Automatic segmentation method and system for cardiology echocardiogram

Abstract

The invention relates to the technical field of image segmentation, in particular to an automatic segmentation method and system for an echocardiogram of the department of cardiology, and the method comprises the following steps: based on image data of the echocardiogram, extracting gray level distribution, edge feature and texture feature information, analyzing gray level change amplitude, screening gray level change abnormal regions, and recognizing connectivity features. According to the method, the segmentation accuracy is effectively improved by extracting image gray, edge and texture features and identifying abnormal regions, noise and artifact interference are reduced by optimizing low connectivity regions, and the segmentation accuracy is improved. A problem area is analyzed and positioned in combination with multi-frame gray level change, a segmentation result is adjusted, the processing stability and consistency are enhanced, meanwhile, an optimized alarm node is output based on a frequency trend, more accurate and stable heart image analysis is supported, and the clinical application practicability is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image segmentation, in particular to an automatic segmentation method and system for echocardiography in cardiology. BACKGROUND

[0002] The technical field of image segmentation involves an important technology in the field of computer vision, aiming to separate the target region from the background in a digital image. The core task of image segmentation is to divide an image into several regions with similar features through algorithms. The regions can be different objects, scenes, or backgrounds. This technology is widely used in medical image processing, autonomous driving, remote sensing image analysis, and other fields. In the medical field, image segmentation is commonly used in medical image analysis and diagnosis to help doctors more accurately identify and locate lesion areas, tissue structures, and organs. With the continuous advancement of computer technology and artificial intelligence, image segmentation technology is also developing towards high precision, automation, and real-time performance. Among them, the automatic segmentation method of traditional internal medicine echocardiogram refers to a technical method based on echocardiogram images for analyzing the structure and function of the heart. Due to the complex noise, artifacts, and low resolution of echocardiogram images, traditional image segmentation methods use techniques such as thresholding, edge detection, region growing, or model matching to identify and segment the heart region. Most of these methods rely on manually set rules or preliminary segmentation results, and are affected by image quality, noise, and changes in heart shape, resulting in inaccurate segmentation results. Traditional methods cannot meet clinical needs and cannot achieve high standards in real-time performance and accuracy.

[0003] Echocardiography is a core means of non-invasive assessment of cardiac structure and function in cardiology, and is widely used in the diagnosis and follow-up of coronary heart disease, heart failure, congenital heart disease, and other diseases. In clinical diagnosis, doctors need to judge the condition through quantitative indicators such as left ventricular ejection fraction, wall thickness, and cardiac volume. The calculation of these indicators depends on the accurate boundary division of the cardiac anatomical structure (such as the left ventricular cavity, myocardial wall, and mitral valve).

[0004] Under the traditional diagnosis and treatment mode, doctors need to manually outline the boundaries of the heart structure. This process has two major problems: first, it is highly subjective, and the results of different doctors or the same doctor at different times can differ by up to 10-15%, which can lead to diagnostic bias, especially for early lesions or borderline cases. Second, it is inefficient, and manual segmentation of a single patient's ultrasound sequence (containing 200-500 dynamic images) takes 15-30 minutes, which cannot meet the needs of emergency rapid assessment or large-scale screening, and hinders the efficient advancement of the diagnosis and treatment process.

[0005] The existing echocardiogram segmentation techniques mainly use traditional image processing methods (threshold segmentation, edge detection, region growing, etc.), which are difficult to cope with the inherent characteristics of ultrasound images, resulting in insufficient segmentation accuracy and stability. The specific manifestations are as follows: Noise and artifact interference: Ultrasound images have speckle noise, acoustic shadow artifact, mirror artifact, etc. Traditional methods rely on single gray scale or edge features, which are prone to misjudgment of noise as structural boundaries or boundary breakage due to artifact obstruction. For example, the traditional segmentation accuracy of the left ventricular apex is often less than 70% due to acoustic shadow artifact.

[0006] Large fluctuations in image quality: Ultrasound image quality is significantly affected by operator technique (such as probe angle, pressure), patient individual differences (such as obesity, lung gas interference), and the contrast ratio and resolution of images from the same patient at different examination times can differ by more than 30%. Traditional methods rely on fixed rules set by humans (such as gray scale threshold, edge intensity threshold), which cannot be adjusted adaptively, and the segmentation failure rate in low-quality images is as high as 40%.

[0007] Poor dynamic structure adaptability: The heart is a dynamic organ that beats, and within a cardiac cycle (about 0.8-1.2 seconds), the volume of the heart chamber and the shape of the myocardium change periodically. Traditional methods are mostly aimed at single-frame static image segmentation and do not consider the temporal correlation between frames, which is prone to "jump" in structural boundaries in dynamic sequences. For example, in left ventricular short-axis plane segmentation, the inter-frame boundary offset can reach 5-8 pixels, which cannot meet the needs of dynamic functional evaluation.

[0008] With the development of cardiovascular disease diagnosis and treatment towards "precision and individualization", the clinical requirements for ultrasound segmentation have been upgraded from "rough positioning" to "sub-millimeter accuracy" and "real-time analysis". For example, in transcatheter aortic valve replacement (TAVR), the aortic valve annulus diameter needs to be accurately measured (error tolerance range ±1mm) to select the valve model, and the error of traditional segmentation methods (±2-3mm) cannot meet the needs of surgical planning; in the follow-up of myocardial infarction patients, dynamic monitoring of changes in abnormal areas of wall motion is required, and the low efficiency of traditional manual segmentation makes it impossible to achieve quantitative comparison at multiple time points, which restricts the accuracy of disease progression evaluation.

[0009] The existing technology mainly relies on traditional methods based on threshold, edge detection, region growing or model matching for image segmentation. Traditional methods are easily affected when dealing with complex noise, artifacts and low-resolution images, resulting in inaccurate segmentation results. Since traditional methods rely on manually set rules or preliminary segmentation results, the segmentation effect is unstable when faced with poor image quality or large changes in heart shape, which cannot meet the high real-time and high-accuracy requirements in clinical applications. For example, in echocardiography, noise and artifacts can severely affect segmentation accuracy, making it difficult to accurately identify some lesion areas and affecting doctors' diagnostic decisions. SUMMARY

[0010] In order to solve the technical problems existing in the prior art, the embodiment of the present application provides an automatic segmentation method for echocardiograms in cardiology, comprising the following steps: In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: an automatic segmentation method for echocardiograms in cardiology, comprising the following steps: S1: based on echocardiogram image data, extracting gray scale distribution, edge feature and texture characteristic information, sorting and classifying the spatial constraint relationship of the differentiated regions in the image, labeling the feature points of each frame of image according to the time label, and generating an image feature and spatial constraint synchronous data set; S2: based on the image feature and spatial constraint synchronous data set, extracting a multi-level gray scale gradient sequence, analyzing the gray scale variation amplitude in the unit area, and comparing the variation law of adjacent frames, screening the regions with abnormal gray scale variation, combining a texture consistency discrimination model to mark the abnormal regions, and obtaining an image original segmentation segment; S3: calling the image original segmentation segment, extracting the boundary gray scale value and texture matching result in each segment, identifying the connectivity feature in the continuous region, and when the connectivity is lower than a preset threshold, marking it as a segmentation defect area, and collecting and outputting the segmentation defect segment; S4: according to the segmentation defect segment, identifying the gray scale gradient arrangement and texture consistency distribution in the corresponding frame, selecting high-frequency defect points for behavior trajectory mapping, judging the difference trajectory of the stable segmentation stage, and marking the interval with an offset amplitude greater than the normal segment, and obtaining a segmentation result correction distance.

[0011] As a further scheme of the present application, the image feature and spatial constraint synchronous data set includes a gray scale distribution label, a texture characteristic timestamp, and a multi-source feature alignment field, the image original segmentation segment includes a gray scale abnormal region identifier, a texture matching difference amount, and a spatial constraint indicator factor, the segmentation defect segment includes a connectivity upper and lower limit value, a texture matching abnormality judgment point, and a continuous defect identification identifier, and the segmentation result correction distance includes a stable trajectory offset amount, a high-frequency defect position point, and a segmentation state comparison feature.

[0012] As a further scheme of the present application, the step of the image feature and spatial constraint synchronous data set is specifically: S101: based on echocardiogram image data, extracting gray scale, edge and texture features, analyzing the spatial constraint of the differentiated regions, matching and sorting the two types of data according to the time label, and generating an image feature and spatial constraint corresponding set; S102: According to the image feature and the space constraint corresponding set, the region gray and the edge feature are distinguished under the time label, the joint change amplitude of the gray and the texture is analyzed, the segment continuously exceeding the abnormal threshold is extracted, and the gray abnormal fluctuation interval value is obtained. S103: Based on the gray abnormal fluctuation interval value, the texture offset and the space constraint change rate are extracted, the synchronous amplitude difference and the duration are combined, the segmentation region connectivity and the boundary integrity abnormality are judged, and the image feature and the space constraint synchronous data set are generated.

[0013] As a further scheme of the application, the abnormal threshold refers to a numerical limit for judging whether the image feature is abnormal. The gray abnormal fluctuation interval value refers to a continuous time period range corresponding to the image gray feature when exceeding the abnormal threshold.

[0014] As a further scheme of the application, the step of the image original segmentation segment is specifically: S201: Based on the image feature and the space constraint synchronous data set, a multi-level gray gradient sequence is extracted, the interframe change rule is identified, the record exceeding the gray change threshold is screened, and the gray abnormal fluctuation interval is generated. The gray change threshold refers to a numerical limit for judging the interframe change of the image gray value. S202: According to the gray abnormal fluctuation interval, the texture characteristic distribution of the corresponding time period is analyzed, the texture combination corresponding to the gray is extracted, and it is judged whether the interframe difference exceeds the segmentation safety threshold, the abnormal region is marked, and the image original segmentation segment is obtained. The segmentation safety threshold refers to a judgment standard for judging whether the texture difference is safe to execute image segmentation.

[0015] As a further scheme of the application, the step of the segmentation defect section is specifically: S301: The continuous gray value and the texture matching result in the image original segmentation segment are identified, pairing is performed according to the same time period, each connectivity feature is identified, and a connectivity fluctuation sequence is generated. S302: Based on the connectivity fluctuation sequence, the upper and lower boundaries of the continuous interval are extracted, the abnormal section is identified in combination with the segmentation reference value, the defect start and end time and the fluctuation range are determined, and the connectivity critical fluctuation interval range is obtained. S303: According to the connectivity critical fluctuation interval range, the corresponding time period gray value and the texture matching result in the gray abnormal segment are extracted, the connectivity offset amplitude is analyzed, the offset combination of the critical section is summarized, and the segmentation defect section is output.

[0016] As a further scheme of the application, the step of the segmentation result correction distance is specifically: S401: According to the segmented defect section, the intra-frame gray scale gradient and texture consistency are extracted, the gray scale instantaneous deviation rate is analyzed, the mutation point is identified and the multi-source sequence is located synchronously, and the high-frequency defect behavior node is obtained; S402: Based on the high-frequency defect behavior node, the texture track and the difference between the stable state baseline are compared, the deviation threshold behavior point is identified, the deviation direction and duration are located, and the texture deviation interference section is established; S403: According to the texture deviation interference section, the intra-frame gray scale deviation trend and the texture disturbance amount are extracted, the deviation total amount of the composite interference period is calculated, the section with low overlap rate and deviation duration exceeding the reference length is identified, and the segmentation result correction distance is obtained.

[0017] As a further scheme of the application, the method further comprises a step S5: S5: Based on the segmentation result correction distance, the trigger frequency of the defect event in the continuous frame is counted, the high-risk area is located according to the frequency accumulation trend, each area is assigned a corresponding grade label, and a segmented abnormal early warning point list is output; The segmented abnormal early warning point list includes defect grade partition, event frequency threshold and alarm node identification number.

[0018] As a further scheme of the application, the step of the segmented abnormal early warning point list is specifically: S501: Based on the segmentation result correction distance, the defect event of the continuous time period is extracted, the trigger frequency of each frame is counted and the cumulative growth trend is identified, the continuous position interval with increasing frequency is identified, and the segmented high-risk section group is obtained. S502: According to the segmented high-risk section group, the trigger frequency, duration and deviation amplitude of the high-risk section are combined to screen the sections exceeding the risk grading threshold, the trigger grade is marked according to the time position, and a segmented abnormal early warning point list is output.

[0019] The automatic segmentation system of echocardiogram of cardiology department comprises: The feature synchronization module is based on the echocardiogram image data, including gray scale distribution, edge feature and texture characteristic information, sorts the time sequence according to the period, extracts the spatial constraint relationship in each period, analyzes the mapping relationship between the time label and the feature, and forms the image feature and spatial constraint association data set; The segmentation and identification module is based on the image feature and spatial constraint association data set, extracts the gray scale gradient sequence of each region, compares the change amplitude and increase-decrease law of adjacent frames, screens the abnormal fluctuation frequency band, identifies the texture characteristic distribution of the corresponding frame, and generates a gray scale abnormal fluctuation feature group; The communication discrimination module extracts the gray value and texture matching result of each group of data based on the gray abnormal fluctuation feature group, analyzes the communication feature in the continuous frames, judges whether it is lower than the segmentation reference threshold, locates the defect paragraph, and obtains the communication abnormal section; The trajectory analysis module matches the gray gradient sequence and texture characteristic sequence of each frame based on the communication abnormal section, extracts the high-frequency point recognition behavior time trajectory, and compares the offset with the stable segmentation behavior trajectory, recognizes the trajectory variation feature, and establishes the segmentation behavior offset path group. The node optimization module counts the abnormal fluctuation trigger times in the path, analyzes the region mapping relationship, performs grade division on the high-frequency points, locodes the trigger concentration area, and outputs the segmentation abnormal early warning point list.

[0020] Compared with the prior art, the advantages and positive effects of the present application are that: In the present application, by carefully extracting and analyzing the gray distribution, edge feature, texture consistency and other information in the image, the abnormal area in the image is accurately captured and processed, and the accuracy of segmentation is improved. By identifying and optimizing the low connectivity area, the segmentation defects caused by noise, artifacts and low resolution can be effectively reduced, the stability and reliability of segmentation are improved, and the problem area can be accurately located in the dynamically changing image and adjusted in time by combining time dimension for multi-frame analysis, optimizing the segmentation process, and finally improving the accuracy and consistency of the image segmentation result. This method not only effectively identifies and corrects the defects in segmentation, but also realizes the optimization warning based on frequency analysis, which helps to perform more accurate and stable cardiac image analysis in clinical or real-time detection, and meets the higher clinical needs. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0022] Figure 1 The step flowchart of the present application is shown in the figure; Figure 2 The S1 refinement schematic diagram of the present application is shown in the figure; Figure 3 The S2 refinement schematic diagram of the present application is shown in the figure; Figure 4 The S3 refinement schematic diagram of the present application is shown in the figure; Figure 5 The S4 refinement schematic diagram of the present application is shown in the figure; Figure 6 The S5 refinement schematic diagram of the present application is shown in the figure; Figure 7 System module diagram of the present application. DETAILED DESCRIPTION

[0023] The technical solutions in the present application will be described below with reference to the drawings.

[0024] In the embodiments of the present application, the words such as "example", "for example" are used to represent an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0025] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times, and it should be pointed out that the meanings expressed are consistent when the distinction is not emphasized. "Of", "corresponding" and "relevant" can be used interchangeably at times, and it should be pointed out that the meanings expressed are consistent when the distinction is not emphasized.

[0026] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1, and the meanings expressed are consistent when the distinction is not emphasized.

[0027] To make the technical problems, technical solutions and advantages to be solved by the present application clearer, specific embodiments will be described in detail below with reference to the drawings.

[0028] Please refer to Figure 1 The embodiments of the present application provide an automatic segmentation method for echocardiograms in cardiology departments, which comprises the following steps: S1: Based on the echocardiogram image data, extract the gray scale distribution, edge feature and texture characteristic information, sort and classify the spatial constraint relationship of the differentiated areas in the image, label the feature points of each frame of image according to the time label, and generate the image feature and spatial constraint synchronous data set; S2: Based on the image feature and spatial constraint synchronous data set, extract the multi-level gray scale gradient sequence, analyze the gray scale change amplitude in the unit area, compare the change law of adjacent frames, screen the areas with abnormal gray scale change, mark the abnormal areas in combination with the texture consistency discrimination model, and obtain the image original segmentation segment; S3: Call the image original segmentation segment, extract the boundary gray scale value and texture matching result in each segment, identify the connectivity feature in the continuous area, and when the connectivity is lower than a preset threshold, mark it as a segmentation defect area, and then aggregate and output the segmentation defect area segment; S4: According to the segmented defect section, the arrangement of the gray scale gradient and the texture consistency distribution in the corresponding frame are identified, the high-frequency defect points are selected for behavior trajectory mapping, the difference trajectory of the stable segmentation stage is judged, and the interval with an offset amplitude greater than the normal section is marked to obtain the segmentation result correction distance; S5: Based on the segmentation result correction distance, the trigger frequency of the defect event in the continuous frame is counted, the high-frequency area is located according to the frequency accumulation trend, each area is assigned a corresponding grade label, and the segmented abnormal early warning point list is output.

[0029] The image feature and space constraint synchronous data set includes a gray scale distribution label, a texture characteristic timestamp, and a multi-source feature alignment field. The image original segmentation segment includes a gray scale abnormal area identifier, a texture matching difference, and a space constraint indication factor. The segmented defect section includes a connectivity upper and lower limit value, a texture matching abnormal judgment point, and a continuous defect recognition identifier. The segmentation result correction distance includes a stable trajectory offset, a high-frequency defect position point, and a segmentation state comparison feature. The segmented abnormal early warning point list includes a defect grade partition, an event frequency threshold, and an alarm node identifier number.

[0030] Please refer to Figure 2 , the steps of the image feature and space constraint synchronous data set are: S101: Based on the echocardiogram image data, the gray scale, edge and texture features are extracted, the space constraint of the difference area is analyzed, the two types of data are matched and sorted according to the time label, and the image feature and space constraint corresponding set is generated; Based on echocardiogram image data, a sequence of continuous real-time ultrasound image frames is received, for each frame, the gray scale feature is extracted, for example, a specific region of the left ventricle is selected, the average gray scale value of all pixels in the region is calculated, for example, the average gray scale value is 120, the edge feature is extracted, the gray scale gradient of each pixel point is calculated by applying the Sobel operator, and the average edge intensity of the region of interest is calculated, for example, the average edge intensity is 45, the texture feature is extracted, the gray level co-occurrence matrix (GLCM) is used to obtain the contrast, energy, homogeneity and other parameters of the local region of the image, for example, the contrast of a certain region is 25, the energy is 0.4, and the homogeneity is 0.85, the spatial constraint of the difference region is analyzed, according to the extracted gray scale, edge and texture features, different heart structures such as heart cavity and myocardial wall are preliminarily identified and distinguished, for example, the myocardial region is determined by the pixel gray scale value range and edge continuity, and its geometric boundary is determined according to its relative position in the image (such as the ventricular wall is located outside the chamber), the relative distance and interaction between regions (for example, the left ventricular cavity diameter is measured as 5.0 centimeters at end diastole, and the myocardial thickness is 1.0 centimeters) form a spatial constraint, the two types of data are matched and sorted according to the time label, the acquisition time of each frame of image is accurately recorded as the time label, for example, the first frame is 0.00 seconds, and the second frame is 0.04 seconds, the image feature data and spatial constraint data corresponding to the frame are associated, and are arranged in chronological order to ensure the time sequence consistency of the data, and an image feature and spatial constraint corresponding set is generated. The corresponding set is a structured data set, which includes image features (gray scale, edge, texture) and spatial constraint information under each time label.

[0031] S102: According to the image feature and spatial constraint corresponding set, the region gray scale and edge feature are distinguished under the time label, the joint change amplitude of gray scale and texture is analyzed, the segment continuously exceeding the abnormal threshold value is extracted, and the gray scale abnormal fluctuation interval value is obtained; The abnormal threshold value refers to the numerical limit for judging whether the image feature is abnormal; The gray scale abnormal fluctuation interval value refers to the continuous time period range corresponding to the image gray scale feature when it exceeds the abnormal threshold value; According to the image feature and space constraint corresponding set, access to pre-generated time-labeled image feature and space constraint data, for example, obtain all image frame information within the time period 0.00-10.00 seconds, judge the region gray scale and edge feature under the time label, the system checks the pixel gray value distribution and edge intensity in the specific region of interest frame by frame, for example, for the left ventricular cavity region, statistics its gray value distribution and calculates the mean value of edge intensity, analyzes the joint change amplitude of gray scale and texture, comprehensively considers the change degree of gray value and texture feature of selected region between continuous frames, for example, if the average gray value of the left ventricular cavity changes from 120 to 80, the change amplitude is 40, the texture energy changes from 0.4 to 0.25, the change amplitude is 0.15, then the joint change amplitude is calculated by multiplying the absolute value of the gray scale change amplitude by the weight (for example 0.6) and the absolute value of the texture change amplitude by the weight (for example 0.4) and adding them together, for example, the calculation result is 24.06, extract the segment continuously exceeding the abnormal threshold, the abnormal threshold refers to the numerical limit for judging whether the image feature is abnormal, for example, the gray scale abnormal threshold is set to be greater than or equal to 30 for the average gray value change or greater than or equal to 20 for the average edge intensity change, the threshold is determined by statistical analysis of a large number of normal echocardiogram sequences, for example, by analyzing 100 normal cases, it is found that the frame-to-frame change is less than 20, so the threshold is set to be twice the standard deviation, identify the segment with joint change amplitude continuously exceeding the preset threshold 25 for multiple frames (for example 3 frames), for example, if the joint change amplitudes of the three consecutive frames are 26.1, 28.5 and 27.3 respectively, all greater than 25, then the three frames are marked as abnormal, and the gray scale abnormal fluctuation interval value is obtained, which refers to the continuous time period range corresponding to the image gray scale feature when exceeding the abnormal threshold.

[0032] S103: Based on the gray scale abnormal fluctuation interval value, extract the texture offset and the space constraint change rate, combine the synchronous amplitude difference and the duration, determine the segmentation region connectivity and boundary integrity abnormality, and generate an image feature and space constraint synchronous data set; Based on the gray abnormal fluctuation interval value, for example, interval [0.00s, 0.08s] and [1.20s, 1.32s], the texture offset and spatial constraint change rate are extracted, in each abnormal fluctuation interval, for the continuous frames, the numerical change amount of the texture features (such as contrast, entropy) and the numerical change amount of the spatial constraints (such as area, shape parameters) are calculated, and the change amount is divided by the corresponding time interval, if the texture entropy changes from 3.5 to 4.0, the time interval is 0.04 seconds, then the texture offset rate is 12.5 entropy units / second, if the area changes from 30cm² to 32cm², then the spatial constraint change rate is 50cm² / second, combined with the synchronous amplitude difference and the duration, the synchronous amplitude difference refers to the amplitude difference of the image features (gray, texture) and the spatial constraint change in the abnormal interval, the duration is the total length of the abnormal fluctuation interval, for example, if the gray, texture and spatial constraint all show changes higher than the normal fluctuation threshold in 0.00s to 0.08s, and the duration is 0.08s, it is determined that the segmentation region connectivity and boundary integrity are abnormal, the determination standard of the segmentation region connectivity abnormality is that the heart region of the continuous frames is broken, merged or appears isolated pixel group in the gray abnormal fluctuation interval, and the number of connected components changes more than 3, the determination standard of the boundary integrity abnormality is that the region boundary length or smoothness changes significantly, for example, the length changes more than 10% or the number of inflection points increases more than 5, for example, in the interval [0.00s, 0.08s], if the left ventricular cavity connected region changes from 1 to 2, and the boundary smoothness index decreases from 0.9 to 0.6, it is determined to be abnormal, the image feature and spatial constraint synchronous data set is generated, which records all the information that the image features and spatial constraints show synchronous abnormalities in a certain period of time.

[0033] Please refer to Figure 3 , the steps of the original image segmentation segment are as follows: S201: Based on the image feature and spatial constraint synchronous data set, a multi-level gray gradient sequence is extracted, the interframe change rule is identified, the records exceeding the gray change threshold are screened, and the gray abnormal fluctuation interval is generated; The gray change threshold refers to the numerical limit of the change of the image gray value between frames; Based on the image feature and the space constraint synchronous dataset, for example, the detailed data of the synchronous abnormal interval [0.00s, 0.08s] and [1.20s, 1.32s] is obtained, and a multi-level gray gradient sequence is extracted. For each synchronous abnormal interval, the gray gradient of each frame image is calculated in different scales, for example, the gray gradient of the pixel point is calculated using 3x3 and 5x5 Laplace operator, a multi-level gradient image is formed, and the gradient value is arranged in time sequence to obtain a gray gradient sequence, for example, the 3x3 gradient mean value of the left ventricular region of the 0.00s frame is 15, and the 0.04s frame is 18. The change rule between frames is identified, the multi-level gray gradient sequence of the same region between consecutive frames is compared, and the significant change of the gradient mean value or distribution mode is found, for example, the 3x3 gradient mean value of the left ventricular region from the 0.00s frame to the 0.04s frame is increased from 15 to 18, and the change amplitude is 3. The records exceeding the gray scale change threshold value are screened out. The gray scale change threshold value refers to the numerical limit for judging the change of image gray scale value between frames. The threshold value is set to be greater than or equal to 5, which is the change amount of the multi-level gray gradient mean value between consecutive frames. This threshold value is determined by statistically analyzing the inter-frame gray gradient change of a large number of normal ultrasound sequences and taking twice the standard deviation of the average change amount, for example, if the normal inter-frame change is less than 2.5, the threshold value is set to 5. The records whose inter-frame gray gradient mean value change amplitude exceeds this threshold value are screened out, for example, if the 3x3 gradient mean value changes from 15 to 21, the change amplitude is 6, and this frame is screened out, and the gray abnormal fluctuation interval is generated.

[0034] S202: According to the gray abnormal fluctuation interval, the texture characteristic distribution of the corresponding time period is analyzed, the texture combination corresponding to the gray scale is extracted, and it is judged whether the difference between frames exceeds the segmentation safety threshold value. The abnormal region is marked, and the original image segmentation fragment is obtained. The segmentation safety threshold value refers to the judgment standard for judging whether the texture difference is safe to execute image segmentation; For example, the gray abnormal fluctuation interval [0.00s, 0.08s], analyze the texture characteristic distribution of the corresponding time period, for each frame image in the interval, recalculate and analyze the texture characteristics of the region, for example, in the interval [0.00s, 0.08s], calculate the contrast 25, energy 0.4, homogeneity 0.85 of the left ventricular region in the 0.00s frame, and the contrast 28, energy 0.38, homogeneity 0.82 of the 0.04s frame, extract the texture combination corresponding to the gray value, combine the gray value and the corresponding texture characteristics of each frame in each gray abnormal fluctuation interval, for example, for the 0.00s frame, the gray mean value is 120, the texture combination is (contrast 25, energy 0.4, homogeneity 0.85), and it is judged whether the difference between frames exceeds the segmentation safety threshold, the segmentation safety threshold refers to the judgment standard for judging whether the texture difference is safe to execute image segmentation, the threshold is set to be greater than or equal to 0.15 in the Euclidean distance of the texture feature vector, the setting of this threshold is based on the experience value and statistical analysis of the texture feature fluctuation in the normal segmentation operation, for example, in the successfully segmented region, the Euclidean distance of the texture features of the consecutive frames is usually less than 0.1, the Euclidean distance of the texture combination corresponding to the gray value between the consecutive frames is calculated, for example, the Euclidean distance of the texture feature vector of the 0.00s frame and the 0.04s frame is 3.0002, and the abnormal region is marked, if the Euclidean distance of the texture combination between frames exceeds the segmentation safety threshold 0.15, the region corresponding to the frame is marked as an abnormal region, for example, since 3.0002 is greater than 0.15, the left ventricular region in the 0.04s frame is marked as abnormal, the image original segmentation segment is obtained, and the image region segment which is judged to be abnormal due to too large texture difference in the gray abnormal fluctuation interval is identified and output, the segment is the starting point of subsequent defect analysis.

[0035] Please refer to Figure 4 The steps of segmenting the defect section are as follows: S301: Identify the continuous gray value and texture matching result in the image original segmentation segment, pair by the same time period, identify each group of connectivity characteristics, and generate a connectivity fluctuation sequence; The continuous gray value in the original segmentation segment of the image and the texture matching result are identified, for example, the original segmentation segment of the left ventricular region in the 0.04 second frame is obtained, the continuous gray value sequence in the segment is identified, for example, for a specific pixel point on the left ventricular wall, the gray values in the 0.04 second, 0.08 second and 0.12 second frames are 150, 152 and 148 respectively, and the corresponding texture matching result is obtained. The texture matching result is obtained by calculating the texture feature similarity of the corresponding region in the continuous frames, for example, the normalized cross-correlation coefficient is used for measurement, the texture matching degree of the region around the pixel point between the 0.04 second frame and the 0.08 second frame is 0.92, and the same time period is paired. The system pairs the continuous gray value and the texture matching result under the same time label, for example, the gray value 150 of the 0.04 second frame is paired with the texture matching result 0.92 between the 0.04 second frame and the 0.08 second frame. The connectivity feature of each group is identified, and the connectivity feature of the image region in each paired time period is identified, including the number of connected components in the region, the area, the perimeter and the geometric center coordinates of each connected component, for example, the abnormal left ventricular region in the 0.04 second frame contains 3 independent connected components, and the areas are 100, 50 and 20 pixels respectively. The change of the connectivity feature (for example, the number of connected components) identified in the continuous time period is recorded to form a time sequence, for example, if the number of connected components in the 0.04 second frame is 3, the number of connected components in the 0.08 second frame is 2, and the number of connected components in the 0.12 second frame is 4, then the connectivity fluctuation sequence is generated.

[0036] S302: Based on the connectivity fluctuation sequence, the upper and lower boundaries of the continuous interval are extracted, the abnormal section is identified in combination with the segmentation reference value, the defect start and end time and the fluctuation range are determined, and the connectivity critical fluctuation interval range is obtained. Based on the connectivity fluctuation sequence, the upper and lower boundaries of the continuous interval are extracted, the connectivity fluctuation sequence is analyzed, the starting point of the connectivity feature (e.g. the number of connected components) fluctuating from the normal range to the abnormal range and the ending point of the connectivity feature recovering from the abnormal range to the normal range are identified, so as to determine the upper and lower time boundaries of the continuous abnormal interval, for example, if the normal number of connected components should be 1, and there are 3, 2 and 4 connected components in the sequence, the starting point is 0.04 seconds when there are 3 connected components, and the ending point is the frame before recovering to 1 connected component, for example, the number of connected components of the 0.16 second frame recovers to 1, the upper boundary of the abnormal interval is 0.04 seconds, and the lower boundary is 0.12 seconds, the abnormal section is identified in combination with the segmentation reference value, the segmentation reference value is a preset standard for evaluating the quality of image segmentation, for example, the region connectivity reference value is set to 1 connected component, and the boundary integrity reference value is set to a boundary smoothness coefficient greater than or equal to 0.95, the reference value is based on the clinical cognition of the morphology of the heart structure and the requirement for segmentation accuracy, the values in the connectivity fluctuation sequence are compared with the segmentation reference value, if the number of connected components in the sequence continuously deviates from the reference value, for example, continuously greater than 1, it is identified as an abnormal section, the start and end time of the defect and the fluctuation range are determined, according to the comparison result, the accurate start and end points of the abnormal section are determined, and the maximum and minimum fluctuation values of the connectivity feature in the time period are recorded, for example, from 0.04 seconds to 0.12 seconds, the fluctuation range of the number of connected components is 2 to 4, the connectivity critical fluctuation interval range is obtained, which represents the time range in which the connectivity feature in the original segmentation fragment of the image continuously deviates from the normal segmentation reference value.

[0037] S303: According to the connectivity critical fluctuation interval range, the corresponding period gray value and texture matching result in the gray abnormal fragment are extracted, the connectivity offset amplitude is analyzed, the offset combination of the critical section is summarized, and the segmentation defect section is output. According to the connectivity critical fluctuation interval range, the gray value and texture matching result of each frame are extracted from the gray abnormal segment in the critical fluctuation interval, for example, for 0.04 second frame, the gray mean value is 110, the texture matching result is 0.92, for 0.08 second frame, the gray mean value is 105, the texture matching result is 0.88, the connectivity offset amplitude is analyzed, the connectivity offset amplitude is calculated by comparing the difference between the actual connectivity component number in the connectivity critical fluctuation interval and the normal connectivity component reference value (for example, 1), for example, if the connectivity component number of 0.04 second frame is 3, the offset amplitude is 2, if the connectivity component number of 0.08 second frame is 2, the offset amplitude is 1, the offset combination of critical section is summarized, the gray value, texture matching result and connectivity offset amplitude at each time point are combined to form the offset combination list of critical section, for example, for 0.04 second frame, the offset combination is (gray 110, texture matching 0.92, connectivity offset 2), for 0.08 second frame, the offset combination is (gray 105, texture matching 0.88, connectivity offset 1), the segmented defect section is output, and finally the segmented defect section is output, which contains all the specific time points showing connectivity, gray or texture abnormalities and their corresponding offset combination information in the connectivity critical fluctuation interval.

[0038] Please refer to Figure 5 The step of segmenting the result correction distance is specifically: S401: According to the segmented defect section, the gray gradient and texture consistency in the frame are extracted, the gray instantaneous offset rate is analyzed, the mutation point is identified and the multi-source sequence is positioned synchronously, and the high-frequency defect behavior node is obtained. According to the segmented defect section, for each defect time point, the gray level gradient of the defect region in the frame image is calculated, for example, the Laplace operator is applied to the defect region in the 0.04 second frame, and the average gray level gradient of the pixels in the region is 25, at the same time, the similarity between different texture features (such as contrast, energy) in the region is evaluated, for example, the Pearson correlation coefficient of different small block texture features in the region is calculated, if the average value is 0.95, the texture consistency is high, the gray level instantaneous shift rate is analyzed, the gray level instantaneous shift rate is calculated by comparing the average gray value change amount of the defect region between consecutive frames and the time interval, for example, the gray level of the 0.04 second frame is 110, the gray level of the 0.08 second frame is 105, the time interval is 0.04 seconds, and the gray level instantaneous shift rate is -125 gray level units per second, the mutation point is identified and the multi-source sequence is located synchronously, the mutation point refers to the frame whose absolute value of gray level instantaneous shift rate exceeds the preset mutation threshold, the mutation threshold is set to 100 gray level units per second, and the threshold is determined by statistical analysis of the gray level change rate in a normal cardiac cycle and taking 3 times the standard deviation, the frame whose absolute value of gray level instantaneous shift rate is greater than 100 is identified as a mutation point, for example, the gray level instantaneous shift rate from 0.04 seconds to 0.08 seconds is -125, which is greater than 100, so the 0.08 second frame is identified as a mutation point, and the data of 0.08 seconds in the original ultrasonic image sequence, gray feature sequence and texture feature sequence corresponding to the mutation point are located at the same time, and the high-frequency defect behavior node is obtained.

[0039] S402: Based on the high-frequency defect behavior node, the difference between the texture track and the steady-state baseline is compared, the offset super-threshold behavior point is identified, the offset direction and the duration are located, and the texture offset interference section is established; Based on the high-frequency defect behavior node, the texture track is compared with the difference of the steady-state baseline. The texture track refers to the numerical sequence of the texture characteristics (such as contrast, energy, and homogeneity) of a specific region in the time sequence. For example, the texture energy of 0.00 second frame is 0.4, the texture energy of 0.04 second frame is 0.38, and the texture energy of 0.08 second frame is 0.25. The steady-state baseline is determined by analyzing the texture characteristics of the same region in a large number of normal echocardiogram sequences to determine the average value and fluctuation range of the region in the normal physiological state. For example, the steady-state baseline of the texture energy of the left ventricular myocardium is 0.35±0.02. The texture track at the high-frequency defect behavior node and its nearby frames is compared with the corresponding steady-state baseline. For example, at 0.08 second frame, the texture energy is 0.25, which is lower than the normal fluctuation range of the steady-state baseline 0.35. The offset threshold behavior point is identified. The offset threshold refers to the difference between the texture track and the steady-state baseline, that is, the texture offset, and the absolute value of the offset threshold is greater than the preset offset threshold. The offset threshold is set to 0.05. The threshold is determined by statistical analysis of the texture energy fluctuation in the normal physiological state and taking 3 times the average absolute deviation. For example, if the normal texture energy absolute deviation is less than 0.015, the frame with an absolute value of the texture offset greater than 0.05 is identified as the offset threshold behavior point. For example, the offset of 0.08 second frame texture energy 0.25 and steady-state baseline 0.35 is 0.1, which is greater than 0.05. Therefore, 0.08 second frame is identified as the offset threshold behavior point. The offset direction and duration are located. For each offset threshold behavior point, it is determined whether the texture feature is higher or lower than the steady-state baseline (offset direction), and it is identified how long the offset state lasts. For example, the texture energy of 0.08 second frame is 0.25, which is lower than the steady-state baseline 0.35, and the offset direction is negative. If the texture energy of 0.12 second frame is still 0.28, the offset state lasts at least 0.08 seconds to 0.12 seconds. The continuous offset threshold behavior points are combined to form a texture offset interference section.

[0040] S403: According to the texture offset interference section, the in-frame gray scale offset trend and the texture disturbance amount are extracted, the offset total amount of the composite interference period is calculated, the section with low overlap rate and offset duration exceeding the reference length is identified, and the segmentation result correction distance is obtained. The offset total amount of the composite interference period is calculated by the formula: ; Wherein, represents the offset total amount of the composite interference period, represents the in-frame gray scale offset amount in the frame, represents the texture disturbance weight factor in the frame, represents the texture disturbance amount in the frame, represents the texture disturbance amount in the frame, represents the texture disturbance amount in the frame, represents the texture disturbance amount in the frame, the arithmetic mean of the texture disturbance quantity of all frames in the interference section, the total number of frames in the interference section; texture disturbance section , the local gray level shift trend and the texture disturbance quantity are extracted, for each frame in the interference section, the local gray level shift trend of the defect area in the frame image is extracted, for example, for the 0.08 second frame, the average gray level value of different small areas in the defect area is calculated relative to the gray level value of the adjacent normal area, to obtain a series of local gray level shift quantities, and the texture disturbance quantity of the frame is extracted, the texture disturbance quantity represents the first frame texture disturbance quantity, which is obtained by calculating the Euclidean distance between the texture feature (such as contrast, energy) of the defect area of the frame and the reference texture feature in the normal physiological state, for example, if the texture feature vector of the left ventricular defect area of the 0.08 second frame is , and the normal reference texture feature vector of the area is , then the texture disturbance quantity is: ; the total shift quantity of the composite interference period is calculated, the total shift quantity of the composite interference period is calculated by the formula , wherein, represents the total shift quantity of the composite interference period, represents the first frame, the local gray level shift quantity in the frame is obtained by measuring the difference between the average gray level value of the defect area in the frame and the average gray level value of the surrounding normal area, for example, if the average gray level value of the defect area of the 0.08 second frame is 105, and the average gray level value of the surrounding normal area is 120, then , represents the first frame texture disturbance weight factor, which is set according to the importance of the texture disturbance quantity to the image segmentation, and the value range is 0 to 1, the setting of the weight is based on the expert experience evaluation and regression analysis of the sensitivity of the texture feature in the image segmentation, for example, if the texture feature is considered to have medium importance in the image segmentation, then 0.5 is set as, represents the first frame texture disturbance quantity, which is calculated as before, for example, the texture disturbance quantity of the 0.08 second frame is 3.0002, represents the arithmetic mean of the texture disturbance quantity of all frames in the interference section, which is obtained by summing the texture disturbance quantities of all frames in the interference section and dividing by the number of frames, for example, if the texture disturbance quantity of the 0.08 second frame is 3.0002, and the texture disturbance quantity of the 0.12 second frame is 2.8005, then , n represents the total number of frames in the interference section, for example, the interference section contains 2 frames, so ; The advantage of the formula is that by combining the gray offset amount and the texture disturbance amount, and introducing a texture disturbance weight factor, the total offset amount of the image under the combined interference can be quantified more comprehensively and more finely, so as to realize accurate evaluation and correction of the segmentation defects in the overall system or method. Now the parameters are assigned and calculated: For 0.08 second frames, , , ; For 0.12 second frames, assume , , ; , ; Substitute the formula to calculate: ; Identify the section with low overlap rate and offset duration exceeding the reference duration, the reference duration refers to the maximum duration allowed for image feature offset under normal circumstances, for example, the reference duration is set to 0.1 seconds (corresponding to 2-3 frames), this duration is set based on statistical analysis of the duration of small fluctuations in normal echocardiogram sequences, check whether the overlap rate of the image segmentation area in the combined interference period is lower than the preset low overlap rate threshold (for example, lower than 80%), and whether its offset duration exceeds the reference duration of 0.1 seconds, for example, if the duration of the interference section is 0.04 seconds, it does not meet the condition of continuously exceeding the reference duration, but if another section has a duration of 0.2 seconds and an overlap rate of only 75%, this section is identified. According to the above conditions, the total amount of combined offset of the section with low overlap rate and long time offset identified is, for example, if after this screening, the section with a total amount of combined offset fails to be identified as needing correction, while another section with a duration of 0.2 seconds and an overlap rate of 75% has a calculated value of 18.75, then this value is the segmentation result correction distance.

[0041] Please refer to Figure 6 , the steps of the segmentation anomaly early warning point list are: S501: Based on the segmentation result correction distance, extract the defect events of the continuous time period, count the trigger times of each frame and identify the cumulative growth trend, identify the continuous position interval with increasing frequency, and obtain the high-risk section group of segmentation; The base statement corrects the segmentation result distance, for example, obtains a correction distance value of 18.75, extracts defect events in a continuous time period, identifies and records all image regions marked as segmentation defects and their corresponding time points in the image sequence corresponding to the correction distance, for example, identifies that there are segmentation defects at 0.08 seconds, 0.12 seconds, and 0.16 seconds, counts the number of triggers for each frame and identifies the cumulative growth trend, counts the number of defect events for each time tag (i.e., each frame), for example, if there is 1 defect in the 0.08 second frame, 2 defects in the 0.12 second frame, and 1 defect in the 0.16 second frame, and calculates the cumulative growth trend of the number of defect triggers, for example, if the number of triggers for three consecutive frames is 1, 2, and 3, respectively, it shows an increasing trend, identifies the continuous position interval with increasing frequency, and identifies the continuous time period in which the number of defect triggers shows a sustained increasing trend. The time period is considered a risk accumulation area, for example, if the number of defect triggers steadily increases from 1 time / frame to 3 times / frame in the time period of 0.40 seconds to 0.60 seconds, obtain the segmentation risk high-risk section group, which includes all identified continuous position intervals with sustained increasing defect trigger frequency, for example, the segmentation risk high-risk section group is ([0.40s, 0.60s], [1.20s, 1.40s]).

[0042] S502: According to the segmentation risk high-risk section group, combined with the trigger frequency, duration and offset amplitude of the high-risk section, filter out the sections that exceed the risk classification threshold, mark the time position according to the trigger level, and output the segmentation anomaly early warning point list; For example, the high-risk segment group is obtained by segmenting the risk area, and the trigger frequency, duration and offset amplitude of the high-risk segment are combined to evaluate each segment in the high-risk segment group, and the internal defect trigger frequency (defect event number per second), the duration of the segment (seconds), and the average offset amplitude (for example, average gray offset or texture disturbance) of the defect events occurring in the segment are considered comprehensively. For example, for the segment [0.40s, 0.60s], the duration is 0.20 seconds, if there are 10 defect events in this segment, the trigger frequency is 50 times / second, and if the average gray offset of all defects in this segment is 15, the segment exceeding the risk classification threshold is screened out. The risk classification threshold is a numerical limit for judging the risk level of image segmentation, for example, the risk classification threshold is set as: the trigger frequency is greater than or equal to 40 times / second, the duration is greater than or equal to 0.15 seconds, and the average offset amplitude is greater than or equal to 10 units. The threshold is set based on the clinical requirements for ultrasound image segmentation quality and diagnostic reliability. Compare each index of each high-risk segment with the risk classification threshold. Only the segment that meets all conditions will be screened out, for example, the trigger frequency of the segment [0.40s, 0.60s] is 50 times / second, the duration is 0.20 seconds, and the average offset amplitude is 15. All conditions are met, so the segment is screened out. Mark the trigger level time position, divide the high-risk segment into different trigger levels (for example, mild, moderate, severe) according to the indexes of the screened high-risk segment, and mark the accurate position of the high-risk segment on the time axis, for example, if the trigger frequency is 40-60 times / second, the duration is 0.15-0.25 seconds, and the offset amplitude is 10-20, it is marked as "moderate risk". The segment [0.40s, 0.60s] is marked as moderate risk. Output the segmentation anomaly warning point list, which includes a summary report of all high-risk segmentation defect segments, trigger levels and time positions, for example, the segmentation anomaly warning point list is: [(0.40s-0.60s, moderate risk), (1.20s-1.40s, severe risk)].

[0043] Please refer to Figure 7 , an automatic segmentation system for echocardiogram in cardiology department, comprising: The feature synchronization module sorts the time sequence according to the cycle based on the ultrasound echocardiogram image data, including gray scale distribution, edge feature and texture characteristic information, extracts the spatial constraint relationship in each cycle, analyzes the mapping relationship between time label and feature, and forms the image feature and spatial constraint association data set. The segmentation recognition module extracts a gray scale gradient sequence of each region based on an image feature and a spatial constraint association dataset, compares a change amplitude and an increase and decrease law of adjacent frames, screens an abnormal fluctuation frequency band, and identifies a texture characteristic distribution of a corresponding frame to generate a gray scale abnormal fluctuation feature group; The connectivity discrimination module extracts a gray scale value and a texture matching result of each group of data based on the gray scale abnormal fluctuation feature group, analyzes a connectivity feature in continuous frames, judges whether the connectivity feature is lower than a segmentation reference threshold, locates a defect paragraph, and obtains a connectivity abnormal section. The trajectory analysis module matches a gray scale gradient sequence and a texture characteristic sequence of each frame based on the connectivity abnormal section, extracts a high-frequency point to identify a behavior time trajectory, and compares the behavior time trajectory with a stable segmentation behavior trajectory to identify a trajectory variation feature and establish a segmentation behavior offset path group. The node optimization module counts a number of times of abnormal fluctuation triggering in the path, analyzes a region mapping relationship, performs grade division on a high-frequency point, locates a triggering concentration area, and outputs a segmentation abnormal early warning point list.

[0044] The application reduces diagnostic bias and guarantees diagnosis and treatment accuracy. The application realizes adaptive and accurate segmentation of a heart structure through multi-feature fusion (gray scale distribution, edge feature, and texture characteristic) and dynamic time sequence analysis. The segmentation accuracy can reach more than 90%, and the inter-frame boundary offset is controlled within 1-2 pixels, which significantly reduces artificial subjective bias. For example, the calculation error of left ventricular ejection fraction (LVEF) can be reduced from ±5% of traditional manual segmentation to ±2%, avoiding misdiagnosis (such as misjudging critical LVEF patients as normal or heart failure) caused by index error.

[0045] The automatic segmentation process of the application can shorten the analysis time of a single patient from 15-30 minutes to 1-2 minutes, and the efficiency is improved by more than 15 times. In the emergency scene (such as acute myocardial infarction), heart function indexes can be quickly obtained to gain time for thrombolysis or interventional treatment decision-making; in large-scale physical examination screening, automatic processing of batch data can be realized to assist early screening and intervention of cardiovascular diseases.

[0046] The application supports complex diagnosis and treatment scenarios and expands the application boundary. The abnormal early warning function (outputting a segmentation abnormal early warning point list) of the application can actively identify regions with poor image quality and abnormal structure morphology to assist doctors in focusing on key lesion sites. For example, in the diagnosis of congenital heart disease (such as ventricular septal defect), the position and range of the defect region can be automatically marked to provide accurate reference for surgical plan development; in heart rehabilitation follow-up, the changes of structure and function in the myocardial repair process can be dynamically tracked to realize adjustment of individualized rehabilitation plans.

[0047] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An automatic segmentation method of echocardiography in cardiology, characterized by, The method comprises the following steps: S1: based on echocardiogram image data, extracting gray scale distribution, edge feature and texture characteristic information, sorting and classifying the spatial constraint relationship of the differentiated regions in the image, labeling the feature points of each frame of image according to the time label, and generating image feature and spatial constraint synchronous data set; S2: based on the image feature and spatial constraint synchronous data set, extracting multi-level gray scale gradient sequence, analyzing the gray scale change amplitude in the unit area, and comparing the change law of adjacent frames, screening the regions with abnormal gray scale change, and marking the abnormal regions by combining the texture consistency discrimination model to obtain the image original segmentation segment; S3: calling the image original segmentation segment, extracting the boundary gray scale value and texture matching result in each segment, identifying the connectivity feature in the continuous region, and marking as a segmentation defect area when the connectivity is lower than the preset threshold, and then collecting and outputting the segmentation defect segment; S4: according to the segmentation defect segment, identifying the gray scale gradient arrangement and texture consistency distribution in the corresponding frame, selecting high-frequency defect points for behavior trajectory mapping, judging the difference trajectory of the stable segmentation stage, and marking the interval with an offset amplitude greater than the normal segment to obtain the segmentation result correction distance.

2. The automatic segmentation method of echocardiography according to claim 1, wherein, The image feature and spatial constraint synchronous data set comprises a gray scale distribution label, a texture characteristic timestamp, and a multi-source feature alignment field. The image original segmentation segment comprises a gray scale abnormal region identifier, a texture matching difference, and a spatial constraint indicator. The segmentation defect segment comprises a connectivity upper and lower limit value, a texture matching abnormal judgment point, and a continuous defect identification mark. The segmentation result correction distance comprises a stable trajectory offset, a high-frequency defect position point, and a segmentation state comparison feature. 3.The automatic segmentation method of intracardiac echocardiogram according to claim 1, characterized in that, The step of the image feature and spatial constraint synchronous data set is specifically: S101: based on echocardiogram image data, extracting gray scale, edge and texture features, analyzing the spatial constraint of the differentiated region, matching and sorting the two types of data according to the time label, and generating an image feature and spatial constraint corresponding set; S102: according to the image feature and spatial constraint corresponding set, identifying the region gray scale and edge feature under the time label, analyzing the joint change amplitude of gray scale and texture, extracting the segment continuously exceeding the abnormal threshold, and obtaining the gray scale abnormal fluctuation interval value; S103: based on the gray scale abnormal fluctuation interval value, extracting the texture offset and spatial constraint change rate, combining the synchronous amplitude difference and the duration, judging the connectivity and boundary integrity abnormality of the segmentation region, and generating the image feature and spatial constraint synchronous data set.

4. The automatic segmentation method of echocardiography according to claim 3, wherein, The abnormal threshold refers to the numerical limit for judging whether the image feature is abnormal; The gray scale abnormal fluctuation interval value refers to the continuous time period range corresponding to the image gray scale feature when it exceeds the abnormal threshold.

5. The automatic segmentation method of echocardiography according to claim 3, wherein, The step of the image original segmentation segment is specifically: S201: based on the image feature and spatial constraint synchronous data set, extracting multi-level gray scale gradient sequence, identifying the interframe change law, screening the records exceeding the gray scale change threshold, and generating the gray scale abnormal fluctuation interval; The gray scale change threshold refers to the numerical limit for judging the change of image gray scale value between frames; S202: According to the gray abnormal fluctuation interval, the texture characteristic distribution of the corresponding time period is analyzed, the texture combination corresponding to the gray is extracted, and it is judged whether the interframe difference exceeds the segmentation safety threshold, the abnormal area is marked, and the image original segmentation fragment is obtained; The segmentation safety threshold refers to a judgment standard for judging whether the texture difference is safe for image segmentation.

6. The automatic segmentation method of echocardiography according to claim 5, wherein, The segmentation defect section step is specifically: S301: Identify the continuous gray value and texture matching result in the image original segmentation fragment, pair by the same time period, identify each group of connectivity characteristics, and generate a connectivity fluctuation sequence; S302: Based on the connectivity fluctuation sequence, extract the upper and lower boundaries of the continuous interval, identify the abnormal section combined with the segmentation reference value, determine the defect start and end time and fluctuation range, and obtain the connectivity critical fluctuation interval range; S303: According to the connectivity critical fluctuation interval range, extract the corresponding time period gray value and texture matching result in the gray abnormal fragment, analyze the connectivity offset amplitude, and output the segmentation defect section by summarizing the offset combination of the critical section.

7. The automatic segmentation method of echocardiography according to claim 6, wherein, The segmentation result correction distance step is specifically: S401: According to the segmentation defect section, extract the intra-frame gray gradient and texture consistency, analyze the gray instantaneous offset rate, identify the mutation point and simultaneously locate the multi-source sequence, and obtain the high-frequency defect behavior node; S402: Based on the high-frequency defect behavior node, compare the texture track and the difference between the stable baseline, identify the offset super-threshold behavior point, locate the offset direction and duration, and establish the texture offset interference section; S403: According to the texture offset interference section, extract the intra-frame gray offset trend and texture disturbance, calculate the offset total amount of the composite interference period, identify the section with low overlap rate and offset duration exceeding the reference length, and obtain the segmentation result correction distance. 8.The automatic segmentation method of intracardiac echocardiography according to claim 1, characterized in that, The method further comprises a step S5: S5: Based on the segmentation result correction distance, count the trigger frequency of the defect event in the continuous frame, locate the high-risk area according to the frequency accumulation trend, assign a corresponding grade label to each area, and output a segmentation anomaly warning point list; The segmentation anomaly warning point list includes defect grade partition, event frequency threshold, and alarm node identification number.

9. The automatic segmentation method of echocardiography according to claim 8, wherein, The segmentation anomaly warning point list step is specifically: S501: Based on the segmentation result correction distance, extract the defect event of the continuous time period, count the trigger times of each frame and identify the cumulative growth trend, identify the continuous position interval with increasing frequency, and obtain the segmentation risk high-risk section group; S502: According to the segmentation risk high-risk section group, combined with the trigger frequency, duration and offset amplitude of the high-risk section, filter the sections exceeding the risk grading threshold, mark the time position according to the trigger level, and output the segmentation anomaly warning point list.

10. An automatic segmentation system for echocardiography in cardiology, characterized in that, The system is used to realize the automatic segmentation method of the intracardiac echocardiogram of the cardiology department according to any one of claims 1-9, and the system comprises: The feature synchronization module sorts the time sequence according to the period based on the echocardiogram image data including gray distribution, edge feature and texture characteristic information, extracts the spatial constraint relationship in each period, analyzes the mapping relationship between the time label and the feature, and forms an image feature and spatial constraint association data set; The segmentation identification module extracts a gray scale gradient sequence of each region based on the image features and the spatial constraint association dataset, compares a change amplitude and an increase and decrease law of adjacent frames, screens an abnormal fluctuation frequency band, and identifies a texture characteristic distribution of a corresponding frame to generate a gray scale abnormal fluctuation feature group; The connectivity discrimination module extracts a gray scale value and a texture matching result of each group of data based on the gray scale abnormal fluctuation feature group, analyzes a connectivity feature in continuous frames, judges whether the connectivity feature is lower than a segmentation reference threshold, locates a defect paragraph, and obtains a connectivity abnormal section; The trajectory analysis module matches a gray scale gradient sequence and a texture characteristic sequence of each frame based on the connectivity abnormal section, extracts a high-frequency point to identify a time trajectory, compares the time trajectory with a stable segmentation behavior trajectory, identifies a trajectory variation feature, and establishes a segmentation behavior offset path group; The node optimization module counts a number of times of triggering an abnormal fluctuation in the path, analyzes a region mapping relationship, performs grade division on a high-frequency point, locates a triggering central area, and outputs a segmentation abnormal early warning point list.

Images