Search

CN-120876510-B - Ischemic cerebrovascular disease angiography image segmentation analysis method

CN120876510BCN 120876510 BCN120876510 BCN 120876510BCN-120876510-B

Abstract

The invention relates to an ischemic cerebrovascular angiography image segmentation analysis method, which comprises the steps of detecting gray transition abnormality, structural fracture and artifact delay signals existing in an angiography image, extracting negative segmentation priori points indicating suspected ischemic areas to aggregate to form priori abnormal areas, introducing a symmetrical disturbance test mechanism to judge potential occlusion or abnormal vessel segments and dynamically adjusting segmentation thresholds of the areas, establishing a local disturbance window in the judging area, extracting frequency and rhythm characteristics of density stripes, performing compensation segmentation on the intermittent vessel segments caused by unsteady pulse changes through a convolution kernel expansion strategy, calculating texture difference values and frequency domain response offset, triggering interpolation complement mechanisms if the offset is within a preset physiological tolerance range, generating a trusted full-image layer for subsequent calibration reference, and performing dynamic feedback adjustment and continuity calibration on a previous segmentation path by analyzing multipath divergence degrees of vessel branch end points and path offset changes in an image sequence.

Inventors

  • CHEN SHAOHUI
  • Zhuang Xiaoyin
  • CAI FUGUI

Assignees

  • 普宁华侨医院

Dates

Publication Date
20260505
Application Date
20250722

Claims (9)

  1. 1. The ischemic cerebrovascular disease angiography image segmentation analysis method is characterized by comprising the following steps of: extracting a plurality of negative segmentation priori points for indicating suspected ischemic areas by detecting gray transition abnormality, structural fracture and artifact delay signals existing in angiography images, aggregating the priori points to form priori abnormal areas, introducing a symmetrical disturbance test mechanism into the priori abnormal areas, analyzing response stability to segmentation boundaries to judge potential occlusion or abnormal vessel segments, and dynamically adjusting segmentation thresholds of the areas; Establishing a local interference window along the trend of the blood vessel in a judging area, extracting the frequency and rhythm characteristics of density stripes, and carrying out compensation segmentation on the discontinuous blood vessel section caused by unsteady pulse change through a flexible strategy of a convolution kernel; And carrying out dynamic feedback adjustment and continuity calibration on the previous segmentation path by analyzing the multipath divergence degree of the vascular branch end point in the segmented vascular segment map and the path offset change of the vascular branch end point in the image sequence.
  2. 2. The method for analyzing the segmentation of the ischemic cerebrovascular disease angiography image according to claim 1, wherein the extraction process of the negative segmentation prior points comprises the steps of establishing a gray level and texture contradiction tensor, identifying a region which has gradient fracture characteristics and lacks texture continuity in the image as a structural artifact abnormal point, and determining an artifact delay signal by adopting an abnormal residual relation among image frames to reversely mark the intensity rise of a static region in an image at the later stage of perfusion time.
  3. 3. The method for analyzing the segmentation of the ischemic cerebrovascular disease angiography image according to claim 2, wherein the symmetrical disturbance test introduces two-way mirror image difference calculation, wherein the image response before and after disturbance is regarded as an unstable segmentation area when the symmetry error on two sides of a central axis of the blood vessel exceeds a set threshold value, the segmentation threshold of the dynamic adjustment area is updated according to the response confidence level field of the disturbance area, and the response unstable area is used for downwards regulating the trigger threshold and expanding the processing range of upstream and downstream blood vessels.
  4. 4. The method for analyzing the segmentation of the ischemic cerebrovascular angiography image according to claim 3, wherein the frequency extraction of the density stripes is combined with short-time Fourier transform and local frequency clustering to distinguish real vascular pulse stripes from contrast pseudo-belt stripes, the rhythmic features comprise space and time periodicity difference degrees, rhythmic unsteady deformation of capillary vessel segments caused by perfusion fluctuation is detected and used for compensating and judging micro-vessel fracture, a flexible strategy of the convolution kernel adopts a leachable deformation kernel, and the size of the flexible strategy is dynamically controlled by the change rate of local frequency energy and is used for keeping the segmentation sensitivity of abnormal regions.
  5. 5. The method for analyzing the segmentation of the ischemic cerebrovascular disease angiography image according to claim 4, wherein the texture difference comprises directional gradient field inversion calculation, the directional interpolation is carried out on the areas with inconsistent main directions of blood vessel segments at two ends and continuous textures, the judgment of the frequency domain response offset adopts the joint matching of a phase difference residual image and a position offset, the trusted image layer output by the completion mechanism simultaneously comprises a completion confidence grading image, and the completion segments at different levels selectively participate in subsequent path correction or manual auditing.
  6. 6. The method for analyzing the segmentation of the ischemic cerebrovascular angiography image according to claim 5, wherein the multipath divergence determination of the vascular branch end point adopts inter-frame direction field difference clustering, an abnormal priority correction list is established for the unstable path in the end point direction, a recursive path scoring system is adopted in the process of dynamically feeding back and adjusting the previous segmentation path, and the path reliability is scored and updated according to the continuity and the structural rationality of the historical path after each segmentation.
  7. 7. The method for analyzing the segmentation of the ischemic cerebrovascular angiography image according to claim 6, wherein the inter-frame direction field difference clustering adopts a combined clustering strategy of a spatial direction vector and an inter-frame relative rotation angle, and the calculation of the direction field difference adopts a cosine similarity weighting scheme based on spindle projection so as to inhibit false clustering caused by directional field mutation of a blood vessel intersection region.
  8. 8. The method for analyzing the segmentation of the ischemic cerebrovascular angiography image according to claim 7, wherein the abnormal priority correction list is subjected to multistage sorting according to the stability of the end point direction, the route access rate and the neighborhood vessel consistency score, the correction operation of the abnormal route comprises the minimum interpolation connection of the structural difference in the direction of the trusted route, and the end point evolution track of the call history image frame is adopted in the process of establishing the abnormal priority correction list for improving the false recognition tolerance under the condition of short-time direction field distortion.
  9. 9. The method for segmenting and analyzing ischemic cerebrovascular disease angiography images according to claim 8, wherein the recursive path scoring system introduces a path topology breakpoint penalty function, gives accumulated negative weight to fragments with repeated breaking behavior in a historical path, and introduces a smoothness index based on path curvature continuity in the path continuity score to distinguish abnormal deviation caused by physiological bending and path misguidance.

Description

Ischemic cerebrovascular disease angiography image segmentation analysis method Technical Field The invention relates to an angiography image segmentation analysis method, in particular to an ischemic cerebrovascular disease angiography image segmentation analysis method. Background In combination with the Chinese patent CN111126403A, a cerebral blood vessel segmentation method and a cerebral blood vessel segmentation system based on magnetic resonance angiography images still have various technical defects and practical application limitations in the current segmentation analysis method for ischemic cerebral vascular angiography images, and mainly show the dimensions of segmentation precision, structural continuity, abnormality identification capability, time sequence adaptability, result reliability and the like. Firstly, the existing method is mainly based on Magnetic Resonance Angiography (MRA) images to carry out cerebrovascular segmentation, and although a double Gaussian model is introduced to model the grey level distribution of the cerebral vessels, and the local continuity of the images is enhanced by utilizing a three-dimensional weighted Markov random field, the overall morphological fidelity of the cerebrovascular segmentation is improved to a certain extent, but the segmentation strategy is still mainly based on static image grey level statistics, the understanding of the dynamic perfusion process of the blood vessels and the evolution behavior of time sequence images is lacking, and the idea of static modeling and local smoothing is difficult to cope with complex pathological features such as unstable perfusion, blood flow fracture, artifact interference and the like which are common in ischemic cerebrovascular diseases. For example, in clinic, due to slow flow rate of contrast agent or delayed filling of vascular occlusion segment, intermittent gray level bands or structural backfill artifacts often appear in images, if only a middle-high gray level region is modeled by a double Gaussian model, micro blood vessels with slow perfusion and coherent structures are easily misjudged as the background, or delayed artifacts are marked as real passages, so that false negative and false positive coexist. Secondly, although three-dimensional Markov random fields are introduced in the invention to try to improve the continuity of a blood vessel structure, the core of the invention is that the spatial similarity among image pixels and the local neighborhood gray co-occurrence relation are depended on, and modeling capability of the directionality of the structure is lacking when the three-dimensional Markov random fields face to fracture, bifurcation or abnormal extension paths. Ischemic cerebrovascular diseases are particularly in branch areas of medium and small blood vessels, often manifested by weak capillary perfusion and unstable running direction, and if a segmentation system cannot judge according to the main direction of blood vessels or the topological connection trend, the problem of pseudo connection or fracture continuation is very easy to occur. The invention does not establish a logic feedback mechanism based on vascular path structure evolution, and can not carry out history judgment or scoring backtracking on abnormal paths, so that an effective screening means is lacked on structure misleading paths. Thirdly, the existing method is basically established on a single frame image to carry out segmentation decision, and lacks an inter-frame path continuity analysis mechanism. For Digital Subtraction Angiography (DSA) or dynamic MRA images, cerebral vessels have significant time-series evolution tracks in image sequences, such as perfusion sequencing, flow rate differences, final arterial perfusion delays, etc., which can be used to assist in judging the connected reality of vessel segments. However, the invention does not relate to joint modeling between image sequences, and does not set any scoring module based on path time consistency or inter-frame direction field consistency, so that the model cannot effectively identify non-structural changes such as direction mutation, transient occlusion or fracture recovery and the like caused by short-time artifacts, thereby influencing the stability and reliability of the final segmentation map. Fourth, the method in the invention mainly relies on gray modeling and markov neighborhood smoothing to adjust the segmentation boundary, and lacks multi-modal feature fusion capability. For example, in an actual angiogram, besides gray scale, a plurality of dimensional information such as texture details, pulse stripe rhythms, boundary response intensity and the like can assist in segmentation judgment, but the method does not introduce a rhythmic unsteady state detection mechanism, frequency domain response analysis or a texture direction complement module, so that key clinical manifestations such as micro-vascular fracture, unstable perfusion and the l