first and second local regions of interest (ROIs) through vascular contour extraction and pixel temporal tracking; performs motion field analysis and gray-scale temporal integration on the first ROI to obtain fluid shear force features and asymmetric retention features; extracts gray-scale spatial decay gradient features by spatial differentiation on the second ROI; topologically fuses the shear force and retention features to obtain a bottom-level spatial constraint mask; uses this mask as a reference to weighted sample the decay gradient features to obtain an internal compactness index; and generates predicted probability data based on time window parameter mapping. This invention effectively eliminates image artifact interference caused by fluctuations in the complex physical environment, achieving accurate decoupling and objective quantitative prediction of the underlying spatial attachment morphology in cerebral angiography.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and more specifically, this application relates to a method and system for predicting cerebral angiography thrombectomy data based on image analysis. Background Technology

[0002] Cerebral angiography and thrombectomy-assisted analysis systems play a crucial role in the preoperative data assessment and prediction stage of cerebrovascular diseases. However, traditional two-dimensional cerebral angiography image processing techniques are generally based on static visual feature recognition, relying solely on single-dimensional pixel grayscale defects for coarse localization of lesion areas, which suffers from serious underlying technical deficiencies.

[0003] In specific preoperative angiography analysis scenarios, the hydrodynamic properties of contrast agents are inevitably significantly affected by fluctuations in the physical environment of the patient's surroundings. Existing prediction systems, when extracting core data features, typically assume the system is in a standard and constant ideal physical state, completely severing the coupling between hemorheological compensatory fluctuations caused by environmental factors and contrast agent emptying delays. They also fail to incorporate the objective deformation of the vessel wall caused by environmental temperature stress into the underlying computation. This processing logic, which forcibly separates the static image matrix from the dynamic physical environment, makes the system highly susceptible to misclassifying non-pathological contrast agent diffusion delays or physiological narrowing of the vessel wall as severe organic occlusion.

[0004] In summary, existing technologies cannot effectively filter out systematic data biases caused by environmental parameters in the context of preoperative image analysis for angiography and thrombectomy in complex physical environments, and it is difficult to non-invasively and accurately decouple the predictive probability of complex lesions. Summary of the Invention

[0005] To address the aforementioned technical problems, this paper provides a method and system for predicting cerebral angiography thrombectomy data based on image analysis. This technical solution solves the problems mentioned in the background section.

[0006] In a first aspect, embodiments of this application provide a method for predicting thrombectomy data based on image analysis in cerebral angiography, comprising the following steps: acquiring a continuous temporal angiography image sequence and event time window parameters of the target vascular region; performing vascular contour extraction and pixel temporal tracking on the continuous temporal angiography image sequence to obtain a first local region of interest (ROI) and a second ROI; performing inter-frame pixel motion field analysis and pixel gray-level time integral calculation on the first ROI to obtain fluid shear force features and asymmetric retention features; performing spatial differentiation of the pixel gray-level matrix on the second ROI to extract gray-level spatial decay gradient features; performing spatial topological fusion of the fluid shear force features and asymmetric retention features to obtain a bottom-level spatial constraint mask of the target vascular region; using the bottom-level spatial constraint mask as a spatial constraint benchmark, performing weighted feature sampling on the gray-level spatial decay gradient features to obtain an internal density index; performing nonlinear time-weighted mapping on the internal density index according to the event time window parameters to generate and output predicted probability data characterizing the composite physical state of the blockage within the target vascular region.

[0007] Secondly, embodiments of this application provide a cerebral angiography thrombectomy data prediction system based on image analysis, comprising: a data acquisition module for acquiring a continuous temporal angiography image sequence and event time window parameters of a target vascular region; a region extraction module for performing vascular contour extraction and pixel temporal tracking on the continuous temporal angiography image sequence to obtain a first local region of interest and a second local region of interest; an image feature extraction module for performing inter-frame pixel motion field analysis and pixel gray-level time integral calculation on the first local region of interest to obtain fluid shear force features and asymmetric retention features; and a gray-level attenuation extraction module for performing inter-frame pixel motion field analysis and pixel gray-level time integral calculation on the second local region of interest. The system performs spatial differentiation of the pixel grayscale matrix in the region of interest to extract grayscale attenuation gradient features. A constraint mask processing module is used to perform spatial topological fusion of fluid shear force features and asymmetric retention features to obtain the underlying spatial constraint mask for the target vascular region. A density index processing module uses the underlying spatial constraint mask as a spatial constraint benchmark to perform weighted feature sampling of the grayscale attenuation gradient features to obtain an internal density index. A prediction output module performs nonlinear time-weighted mapping of the internal density index based on event time window parameters to generate and output predicted probability data characterizing the composite physical state of the blockage within the target vascular region.

[0008] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described image analysis-based method for predicting cerebral angiography thrombectomy data.

[0009] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0010] 1. This innovative approach acquires contrast agent filling and emptying sequences, and performs inter-frame pixel motion field analysis, pixel gray-scale time integral calculation, and spatial differentiation of the pixel gray-scale matrix for different sequence periods. This processing logic breaks through the limitations of traditional static single-image recognition, transforming the homogeneous pixel matrix into a multi-dimensional decoupled extraction of fluid shear force features, asymmetric retention features, and gray-scale spatial attenuation gradient features. This ensures the objectivity and comprehensiveness of the extraction of deep physical features of complex lesions from the source of data processing.

[0011] 2. This scheme constructs a bottom-layer spatial constraint mask for the target vascular region by spatially topologically fusing the extracted fluid shear force features with asymmetric retention features. This technical step not only achieves precise anchoring of fluid dynamics data of different dimensions on a two-dimensional spatial plane, but also provides a rigorous physical spatial benchmark for subsequent feature calculations, significantly reducing the contamination of core calculation results by invalid background pixels and restoring the true underlying attachment morphology data of the lesion.

[0012] 3. This scheme uses a bottom-level spatial constraint mask as the spatial constraint benchmark, performs weighted feature sampling on the gray-scale attenuation gradient features, and accurately calculates the internal density index characterizing the micropore permeability of the lesion. Then, combined with the event time window parameter, a nonlinear time-weighted mapping is performed on this internal density index. This closed-loop data processing flow deeply nests and couples the spatial dimension density features with the temporal dimension evolution parameters, transforming simple image gray-scale data across dimensions into predictive probability data characterizing the complex physical state of the obstruction, providing a solid data foundation for the reliability of the prediction results. Attached Figure Description

[0013] Figure 1 A schematic diagram illustrating the steps of the image analysis-based cerebral angiography thrombectomy data prediction method provided in this application embodiment;

[0014] Figure 2 A schematic diagram of the logic flow of the image analysis-based cerebral angiography thrombectomy data prediction method provided in the embodiments of this application;

[0015] Figure 3 This is a schematic diagram of the structure of the image analysis-based cerebral angiography thrombectomy data prediction system provided in the embodiments of this application. Detailed Implementation

[0016] This application's embodiments address the technical problems in the prior art where, in the context of complex physical environment fluctuations during preoperative image analysis for cerebral angiography thrombectomy, systematic data biases caused by environmental parameters cannot be effectively filtered out, and it is difficult to non-invasively and accurately decouple the predicted probability of complex lesions.

[0017] In the context of preoperative auxiliary analysis for cerebral angiography thrombectomy, the actual occlusive lesion is often a complex structure composed of lesions at the underlying vascular layer and occlusions at different stages of evolution. Furthermore, fluctuations in the patient's physical environment can trigger hemorheological compensation and delayed contrast agent emptying. Existing conventional static visual feature recognition processing logic forcibly separates the image matrix from the dynamic physical environment, making it difficult to non-invasively decouple the true three-dimensional morphology and microscopic physical properties of complex lesions. This easily leads to the misclassification of non-pathological contrast agent diffusion delays or physiological narrowing of the vessel wall as complete organic occlusion.

[0018] To address the underlying systemic errors caused by the complex environmental fluctuations and homogeneous image appearances mentioned above, this solution starts with independent decoupling across the entire spatiotemporal dimension, constructing an exploratory data processing underlying logic. Abandoning static recognition of single image frames, this solution first acquires a continuous temporal sequence of angiographic images of the target vascular region and event time window parameters, establishing a spatiotemporal data foundation for capturing dynamic fluid dynamics changes. To accurately isolate the physical mapping differences of the contrast agent at different flow stages, this solution performs vascular contour extraction and pixel temporal tracking on the continuous temporal sequence of angiographic images, respectively locating the first and second regions of interest.

[0019] After identifying the target region, this scheme performs pixel-level feature extraction based on the physical characteristics of different regions. For the first local region of interest (MOI), inter-frame pixel motion field analysis and pixel gray-level time integration calculation are performed, thereby transforming the pure image pixel displacement and accumulation patterns into fluid shear force features and asymmetric retention features characterizing the hydrodynamic state. For the second MOI, gray-level attenuation gradient features reflecting the obstruction of contrast agent penetration are extracted by performing spatial differentiation of the pixel gray-level matrix. These multi-dimensional feature data provide objective quantitative evidence for revealing the microscopic properties of lesions.

[0020] To eliminate artifact interference from single features in complex backgrounds, this scheme performs spatial topological fusion of fluid shear force features and asymmetric retention features to reconstruct the true physical foundation of the underlying lesions in the target vascular region, generating a bottom-level spatial constraint mask. This bottom-level spatial constraint mask can effectively filter out invalid background pixels, providing a rigorous physical spatial boundary constraint for subsequent density feature extraction.

[0021] After obtaining a high-confidence spatial benchmark, the underlying spatial constraint mask is used as the spatial constraint benchmark. Weighted feature sampling of the gray-scale spatial attenuation gradient features is then performed to accurately extract internal density indicators characterizing the micropore permeability of the lesion. Finally, by combining event time window parameters reflecting the timeliness of lesion evolution, a nonlinear time-weighted mapping is applied to the internal density indicators, deeply coupling static spatial density features with dynamic time evolution parameters to generate and output predicted probability data characterizing the composite physical state of the blocker within the target vascular region.

[0022] The underlying data processing logic of this solution achieves a cross-dimensional transformation of image grayscale data into microscopic physical properties through the independent extraction and spatial topology nesting fusion of multidimensional spatiotemporal features. The technical effect of this logic is that it effectively eliminates image artifact interference caused by fluctuations in complex physical environments from the data level, completely solving the core technical problems of existing technologies' inability to penetrate two-dimensional homogeneous image appearances and the difficulty in accurately decoupling the underlying spatial attachment morphology and microscopic density properties of vascular occlusions. This provides highly objective data support for clinical auxiliary intervention and device matching.

[0023] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0024] Figure 1 This document illustrates the steps of a method for predicting cerebral angiography thrombectomy data based on image analysis, as provided in an embodiment of this application. The method includes the following steps: acquiring a continuous temporal angiography image sequence and event time window parameters for the target vascular region; extracting vascular contours and tracking pixels temporally from the continuous temporal angiography image sequence to obtain a first and second local region of interest (ROI); performing inter-frame pixel motion field analysis and pixel gray-level time integral calculation on the first ROI to obtain fluid shear force features and asymmetric retention features; performing spatial differentiation of the pixel gray-level matrix on the second ROI to extract gray-level spatial decay gradient features; performing spatial topological fusion of the fluid shear force features and asymmetric retention features to obtain a bottom-level spatial constraint mask for the target vascular region; using the bottom-level spatial constraint mask as a spatial constraint benchmark, performing weighted feature sampling on the gray-level spatial decay gradient features to obtain an internal density index; and performing nonlinear time-weighted mapping on the internal density index according to the event time window parameters to generate and output predicted probability data characterizing the composite physical state of the blockage within the target vascular region.

[0025] Figure 2This is a schematic diagram of the logical flow of the image analysis-based thrombectomy data prediction method for cerebral angiography provided in this application embodiment; this embodiment can be executed by a computer device deployed on a medical edge computing node or a cloud server. Based on the fluid / morphological extraction and multi-source environmental compensation of continuous image frames, and through deep nesting of spatial mask constraints and temporal time-effect mapping, this method completely solves the core technical defects of existing prediction systems that forcibly separate static image matrices from dynamic physical environments, leading to severe systemic artifacts caused by environmental parameters, and are unable to non-invasively decouple the three-dimensional morphology and microscopic attributes of complex lesions from homogeneous images. This solution achieves accurate and interference-resistant quantitative assessment of the spatial attachment morphology and microscopic density of blood vessels in complex environmental fluctuations during thrombectomy angiography, providing crucial underlying data support for subsequent clinical intervention decisions and high-risk device avoidance.

[0026] Furthermore, the specific acquisition process of the first and second local regions of interest is as follows: the continuous temporal contrast imaging image sequence is divided into a contrast agent filling period sequence and a contrast agent emptying period sequence; vascular contour extraction and pixel temporal tracking are performed on the contrast agent filling period sequence and the contrast agent emptying period sequence to calculate the time decay curve of the contrast agent flow rate; the pixel coordinates corresponding to the extreme points of flow rate decay in the time decay curve are determined as flow occlusion anchor points; spatial expansion is performed based on the flow occlusion anchor points to delineate the first local regions of interest representing the boundary of the vascular wall in the contrast agent emptying period sequence and the second local regions of interest representing the end of the occlusion device in the contrast agent filling period sequence.

[0027] In this embodiment, the blood vessel contour extraction adopts the Frangi blood vessel filtering algorithm based on the Hessian matrix. Its input is a two-dimensional angiographic image, and its output is an enhanced tubular blood vessel mask.

[0028] Pixel-time tracking employs optical flow, calculating the flow velocity by tracking the pixel displacement of the contrast agent wavefront.

[0029] The time decay curve is calculated by statistically analyzing the rate of change of the advance distance of the contrast agent pixel leading edge between consecutive frames over time.

[0030] The extreme point of flow velocity decay is obtained by taking the first derivative of the time decay curve and finding the inflection point where the derivative abruptly changes from a negative minimum to approach zero. In engineering implementation, the scale of spatial expansion is combined with prior knowledge of the diameter of cerebral blood vessels. Typically, it is centered on the occlusion anchor point and extends upstream along the vessel axis by 5-15 mm, with a typical value of 10 mm, as the second local region of interest. The expansion downstream and along the lateral walls to cover any possible eddy current areas is the first local region of interest.

[0031] This embodiment solves the technical problems of inaccurate lesion feature extraction range and overlapping filling and emptying features caused by the reliance on manual static selection in the existing technology by dynamically locking the extreme points based on the flow rate time decay curve and expanding the temporal separation. It achieves the technical effect of automatic and accurate positioning and targeted separation of the macroscopic spatial region of the multi-temporal imaging sequence. This distinction enables the system to automatically distinguish between the underlying foundation of the lesion and the surface attachments.

[0032] Furthermore, the specific process for acquiring fluid shear force characteristics is as follows: acquire current altitude data; acquire preset absolute viscosity benchmark values ​​for the contrast agent, preset first fitting constants, preset reference perfusion slope benchmark values, and preset second fitting constants; perform pixel grayscale time integration calculation on the first local region of interest to extract the initial perfusion slope characteristics of the contrast agent; perform inter-frame pixel motion field analysis on the first local region of interest to extract the initial shear force characteristics of the contrast agent; calculate the contrast agent fluid viscosity correction factor based on the current altitude data and the initial perfusion slope characteristics of the contrast agent; the specific calculation formula for the contrast agent fluid viscosity correction factor is as follows: ,in, This is the contrast agent fluid viscosity correction factor. This represents an exponential function with the natural constant as its base. To preset the standard absolute viscosity reference value for contrast agents, This indicates the preset first fitting constant. This indicates the current altitude data. The initial perfusion slope characteristics of the contrast agent. To preset a reference infusion slope baseline value, This represents the preset second fitting constant; the initial shear force characteristics of the contrast agent are corrected according to the contrast agent fluid viscosity correction factor to obtain the fluid shear force characteristics.

[0033] In this embodiment, the current altitude data can be obtained in real time through the on-board barometer of the mobile stroke unit.

[0034] The preset absolute viscosity benchmark value for contrast agents is derived from the physical property testing standards of the contrast agent manufacturers, and its typical value under normal temperature and plain conditions is [value missing]. ;

[0035] The first preset fitting constant is an empirical coefficient characterizing the compensatory increase and decreased deformability of red blood cells caused by altitude, with a typical value of .

[0036] The preset reference perfusion slope benchmark is the statistically derived standard proximal artery perfusion grayscale change rate, with a typical value range of [missing value]. .

[0037] The second fitting constant is preset as a fitting exponent characterizing the shear thinning properties of non-Newtonian fluids, calibrated based on the Casson fluid model, and its value range is [value range missing]. .

[0038] The initial perfusion slope characteristics of the contrast agent were obtained by calculating the slope of the linear regression of the average gray-time curve of the contrast agent in the local region of interest.

[0039] The initial shear stress characteristics of the contrast agent were extracted by calculating the velocity gradient of the pixels at the tube wall edge using the Lucas-Kanade optical flow method, and then converted into a preliminary estimate of the initial shear stress by combining the physical spatial resolution of the image.

[0040] Through the above technical solution, this embodiment constructs a non-Newtonian fluid compensation formula by introducing altitude environmental parameters and objective perfusion slope of the image to correct viscosity, which solves the technical problem of neglecting the hysteresis artifact caused by high altitude and changes in blood rheology in the prior art, and achieves the technical effect of objectively restoring the underlying fluid dynamic characteristics in complex physical environments.

[0041] Furthermore, the specific process for obtaining the grayscale space attenuation gradient features is as follows: Obtain the preset pixel deformation extreme value limit, the preset human core temperature reference value, and the preset temperature pixel deformation mapping coefficient; obtain the ambient absolute temperature data and calculate the pipe wall deformation baseline accordingly. The specific calculation formula is as follows: ,in, The baseline for pipe wall deformation. To find the maximum value function, To limit the preset pixel deformation extreme values, It is the hyperbolic tangent function. This indicates the preset human core temperature reference value. This is the absolute ambient temperature data. The preset temperature pixel deformation mapping coefficient is used; the initial attenuation gradient is obtained by taking the spatial derivative of the pixel grayscale matrix of the second local region of interest; the geometric shrinkage artifact component caused by temperature deviation is filtered out from the initial attenuation gradient based on the pipe wall deformation baseline to obtain the grayscale space attenuation gradient feature.

[0042] In this embodiment, the absolute ambient temperature data is acquired in real time using a temperature and humidity sensor.

[0043] The preset pixel deformation extreme value limit is derived from the statistical analysis of the maximum physiological contraction threshold of vascular smooth muscle, and its value range is [value range missing]. The preset human core temperature reference value is set to the standard human deep body temperature. .

[0044] The preset temperature pixel deformation mapping coefficient characterizes the sensitivity of smooth muscle calcium ion channels to temperature steps, with a value range of [value range missing]. .

[0045] The formula for calculating the baseline of tube wall deformation uses the smooth saturation characteristics of the hyperbolic tangent function to simulate the nonlinear stress limit of biological tissues, and uses the maximum value function to achieve truncation.

[0046] The gradient magnitude is calculated using the Sobel operator to differentiate the pixel grayscale matrix in space. When filtering out artifact components, a spatial low-pass filter is constructed using the shrinkage ratio corresponding to the pipe wall deformation baseline to attenuate the edge gradient of high-frequency artifacts.

[0047] Through the above technical solution, this embodiment introduces the absolute ambient temperature and uses the hyperbolic tangent extreme value limiting model to extract the tube wall deformation baseline to implement high-frequency gradient attenuation. This solves the technical problem of physiological spasm being misjudged as organic occlusion under extreme cold stress, and achieves the technical effect of accurately decoupling the true permeability resistance of the lesion from the mixed gray-scale gradient.

[0048] Furthermore, the specific process for obtaining the underlying spatial constraint mask is as follows: Integrating the pixel grayscale values ​​within the first local region of interest (ROI) over time to generate a local grayscale accumulation matrix; extracting the local vessel centerline within the ROI, and using the local vessel centerline as a symmetrical segmentation reference, dividing the local grayscale accumulation matrix into a reference side matrix and a retention side matrix; calculating the sum of pixel grayscale values ​​in the reference side matrix and the retention side matrix respectively, and calculating the absolute difference between them to obtain the spatial eccentricity integral; numerically concatenating the global integral value of the local grayscale accumulation matrix with the spatial eccentricity integral to obtain the asymmetric retention feature; and extracting pixels whose fluid shear force features are greater than a preset shear force threshold. The region is used as the basic spatial region; the pixel connected domain composed of non-zero pixels in the asymmetric retention feature is extracted; the spatial intersection region between the basic spatial region and the pixel connected domain is extracted, and the geometric eccentricity of the spatial intersection region is calculated; it is determined whether the geometric eccentricity is greater than a preset narrow threshold; if the geometric eccentricity is not greater than the preset narrow threshold, the boundary of the spatial intersection region is used as the morphological contour to generate the bottom spatial constraint mask; if the geometric eccentricity is greater than the preset narrow threshold, it is determined that the spatial intersection region represents the abnormal permeation artifact caused by the rupture of the intervascular layer, an extreme value rejection mask signal is generated, and the weighted feature sampling of the gray space attenuation gradient feature is interrupted based on the extreme value rejection mask signal, and the intervascular layer prediction probability data is output.

[0049] In this embodiment, the extraction of the local blood vessel centerline employs a morphological thinning algorithm based on skeletonization.

[0050] The typical value for the preset shear force threshold is: The geometric eccentricity is calculated using the second-order central moment of the region image.

[0051] The preset narrow threshold is determined through statistical analysis of large sample images, limiting its value range to [specific range]. between.

[0052] If the geometric eccentricity is greater than the preset narrow threshold, the spatial intersection region is determined to represent the abnormal permeation artifact caused by the rupture of the vascular interlayer. An extreme rejection mask signal is generated, and the gray space attenuation gradient features are sampled by weighted feature sampling based on the interruption of the extreme rejection mask signal, and the vascular interlayer prediction probability data is output. Through the above technical solution, this embodiment solves the technical problem that traditional methods cannot distinguish between plaque base retention and fatal vascular dissection false lumen retention by combining spatial eccentric integral and geometric eccentricity threshold determination. It achieves the technical effect of accurate morphological recognition of complex underlying foundation lesions and automatic physical foolproof fuse in extremely high-risk failure scenarios.

[0053] Furthermore, before performing inter-frame pixel motion field analysis and pixel gray-scale time integration calculation on the first local region of interest, the method further includes: calculating the time series variance of the overall gray-scale contrast of the continuous temporal imaging image sequence within a preset sliding time window; determining whether the time series variance is greater than a preset flow pulsation threshold; if the time series variance is greater than the preset flow pulsation threshold, then based on the difference between the time series variance and the preset flow pulsation threshold, increasing the initial sampling frame number parameter used for pixel gray-scale time integration calculation to obtain the target sampling frame number parameter, and performing subsequent pixel gray-scale time integration calculation based on the target sampling frame number parameter.

[0054] In this embodiment, before performing correlation analysis on the first local region of interest, the time series variance of the overall grayscale contrast of the continuous time-series imaging image sequence within a preset sliding time window is calculated in advance; and it is determined whether it is greater than a preset flow pulsation threshold.

[0055] The preset sliding time window typically includes The duration of a frame.

[0056] If the time series variance is greater than the preset flow pulsation threshold, the initial sampling frame number parameter used for pixel gray-scale time integration calculation is increased based on the difference between the two to obtain the target sampling frame number parameter. Subsequent pixel gray-scale time integration calculation is then performed based on the target sampling frame number parameter. The corresponding compensation logic is as follows: the larger the difference, the more intense the blood flow pulsation. The target sampling frame number will be automatically increased proportionally to the initial sampling frame number to smooth out high-frequency pulsation noise in the time domain.

[0057] Through the above technical solution, this embodiment solves the technical problem that abnormal fluctuations in transient flow velocity seriously interfere with static feature extraction by monitoring grayscale contrast variance and adaptively increasing the number of integral sampling frames, and achieves the technical effect of feedforward smoothing and data fidelity of dynamic calculation features under unstable flow conditions.

[0058] Furthermore, the specific process for obtaining the internal density index is as follows: obtain the pixel gradient matrix corresponding to the grayscale space attenuation gradient feature; mark the pixel values ​​within the coverage area of ​​the bottom spatial constraint mask as 1 and the pixel values ​​outside the area as 0, and convert them into a Boolean value space matrix; perform element-wise Hadamard product operation on the Boolean value space matrix and the pixel gradient matrix to obtain the filter gradient matrix; perform spatial integration and summation operation on the non-zero elements in the filter gradient matrix to obtain the internal density index.

[0059] In this embodiment, the element-wise Hadamard product operation, as a means of implementing the spatial hard attention mechanism, can accurately eliminate invalid gradients in the lesion area and ensure that the density index obtained by integration only reflects the microscopic permeability of the interface between the underlying plaque and the thrombus.

[0060] Through the above technical solution, this embodiment solves the technical problem of density misjudgment caused by background specular interference in global gradient calculation by Boolean matrix masking and Hadamard product precise mapping, and achieves the technical effect of targeted extraction of the micro-density of the core region of occluded objects in complex backgrounds.

[0061] Furthermore, the specific process for obtaining the predicted probability data is as follows: based on the event time window parameter, the corresponding nonlinear penalty weight is obtained by querying the preset time feature mapping table; the internal density index is multiplied by the nonlinear penalty weight to obtain the mechanistic density parameter; the mechanistic density parameter is input into the preset probability distribution function to calculate the predicted probability data containing the target physical state classification confidence.

[0062] In this embodiment, the event time window parameter is the absolute time difference between the time the target patient takes the contrast agent and the time of data acquisition.

[0063] The preset time feature mapping table is constructed based on the statistical analysis of in vitro thrombus organization and pathology. The longer the time window, the greater the corresponding nonlinear penalty weight index.

[0064] The probability distribution function uses the Softmax function to map the machined dense parameters into a normalized multi-class probability vector.

[0065] Through the above technical solution, this embodiment amplifies the compactness by introducing an event time window query nonlinear penalty weight, which solves the technical problem that static images alone cannot assess the degree of dynamic hardening of occlusions over time, and achieves the technical effect of high confidence physical state prediction with deep coupling of spatial morphology and temporal evolution parameters.

[0066] Figure 3This is a schematic diagram of the image analysis-based cerebral angiography thrombectomy data prediction system provided in an embodiment of this application. The image analysis-based cerebral angiography thrombectomy data prediction system includes: a data acquisition module for acquiring a continuous temporal angiography image sequence and event time window parameters of the target vascular region; a region extraction module for extracting vascular contours and tracking pixel time sequences from the continuous temporal angiography image sequence to obtain a first local region of interest (MOI) and a second MOI; an image feature extraction module for performing inter-frame pixel motion field analysis and pixel gray-level time integration calculation on the first MOI to obtain fluid shear force features and asymmetric retention features; and a gray-level attenuation extraction module for performing image feature extraction on the second MOI. The grayscale matrix is ​​differentiated in space to extract grayscale attenuation gradient features; the constraint mask processing module is used to perform spatial topological fusion of fluid shear force features and asymmetric retention features to obtain the bottom-level spatial constraint mask of the target vascular region; the compactness index processing module is used to use the bottom-level spatial constraint mask as a spatial constraint benchmark to perform weighted feature sampling of grayscale attenuation gradient features to obtain the internal compactness index; the prediction output module is used to perform nonlinear time-weighted mapping of the internal compactness index according to the event time window parameter to generate and output predicted probability data characterizing the composite physical state of the blockage within the target vascular region.

[0067] This application also provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements a method for predicting cerebral angiography thrombectomy data based on image analysis.

[0068] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks of the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0073] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for predicting cerebral angiography thrombectomy data based on image analysis, characterized in that, Includes the following steps: Acquire a continuous temporal angiography image sequence and event time window parameters for the target vascular region; The continuous temporal angiography image sequence is subjected to vascular contour extraction and pixel temporal tracking to obtain a first local region of interest and a second local region of interest; Inter-frame pixel motion field analysis and pixel grayscale time integration calculation are performed on the first local region of interest to obtain fluid shear force characteristics and asymmetric retention characteristics. The grayscale matrix space derivative of the second local region of interest is calculated to extract the grayscale space attenuation gradient feature. The fluid shear force feature and the asymmetric retention feature are spatially topologically fused to obtain the underlying spatial constraint mask of the target blood vessel region. Using the underlying spatial constraint mask as a spatial constraint reference, the internal compactness index is obtained by weighted feature sampling of the gray-scale spatial attenuation gradient features. Based on the event time window parameter, a nonlinear time-weighted mapping is performed on the internal density index to generate and output predicted probability data characterizing the composite physical state of the blocker within the target vascular region.

2. The method for predicting cerebral angiography thrombectomy data based on image analysis according to claim 1, characterized in that, The specific process for obtaining the first and second local regions of interest is as follows: The continuous time-series contrast imaging image sequence is divided into a contrast agent filling period sequence and a contrast agent emptying period sequence; Vascular contour extraction and pixel temporal tracking were performed on the contrast agent filling phase sequence and the contrast agent emptying phase sequence to calculate the time decay curve of the contrast agent flow rate. The pixel coordinates corresponding to the extreme points of flow velocity decay in the time decay curve are determined as flow blocking anchor points. Using the flow blocking anchor point as a reference, spatial expansion is performed to delineate the first local region of interest (ROI) representing the boundary of the blood vessel wall in the contrast agent emptying sequence, and the second ROI representing the end of the blocker in the contrast agent filling sequence.

3. The image analysis-based method for predicting cerebral angiography thrombectomy data according to claim 2, characterized in that, The specific process for obtaining the fluid shear force characteristics is as follows: Get the current altitude data; Obtain the preset standard absolute viscosity reference value of the contrast agent, the preset first fitting constant, the preset reference perfusion slope reference value, and the preset second fitting constant; Pixel grayscale time integration is performed on the first local region of interest to extract the initial perfusion slope feature of the contrast agent. Inter-frame pixel motion field analysis was performed on the first local region of interest to extract the initial shear force characteristics of the contrast agent. The contrast agent fluid viscosity correction factor is calculated based on the current altitude data and the initial perfusion slope characteristics of the contrast agent; the specific calculation formula for the contrast agent fluid viscosity correction factor is as follows: ,in, This is the contrast agent fluid viscosity correction factor. This represents an exponential function with the natural constant as its base. To preset the standard absolute viscosity reference value for contrast agents, This indicates the preset first fitting constant. This indicates the current altitude data. The initial perfusion slope characteristics of the contrast agent. To preset a reference infusion slope baseline value, This indicates a preset second fitting constant; The initial shear force characteristics of the contrast agent are corrected by the contrast agent fluid viscosity correction factor to obtain the fluid shear force characteristics.

4. The method for predicting cerebral angiography thrombectomy data based on image analysis according to claim 1, characterized in that, The specific process for obtaining the grayscale space attenuation gradient feature is as follows: Obtain the preset pixel deformation extreme value limit, the preset human core temperature reference value, and the preset temperature pixel deformation mapping coefficient; Obtain ambient absolute temperature data and calculate the pipe wall deformation baseline accordingly. The specific calculation formula is as follows: ,in, The baseline for pipe wall deformation. To find the maximum value function, To limit the preset pixel deformation extreme values, It is the hyperbolic tangent function. This indicates the preset human core temperature reference value. This is the absolute ambient temperature data. Preset temperature pixel deformation mapping coefficient; The initial attenuation gradient is obtained by taking the spatial derivative of the pixel grayscale matrix of the second local region of interest. Based on the pipe wall deformation baseline, the geometric shrinkage artifact component caused by temperature deviation is filtered out from the initial attenuation gradient to obtain the grayscale space attenuation gradient feature.

5. The method for predicting cerebral angiography thrombectomy data based on image analysis according to claim 1, characterized in that, The specific process for obtaining the underlying spatial constraint mask is as follows: Perform time-dimensional integration on the pixel grayscale values ​​within the first local region of interest to generate a local grayscale accumulation matrix; Extract the local blood vessel centerline within the first local region of interest, and use the local blood vessel centerline as a symmetrical segmentation reference to divide the local gray-level accumulation matrix into a reference side matrix and a stagnant side matrix; The pixel grayscale sums of the reference side matrix and the lingering side matrix are calculated respectively, and the absolute difference between the two is calculated to obtain the spatial eccentricity integral. The global integral value of the local gray-level accumulation matrix is ​​numerically concatenated with the spatial eccentric integral value to obtain the asymmetric retention feature. Pixel regions whose fluid shear force characteristics are greater than a preset shear force threshold are extracted as the basic spatial region; Extract the pixel connected components formed by non-zero pixels in the asymmetric retention feature; Extract the spatial intersection region between the basic spatial region and the pixel connected region, and calculate the geometric eccentricity of the spatial intersection region; Determine whether the geometric eccentricity is greater than a preset narrow threshold; If the geometric eccentricity is not greater than the preset narrow threshold, then the boundary of the spatial intersection region is used as the shape contour to generate the underlying spatial constraint mask. If the geometric eccentricity is greater than the preset narrow threshold, the spatial intersection region is determined to represent abnormal permeation artifacts caused by inter-vascular rupture. An extreme rejection mask signal is generated, and the weighted feature sampling of gray-scale spatial attenuation gradient features is interrupted based on the extreme rejection mask signal, and inter-vascular rupture prediction probability data is output.

6. The method for predicting cerebral angiography thrombectomy data based on image analysis according to claim 1, characterized in that, Before performing inter-frame pixel motion field analysis and pixel gray-scale time integration calculation on the first local region of interest, the following steps are also included: Calculate the time series variance of the overall grayscale contrast of the continuous temporal imaging image sequence within a preset sliding time window; Determine whether the variance of the time series is greater than a preset flow pulsation threshold; If the time series variance is greater than the preset flow pulsation threshold, then based on the difference between the time series variance and the preset flow pulsation threshold, the initial sampling frame number parameter used for pixel grayscale time integration calculation is increased to obtain the target sampling frame number parameter, and subsequent pixel grayscale time integration calculation is performed based on the target sampling frame number parameter.

7. The method for predicting cerebral angiography thrombectomy data based on image analysis according to claim 1, characterized in that, The specific process for obtaining the internal density index is as follows: Obtain the pixel gradient matrix corresponding to the grayscale space attenuation gradient feature; The pixel values ​​within the area covered by the underlying spatial constraint mask are marked as 1, and the pixel values ​​outside the area are marked as 0, thus converting it into a Boolean space matrix; The element-wise Hadamard product operation is performed between the Boolean space matrix and the pixel gradient matrix to obtain the filter gradient matrix. The internal compactness index is obtained by performing spatial integration and summation on the non-zero elements in the filter gradient matrix.

8. The method for predicting cerebral angiography thrombectomy data based on image analysis according to claim 1, characterized in that, The specific process for obtaining the predicted probability data is as follows: Based on the event time window parameter, the corresponding nonlinear penalty weight is obtained by querying the preset time feature mapping table. Multiplying the internal compactness index by the nonlinear penalty weight yields the mechanistic compactness parameter; The mechanical density parameter is input into a preset probability distribution function to calculate the predicted probability data, which includes the target physical state classification confidence.

9. A cerebral angiography thrombectomy data prediction system based on image analysis, characterized in that, include: Data acquisition module: used to acquire continuous time-series angiographic image sequences and event time window parameters of the target vascular region; Region extraction module: used to extract blood vessel contours and track pixel temporal sequence on the continuous temporal angiography image sequence to obtain a first local region of interest and a second local region of interest; Image feature extraction module: used to perform inter-frame pixel motion field analysis and pixel gray-scale time integration calculation on the first local region of interest to obtain fluid shear force features and asymmetric retention features; Gray-level attenuation extraction module: used to perform spatial differentiation of the pixel gray-level matrix of the second local region of interest to extract the gray-level attenuation gradient features; Constraint mask processing module: used to perform spatial topological fusion of the fluid shear force feature and the asymmetric retention feature to obtain the bottom spatial constraint mask of the target blood vessel region; Compactness index processing module: used to obtain the internal compactness index by weighted feature sampling of the gray-scale space attenuation gradient features with the underlying spatial constraint mask as the spatial constraint reference; Prediction output module: used to perform nonlinear time-weighted mapping on the internal density index according to the event time window parameter, and generate and output predicted probability data characterizing the composite physical state of the blocker in the target vascular region.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-8.

Images