A mobile device remote evaluation of cardiovascular image assisted processing system

Abstract

The present application relates to the technical field of image analysis, in particular to a kind of mobile device remote evaluation cardiovascular image auxiliary processing system of piece, the system can realize the following steps by the mutual cooperation between multiple modules: the corresponding preset quantity frame cardiovascular image of the patient to be detected is obtained, and each frame cardiovascular image is equally divided;Filter out initial cardiovascular region, determine the overall edge irregularity corresponding to each image block;According to the edge information loss condition analysis of the down-sampling result of each image block;Determine the overall abnormal motion index corresponding to each image block;According to the overall edge irregularity, edge information loss degree and overall abnormal motion index corresponding to the image block in each frame cardiovascular image, identify target cardiovascular region from each frame cardiovascular image.The present application realizes cardiovascular region identification by analyzing multiple frame cardiovascular images, and improves the accuracy of cardiovascular region identification on cardiovascular image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image analysis, in particular to a cardiovascular image auxiliary processing system for mobile device remote evaluation. BACKGROUND

[0002] The cardiovascular image auxiliary processing system for mobile device remote evaluation is a system combining modern communication technology, electronic technology, computer technology and artificial intelligence algorithm, aiming to provide a convenient and efficient platform for doctors to perform remote evaluation of cardiovascular images on mobile devices. In cardiovascular image auxiliary processing, it is often necessary to identify the cardiovascular region from the cardiovascular image to assist doctors in cardiovascular evaluation. Currently, when performing region identification on an image, the commonly used method is to perform region segmentation on the image according to the difference in gray value.

[0003] However, when identifying the cardiovascular region from the cardiovascular image according to the difference in gray value, the accuracy of cardiovascular region segmentation may be poor. For example, due to the patient's involuntary movement or failure to fully cooperate with the breath-holding instruction, the collected cardiovascular image may be blurred or ghosted, thereby producing motion artifacts. At this time, if the cardiovascular region is identified only based on the difference in gray value, the accuracy of cardiovascular region segmentation may be poor due to the presence of motion artifacts. SUMMARY

[0004] To solve the technical problem of poor accuracy of cardiovascular region identification from a cardiovascular image, the present application provides a cardiovascular image auxiliary processing system for mobile device remote evaluation.

[0005] In a first aspect, the present application provides a cardiovascular image auxiliary processing system for mobile device remote evaluation, which comprises:

[0006] An acquisition and equalization module for acquiring a preset number of frames of cardiovascular images corresponding to a patient to be detected, and equally dividing each frame of cardiovascular image to obtain image blocks;

[0007] A screening and determination module for screening an initial cardiovascular region from each image block, and determining the overall edge irregularity corresponding to each image block based on the difference between the initial cardiovascular region in each image block before and after morphological opening operation, and the included angle between the gradient direction and the normal direction corresponding to the edge pixel points of the initial cardiovascular region in each image block;

[0008] An edge information loss condition analysis module for analyzing the edge information loss condition according to the down-sampling result of each image block to obtain the edge information loss degree corresponding to each image block;

[0009] The acquisition and determination module is configured to acquire the feature points in each image block, and determine an overall abnormal motion index corresponding to each image block based on the pre-acquired cardiovascular suspected skeleton corresponding to each image block and a motion condition of the feature points in each image block;

[0010] The region identification module is configured to identify a target cardiovascular region from each frame of cardiovascular image based on the overall edge irregularity, the edge information loss degree and the overall abnormal motion index corresponding to the image block in each frame of cardiovascular image through the cardiovascular identification network.

[0011] In combination with the first aspect, in a possible implementation manner, the determination of the overall edge irregularity corresponding to each image block based on the difference between the initial cardiovascular region in each image block before and after the morphological opening operation and the included angle between the gradient direction and the normal direction of the edge pixel points of the initial cardiovascular region in each image block comprises:

[0012] The morphological opening operation is performed on each initial cardiovascular region to obtain a reference region corresponding to each initial cardiovascular region;

[0013] The edge smoothing factor corresponding to each image block is determined according to the overlap difference between the initial cardiovascular region in each image block and the reference region corresponding thereto;

[0014] The direction irregularity corresponding to each image block is determined according to the included angle between the gradient direction and the normal direction of all edge pixel points on the initial cardiovascular region in each image block;

[0015] The overall edge irregularity corresponding to each image block is determined according to the edge smoothing factor and the direction irregularity corresponding to each image block.

[0016] In combination with the first aspect, in a possible implementation manner, the determination of the edge smoothing factor corresponding to each image block according to the overlap difference between the initial cardiovascular region in each image block and the reference region corresponding thereto comprises:

[0017] The number of pixel points in the intersection of the initial cardiovascular region in each image block and the reference region corresponding thereto is determined as a target intersection number corresponding to each image block;

[0018] The number of pixel points in the union of the initial cardiovascular region in each image block and the reference region corresponding thereto is determined as a target union number corresponding to each image block;

[0019] The ratio of the target intersection number to the target union number corresponding to each image block is determined as the edge smoothing factor corresponding to each image block.

[0020] In a possible implementation manner of the first aspect, the direction irregularity corresponding to each image block is determined according to an included angle between a gradient direction and a normal direction corresponding to all edge pixel points on the initial cardiovascular region in each image block.

[0021] The included angle between the gradient direction and the normal direction corresponding to each edge pixel point on the initial cardiovascular region in each image block is normalized to obtain a local direction difference corresponding to each edge pixel point on the initial cardiovascular region in each image block.

[0022] A mean value of the local direction differences corresponding to all edge pixel points on the initial cardiovascular region in each image block is determined as the direction irregularity corresponding to each image block.

[0023] In a possible implementation manner of the first aspect, the edge information loss degree corresponding to each image block is obtained by performing edge information loss analysis according to the down-sampling result of each image block.

[0024] The pyramid down-sampling manner is adopted to perform down-sampling on each image block for a preset number of times to obtain a down-sampled image block sequence corresponding to each image block.

[0025] An edge point number ratio is determined according to a proportion of edge pixels in each down-sampled image block in the down-sampled image block sequence corresponding to each image block to obtain an edge point number ratio sequence corresponding to each image block.

[0026] An edge gradient feature value is determined according to a mean value of gradient values corresponding to all edge pixels on the initial cardiovascular region in each down-sampled image block in the down-sampled image block sequence corresponding to each image block to obtain an edge gradient feature value sequence corresponding to each image block.

[0027] A straight line fitting is performed on the edge point number ratio sequence and the edge gradient feature value sequence corresponding to each image block respectively to obtain an edge point number ratio change straight line and an edge gradient feature change straight line corresponding to each image block.

[0028] The edge information loss degree corresponding to each image block is determined according to a slope of the edge point number ratio change straight line and a slope of the edge gradient feature change straight line.

[0029] In a possible implementation manner of the first aspect, the feature points in each image block are obtained by:

[0030] The Harris corner point detection algorithm is adopted to obtain a corner point of the initial cardiovascular region in each image block as the feature point in each image block.

[0031] In a possible implementation manner of the first aspect, the determining of the overall abnormal motion index corresponding to each image block based on the pre-acquired cardiovascular suspected skeleton corresponding to each image block and the motion condition of the feature points in each image block comprises:

[0032] screening, from all the cardiovascular images, the pixel points matched with the same feature point to form a motion trajectory point sequence corresponding to the feature point;

[0033] performing a morphological thinning operation on the initial cardiovascular region in each image block to obtain a cardiovascular suspected skeleton corresponding to each image block;

[0034] determining a target distance and a target direction between each adjacent pixel point in the motion trajectory point sequence corresponding to each feature point according to the positions of the adjacent pixel points in the motion trajectory point sequence corresponding to each feature point;

[0035] determining a blood flow direction inconsistency corresponding to each feature point according to the target direction between the adjacent pixel points in the motion trajectory point sequence corresponding to each feature point and the cardiovascular suspected skeleton corresponding to the image block to which each feature point belongs;

[0036] determining a blood flow velocity fluctuation corresponding to each feature point according to the target distance between the adjacent pixel points in the motion trajectory point sequence corresponding to each feature point;

[0037] determining an overall abnormal motion index corresponding to each image block according to the blood flow direction inconsistency and the blood flow velocity fluctuation corresponding to all the feature points in each image block.

[0038] In a possible implementation manner of the first aspect, the determining of the target distance and the target direction between each adjacent pixel point in the motion trajectory point sequence corresponding to each feature point according to the positions of the adjacent pixel points in the motion trajectory point sequence corresponding to each feature point comprises:

[0039] determining any one image block as a marker image block and determining any one feature point in the marker image block as a marker feature point;

[0040] determining any two adjacent pixel points in the motion trajectory point sequence corresponding to the marker feature point as a first to-be-detected point and a second to-be-detected point respectively;

[0041] screening, from a cardiovascular image to which the second to-be-detected point belongs, a pixel point having the same position as the first to-be-detected point as a reference pixel point;

[0042] determining a distance between the second to-be-detected point and the reference pixel point as a target distance between the first to-be-detected point and the second to-be-detected point;

[0043] A direction from the reference pixel point to the second to-be-detected point is determined as a target direction between the first to-be-detected point and the second to-be-detected point.

[0044] In a possible implementation manner of the first aspect, the blood flow direction inconsistency of each feature point is determined according to a target direction between adjacent pixel points in the motion trajectory point sequence corresponding to each feature point and a cardiovascular suspected skeleton corresponding to the image block to which each feature point belongs.

[0045] An arbitrary image block is determined as a marker image block, and an arbitrary feature point in the marker image block is determined as a marker feature point.

[0046] A pixel point closest to each pixel point in the motion trajectory point sequence corresponding to the marker feature point is filtered out from a cardiovascular suspected skeleton corresponding to the image block to which the marker feature point belongs, as a reference skeleton point corresponding to each pixel point in the motion trajectory point sequence corresponding to the marker feature point.

[0047] A direction between reference skeleton points corresponding to each adjacent pixel point in the motion trajectory point sequence corresponding to the marker feature point is determined as a reference direction between each adjacent pixel point in the motion trajectory point sequence corresponding to the marker feature point.

[0048] An included angle between a target direction and the reference direction between each adjacent pixel point in the motion trajectory point sequence corresponding to the marker feature point is determined as a target included angle between each adjacent pixel point in the motion trajectory point sequence corresponding to the marker feature point.

[0049] A blood flow direction inconsistency of the marker feature point is determined according to a mean value of the target included angles between all adjacent pixel points in the motion trajectory point sequence corresponding to the marker feature point and a mean value of the target included angles between the marker feature point and adjacent pixel points in the motion trajectory point sequence corresponding to the marker feature point.

[0050] In a possible implementation manner of the first aspect, the target cardiovascular region in each frame of cardiovascular image is identified by the cardiovascular recognition network according to the overall edge irregularity, the edge information loss degree and the overall abnormal motion index of the image block corresponding to each frame of cardiovascular image.

[0051] A target weight corresponding to each image block is determined according to the overall edge irregularity, the edge information loss degree and the overall abnormal motion index corresponding to each image block.

[0052] A weighted mean square error loss function corresponding to each frame of cardiovascular image is constructed according to the target weights corresponding to all image blocks in each frame of cardiovascular image.

[0053] A loss function corresponding to each frame of cardiovascular image is taken as a loss function corresponding to the cardiovascular image, and a target cardiovascular region in each frame of cardiovascular image is identified through a cardiovascular recognition network that is pre-trained.

[0054] In a second aspect, the present application provides a cardiovascular image auxiliary processing method for mobile device remote evaluation implemented by a cardiovascular image auxiliary processing system for mobile device remote evaluation, which comprises:

[0055] A preset number of frames of cardiovascular images corresponding to a patient to be detected are acquired, and each frame of cardiovascular image is equally divided to obtain image blocks;

[0056] An initial cardiovascular region is screened from each image block, and a global edge irregularity corresponding to each image block is determined based on a difference between the initial cardiovascular region in each image block before and after a morphological opening operation, and an included angle between a gradient direction and a normal direction corresponding to an edge pixel point of the initial cardiovascular region in each image block;

[0057] An edge information loss analysis is performed according to a down-sampling result of each image block to obtain an edge information loss degree corresponding to each image block;

[0058] Feature points in each image block are acquired, and a global abnormal motion index corresponding to each image block is determined based on a pre-acquired cardiovascular suspected skeleton corresponding to each image block and a motion condition corresponding to the feature points in each image block;

[0059] A target cardiovascular region is identified from each frame of cardiovascular image through a cardiovascular recognition network according to the global edge irregularity, the edge information loss degree and the global abnormal motion index corresponding to the image blocks in each frame of cardiovascular image.

[0060] In a third aspect, a server is provided, which comprises a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes the above-mentioned cardiovascular image auxiliary processing method for mobile device remote evaluation.

[0061] In a fourth aspect, a computer program product is provided, which comprises computer program code, which, when running on a computer, causes the computer to execute the above-mentioned cardiovascular image auxiliary processing method for mobile device remote evaluation.

[0062] In a fifth aspect, a computer readable storage medium is provided, which stores computer program code, which, when running on a computer, causes the computer to execute the above-mentioned cardiovascular image auxiliary processing method for mobile device remote evaluation.

[0063] The present application has the following beneficial effects:

[0064] The cardiovascular image auxiliary processing system for mobile device remote evaluation of the present application realizes cardiovascular region identification by analyzing multiple frames of cardiovascular images, solves the technical problem that the accuracy of cardiovascular region identification of cardiovascular images is poor, and improves the accuracy of cardiovascular region identification of cardiovascular images. Specifically, in actual situations, the edge of the cardiovascular region is smooth under normal circumstances, but motion artifacts generated during the imaging process often interfere with the clarity of the edge, making it irregular, thereby affecting the accurate segmentation of the cardiovascular region. Therefore, the present application quantifies the overall edge irregularity representing the regularity of the edge of the initial cardiovascular region by analyzing the difference between the initial cardiovascular region before and after the morphological opening operation, and the included angle between the gradient direction and the normal direction corresponding to the edge pixel points of the initial cardiovascular region, which can reduce the interference of motion artifacts on the segmentation of the cardiovascular region to a certain extent. Secondly, the present application also quantifies multiple indexes related to the accuracy of cardiovascular region segmentation, such as the degree of edge information loss and the overall abnormal motion index, and based on the overall edge irregularity, the degree of edge information loss and the overall abnormal motion index, the target cardiovascular region is relatively objectively identified from each frame of cardiovascular image, thereby improving the accuracy of cardiovascular region identification of cardiovascular images. BRIEF DESCRIPTION OF DRAWINGS

[0065] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0066] Figure 1 The structure diagram of the cardiovascular image auxiliary processing system for mobile device remote evaluation of the present application;

[0067] Figure 2 The flowchart of the cardiovascular image auxiliary processing method for mobile device remote evaluation of the present application;

[0068] Figure 3 The structure diagram of the computer device of the present application. DETAILED DESCRIPTION

[0069] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined inventive objectives, the specific implementation, structure, features and effects of the technical solutions according to the present application are described in detail below in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0070] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0071] In the cardiovascular image auxiliary processing, the cardiovascular segmentation technology can accurately identify and extract the cardiovascular structure from the cardiovascular image, which is crucial for the diagnosis and treatment of cardiovascular diseases such as remote evaluation of vascular stenosis degree by mobile devices. However, due to the involuntary movement of the patient or the inability to fully cooperate with the breath-holding instruction, it may cause the cardiovascular image to be blurred or ghosted, that is, motion artifacts are generated, wherein the involuntary movement can include: breathing and heartbeat, and displacement artifacts can also occur when adjacent parts of the anatomical structure are displaced relative to each other, making the tissue appear discontinuous, and even when there are metal objects in the patient's body that interfere with the penetration of X-rays, bright spots or shadows on the image, which can cause inaccurate segmentation of the cardiovascular region, thereby increasing the difficulty for doctors to perform remote reading on mobile devices. Among them, the metal objects present in the patient's body can include: braces and surgical implants. Therefore, the present application provides a mobile device remote reading cardiovascular image auxiliary processing system, which realizes accurate identification of the cardiovascular region by analyzing multiple frames of cardiovascular images.

[0072] Reference Figure 1 , shows a structural schematic diagram of a mobile device remote reading cardiovascular image auxiliary processing system according to the present application. The mobile device remote reading cardiovascular image auxiliary processing system comprises:

[0073] The acquisition and division module 101 is configured to acquire a preset number of frames of cardiovascular images corresponding to a to-be-detected patient, and divide each frame of cardiovascular image to obtain an image block.

[0074] Among them, the to-be-detected patient can be a patient to be subjected to cardiovascular identification. The preset number can be a pre-set number, which can be 20. The cardiovascular image can be an image of the cardiovascular of the to-be-detected patient. The image block can be a rectangular region obtained by dividing the cardiovascular image.

[0075] As an example, the acquisition and division module 101 can specifically implement the following steps:

[0076] In the first step, a plurality of frames of cardiovascular images of a patient to be detected can be acquired by MRI (Magnetic Resonance Imaging) scanning.

[0077] It should be noted that the continuous frames of cardiovascular images of the patient are acquired to capture the blood flow of the cardiovascular. These continuous frames of cardiovascular images can provide detailed information about the blood flow dynamics, which can assist doctors in diagnosing heart diseases and evaluating treatments. These continuous frames of cardiovascular images can be transmitted to a remote server or a cloud platform through a high-speed communication network.

[0078] In the second step, each frame of cardiovascular images can be equally divided into 100 rectangular regions, and each rectangular region obtained at this time can be taken as an image block.

[0079] The screening and determination module 102 is configured to screen an initial cardiovascular region from each image block, and determine the overall edge irregularity corresponding to each image block based on the difference between the initial cardiovascular region in each image block before and after a morphological opening operation, and the included angle between the gradient direction and the normal direction corresponding to the edge pixel points of the initial cardiovascular region in each image block.

[0080] It should be noted that the edge of the cardiovascular is smooth under normal circumstances, but cardiovascular diseases often cause thickening of the blood vessel wall, plaque formation or changes in the structure of the heart, so that the edge becomes irregular. Inflammation or tissue damage can also cause irregularity of the edge, and artifacts generated during the imaging process often interfere with the clarity of the edge, making it irregular. The cardiovascular diseases can include atherosclerosis and myocardial hypertrophy, etc. The artifacts generated during the imaging process can include motion artifacts and metal artifacts, etc.

[0081] As an example, the screening and determination module 102 can implement the following steps:

[0082] In the first step, the Otsu algorithm is used to perform threshold segmentation on each image block to obtain two sub-regions corresponding to each image block, and the sub-region with smaller gray value in the two sub-regions corresponding to each image block is determined as an initial cardiovascular region.

[0083] In the second step, a morphological opening operation is performed on each initial cardiovascular region to obtain a reference region corresponding to each initial cardiovascular region.

[0084] The reference region corresponding to the initial cardiovascular region can be the region after the morphological opening operation of the initial cardiovascular region.

[0085] It should be noted that the morphological opening operation on the region can include: performing the operation of first erosion and then inflation on the binary image corresponding to the region to obtain the binary image after the opening operation, so as to obtain the region after the morphological opening operation.

[0086] In the third step, the edge smoothing factor corresponding to each image block can be determined according to the overlap difference between the initial cardiovascular region in each image block and the reference region corresponding thereto, and can include the following sub-steps:

[0087] In the first sub-step, the number of pixel points in the intersection of the initial cardiovascular region in each image block and the reference region corresponding thereto is determined as the target intersection quantity corresponding to each image block.

[0088] In the second sub-step, the number of pixel points in the union of the initial cardiovascular region in each image block and the reference region corresponding thereto is determined as the target union quantity corresponding to each image block.

[0089] In the third sub-step, the ratio of the target intersection quantity to the target union quantity corresponding to each image block is determined as the edge smoothing factor corresponding to each image block.

[0090] For example, the formula for determining the edge smoothing factor corresponding to each image block can be:

[0091] ;

[0092] Wherein, C is the edge smoothing factor corresponding to the image block. is the target intersection quantity corresponding to the image block, that is, the number of pixel points in the intersection of the initial cardiovascular region in the image block and the reference region corresponding thereto. is the target union quantity corresponding to the image block, that is, the number of pixel points in the union of the initial cardiovascular region in the image block and the reference region corresponding thereto.

[0093] It should be noted that when C is larger, it often indicates that the overlap between the initial cardiovascular region in the image block and the region after the morphological opening operation is larger, it often indicates that the display of the cardiovascular edge in the image block is more likely to be smooth, it often indicates that the position of the cardiovascular region boundary before and after the opening operation changes less, it often indicates that the edge of the initial cardiovascular region in the image block is more likely to be smooth, and it often indicates that the edge of the initial cardiovascular region in the image block is more likely to be regular.

[0094] In the fourth step, the direction irregularity corresponding to each image block can be determined according to the included angle between the gradient direction and the normal direction of all edge pixel points on the initial cardiovascular region in each image block, and can include the following sub-steps:

[0095] The first sub-step is to normalize the included angle between the gradient direction and the normal direction corresponding to each edge pixel point on the initial cardiovascular region in each image block to obtain the local direction difference corresponding to each edge pixel point on the initial cardiovascular region in each image block.

[0096] The gradient direction corresponding to the edge pixel point can be the direction of the gradient vector of the edge pixel point. The normal direction corresponding to the edge pixel point can be the direction of the normal to the edge line to which the edge pixel point belongs. The included angle between different directions can correspond to a value range of [0°, 180°].

[0097] The second sub-step is to determine the direction irregularity corresponding to each image block as the mean value of the local direction differences corresponding to all edge pixel points on the initial cardiovascular region in each image block.

[0098] For example, the formula for determining the direction irregularity corresponding to the image block can be:

[0099] ;

[0100] wherein, D is the direction irregularity corresponding to the image block. N is the number of edge pixel points on the initial cardiovascular region in the image block. i is the serial number of the edge pixel point on the initial cardiovascular region in the image block. is the included angle between the gradient direction and the normal direction corresponding to the edge pixel point on the initial cardiovascular region in the image block. i is the maximum value of the value range of the included angle between different directions, which can be 180°, and is mainly used for normalization. The value range of the included angle between different directions can be [0°, 180°]. is the local direction difference corresponding to the edge pixel point on the initial cardiovascular region in the image block. is the serial number of the edge pixel point on the initial cardiovascular region in the image block. i is the local direction difference corresponding to the edge pixel point on the initial cardiovascular region in the image block.

[0101] It should be noted that, under normal circumstances, the gradient direction of the pixel point on the cardiovascular edge line should be perpendicular to the edge line, i.e., the gradient direction of the pixel point on the cardiovascular edge line is often the same as the normal direction of the pixel point on the cardiovascular edge line. However, when a lesion or artifact exists, it often makes the cardiovascular edge irregular, and often makes the gradient direction of the pixel point on the cardiovascular edge line different from the normal direction. Therefore, when D is larger, it often means that the gradient direction and the normal direction corresponding to the edge pixel point on the initial cardiovascular region in the image block are more likely to be different, and it often means that the initial cardiovascular region in the image block is more likely to have a lesion or artifact.

[0102] In the fifth step, the overall edge irregularity corresponding to each image block is determined according to the edge smoothness factor and the directional irregularity corresponding to each image block.

[0103] For example, the formula for determining the overall edge irregularity corresponding to each image block can be:

[0104] ;

[0105] wherein, S is the overall edge irregularity corresponding to the image block. C is the edge smoothness factor corresponding to the image block. D is the directional irregularity corresponding to the image block.

[0106] It should be noted that, when C is smaller, it often means that the edge of the initial cardiovascular region in the image block is more likely to be smoother, and it often means that the edge of the initial cardiovascular region in the image block is more likely to be more regular. When D is larger, it often means that the gradient direction corresponding to the edge pixel point on the initial cardiovascular region in the image block is more likely to be different from the normal direction, and it often means that the initial cardiovascular region in the image block is more likely to have lesions or artifacts. Therefore, when S is larger, it often means that the edge of the initial cardiovascular region in the image block is more likely to be irregular, it often means that the initial cardiovascular region in the image block is more likely to have lesions or artifacts, and it often means that the cardiovascular edge in the image block is more likely to be disturbed to a greater extent, and it should be given a higher weight in the loss function when subsequently recognized by the neural network. In the network segmentation process, the cross-entropy loss function of the image block with a greater degree of disturbance to the cardiovascular edge needs to be given a higher weight, so as to encourage the model to pay more attention to maintaining the clear boundary of the cardiovascular region during segmentation.

[0107] The edge information loss analysis module 103 is configured to analyze the edge information loss of each image block according to the down-sampling result of each image block, and obtain the edge information loss degree corresponding to each image block.

[0108] It should be noted that the method for subsequent cardiovascular recognition of the embodiment of the present application can be recognition by a neural network, and the neural network used for recognition can be a deep learning network based on U-Net. Since the structure of the U-Net network includes an encoder (down-sampling path) and a decoder (up-sampling path). In the down-sampling process, if the cardiovascular edge information in the image block is lost seriously, it may lead to a decrease in the segmentation accuracy of the network. This is because edge information is crucial for image segmentation tasks, especially in medical image segmentation, edge information can help the network to more accurately distinguish different tissues and structures. Therefore, it is necessary to further analyze the loss of cardiovascular edge information of the image block in the down-sampling process.

[0109] As an example, determining the edge information loss degree corresponding to each image block can include the following steps:

[0110] Firstly, a pyramid down-sampling manner is adopted to down-sample each image block for a preset number of times to obtain a sequence of down-sampled image blocks corresponding to each image block.

[0111] The preset number can be a preset number, which can be 5. The down-sampled image block in the sequence of down-sampled image blocks corresponding to the image block can be the result obtained by down-sampling the image block. The down-sampling size corresponding to the down-sampled image block in the sequence of down-sampled image blocks can gradually decrease.

[0112] Secondly, the proportion of the initial cardiovascular region edge pixel in each down-sampled image block in the sequence of down-sampled image blocks corresponding to each image block is determined as the edge point number proportion to obtain a sequence of edge point number proportions corresponding to each image block.

[0113] For example, the formula corresponding to the edge point number proportion in the sequence of edge point number proportions corresponding to the image block can be:

[0114] ;

[0115] Wherein, is the edge point number proportion in the sequence of edge point number proportions corresponding to the image block. j is the number of initial cardiovascular region edge pixels in the down-sampled image block in the sequence of down-sampled image blocks corresponding to the image block. is the down-sampled image block in the sequence of down-sampled image blocks corresponding to the image block, that is, the region block obtained by down-sampling the image block for the th time. j is the number of initial cardiovascular region edge pixels in the down-sampled image block in the sequence of down-sampled image blocks corresponding to the image block. j is the down-sampled image block in the sequence of down-sampled image blocks corresponding to the image block, that is, the region block obtained by down-sampling the image block for the th time. j is the number of initial cardiovascular region edge pixels in the down-sampled image block in the sequence of down-sampled image blocks corresponding to the image block. is the number of pixels in the down-sampled image block in the sequence of down-sampled image blocks corresponding to the image block. j It should be noted that when the larger, the edge pixel proportion in the region block obtained by down-sampling the image block for the th time is relatively more.

[0116] Thirdly, the mean value of the gradient values corresponding to all edge pixels on the initial cardiovascular region in each down-sampled image block in the sequence of down-sampled image blocks corresponding to each image block is determined as the edge gradient feature value to obtain a sequence of edge gradient feature values corresponding to each image block. j

[0117]

[0118] ​​​For example, the formula for determining the edge gradient feature values ​​in the edge gradient feature value sequence corresponding to an image patch can be:

[0119] ;

[0120] in, It is the first in the sequence of edge gradient feature values ​​corresponding to the image patch. j Each edge gradient feature value. It is the first downsampled image block in the sequence corresponding to the image block. j The number of pixels at the upper edge of the initial cardiovascular region in each downsampled image block. a It is the first downsampled image block in the sequence corresponding to the image block. j The sequence number of the upper edge pixel of the initial cardiovascular region in each downsampled image block. It is the first downsampled image block in the sequence corresponding to the image block. j In the first downsampled image patch, the initial cardiovascular region on the first a The gradient value corresponding to each edge pixel.

[0121] It should be noted that when The larger the value, the more likely it is to indicate that the image patch has undergone the first... j The larger the gradient corresponding to the edge pixels within the region obtained by the second downsampling, the more likely it is that the image patch has undergone the first downsampling. j The edges of the regions obtained by the next downsampling are relatively more distinct.

[0122] The fourth step involves performing linear fitting on the edge point quantity ratio sequence and edge gradient feature value sequence corresponding to each image patch to obtain the linear changes in the edge point quantity ratio and edge gradient feature value for each image patch. This may include the following sub-steps:

[0123] The first sub-step involves fitting a straight line using the index of the edge point quantity ratio sequence in the image patch as the x-axis and the edge point quantity ratio sequence as the y-axis. The straight line obtained at this point is then denoted as the straight line of change in the edge point quantity ratio corresponding to the image patch.

[0124] The second sub-step involves fitting a straight line using the index of the edge gradient feature value in the edge gradient feature value sequence corresponding to the image patch as the x-axis and the edge gradient feature value in the edge gradient feature value sequence as the y-axis. The straight line obtained at this time is recorded as the edge gradient feature change line corresponding to the image patch.

[0125] The fifth step is to determine the degree of edge information loss for each image block based on the slope of the line showing the change in the proportion of edge points corresponding to each image block and the slope of the line showing the change in edge gradient features.

[0126] For example, the formula for determining the degree of loss of edge information corresponding to the image block can be:

[0127] ;

[0128] wherein, G is the degree of loss of edge information corresponding to the image block. is a normalization function. is an absolute value function. is the slope of the straight line of the change of the proportion of the number of edge points corresponding to the image block. is the slope of the straight line of the change of the edge gradient feature corresponding to the image block.

[0129] It should be noted that during the downsampling process of the image block, if the edge information loss is less, the elements in the edge point proportion sequence and the edge gradient feature value sequence are often more stable, at this time the straight line of the change of the proportion of the number of edge points and the straight line of the change of the edge gradient feature often present as a horizontal straight line, and the slope should be close to 0. Therefore, when G is greater, it often means that the edge information loss of the image block during the downsampling process is relatively more, and it is more necessary to give a greater weight to the cross-entropy loss function during the network segmentation process, so that the model pays more attention to this image block and ensures the segmentation accuracy.

[0130] The acquisition and determination module 104 is configured to acquire feature points in each image block, and determine an overall abnormal motion indicator corresponding to each image block based on a pre-acquired cardiovascular suspected skeleton corresponding to each image block and a motion condition corresponding to the feature points in each image block.

[0131] As an example, the acquisition and determination module 104 can specifically implement the following steps:

[0132] First, the Harris corner point detection algorithm is used to acquire the corner points of the initial cardiovascular region in each image block, which are denoted as the feature points in each image block.

[0133] Second, the pixel points matched with the same feature point are screened out from all the cardiovascular images to form a motion trajectory point sequence corresponding to the feature point.

[0134] The motion trajectory point sequence can be a time sequence, and the pixel points in the motion trajectory point sequence can represent the same part of the patient.

[0135] For example, any one image block can be determined as a marker image block, and any one feature point in the marker image block can be determined as a marker feature point. The optical flow method and the inverse optical flow method are used to screen out the pixel points matched with the marker feature point from all the cardiovascular images, and the motion trajectory point sequence is formed in the order of the acquisition time.

[0136] It should be noted that the optical flow method is a computer vision technique for estimating the motion pattern of a pixel point in an image sequence. In cardiovascular image analysis, the optical flow method can be used to analyze blood flow conditions by tracking feature points in the blood flow to estimate blood flow velocity and direction.

[0137] In the third step, morphological thinning operation is performed on the initial cardiovascular region in each image block, and the obtained skeleton is referred to as the cardiovascular suspected skeleton corresponding to each image block.

[0138] In the fourth step, the target distance and the target direction between each adjacent pixel point in the motion trajectory point sequence corresponding to each feature point are determined according to the positions of the adjacent pixel points.

[0139] It should be noted that the target direction between adjacent pixel points in the motion trajectory point sequence can represent the blood flow direction between the adjacent pixel points. Since the time length between adjacent pixel points in the motion trajectory point sequence is the same, the target distance between adjacent pixel points in the motion trajectory point sequence can represent the motion velocity between the adjacent pixel points to some extent.

[0140] For example, determining the target distance and the target direction between each adjacent pixel point in the motion trajectory point sequence corresponding to each feature point can include the following sub-steps:

[0141] In the first sub-step, any one image block is determined as a marker image block, and any one feature point in the marker image block is determined as a marker feature point.

[0142] In the second sub-step, any two adjacent pixel points in the motion trajectory point sequence corresponding to the marker feature point are determined as a first to-be-detected point and a second to-be-detected point, respectively.

[0143] In the third sub-step, a pixel point having the same position as the first to-be-detected point is selected from the cardiovascular image to which the second to-be-detected point belongs as a reference pixel point.

[0144] In the fourth sub-step, the distance between the second to-be-detected point and the reference pixel point is determined as the target distance between the first to-be-detected point and the second to-be-detected point.

[0145] In the fifth sub-step, the direction of the reference pixel point pointing to the second to-be-detected point is determined as the target direction between the first to-be-detected point and the second to-be-detected point.

[0146] In the fifth step, the blood flow direction inconsistency corresponding to each feature point can be determined according to the target direction between adjacent pixel points in the motion trajectory point sequence corresponding to each feature point and the cardiovascular suspected skeleton corresponding to the image block to which each feature point belongs, which can include the following sub-steps:

[0147] A first sub-step is to determine any one of the image blocks as a marker image block and any one of the feature points in the marker image block as a marker feature point.

[0148] A second sub-step is to screen out, from the cardiovascular suspected skeleton corresponding to the image block to which the marker feature point belongs, a pixel point closest to each pixel point in a motion trajectory point sequence corresponding to the marker feature point as a reference skeleton point corresponding to each pixel point in the motion trajectory point sequence corresponding to the marker feature point.

[0149] For example, any one of the pixel points in the motion trajectory point sequence corresponding to the marker feature point can be determined as a temporary pixel point, and a pixel point closest to the temporary pixel point can be screened out from the cardiovascular suspected skeleton corresponding to the image block to which the marker feature point belongs as a reference skeleton point corresponding to the temporary pixel point.

[0150] A third sub-step is to determine a direction between reference skeleton points corresponding to each adjacent pixel point in the motion trajectory point sequence corresponding to the marker feature point as a reference direction between the adjacent pixel points in the motion trajectory point sequence corresponding to the marker feature point.

[0151] The direction between the reference skeleton points corresponding to the adjacent two pixel points can be a direction in which the reference skeleton point corresponding to the pixel point with an earlier collection time of the two pixel points points to the reference skeleton point corresponding to the pixel point with a later collection time.

[0152] It should be noted that the reference direction between the adjacent pixel points in the motion trajectory point sequence can represent a blood vessel extension direction between the adjacent pixel points.

[0153] For example, if the motion trajectory point sequence is a time sequence, the direction between the reference skeleton point corresponding to the second pixel point and the reference skeleton point corresponding to the third pixel point in the motion trajectory point sequence can be a direction in which the reference skeleton point corresponding to the second pixel point points to the reference skeleton point corresponding to the third pixel point.

[0154] A fourth sub-step is to determine an included angle between a target direction and a reference direction between each adjacent pixel point in the motion trajectory point sequence corresponding to the marker feature point as a target included angle between the adjacent pixel points in the motion trajectory point sequence corresponding to the marker feature point.

[0155] The target included angle can correspond to a value range of [0°, 180°].

[0156] In the fifth sub-step, the blood flow direction inconsistency of the marked feature point is determined according to the average of the target angles between all adjacent pixel points in the motion trajectory point sequence corresponding to the marked feature point and the average of the target angles between the marked feature point and its adjacent pixel points in the motion trajectory point sequence corresponding to the marked feature point.

[0157] For example, the formula for determining the blood flow direction inconsistency of the marked feature point can be:

[0158] ;

[0159] ;

[0160] ;

[0161] wherein, Q is the blood flow direction inconsistency of the marked feature point. M is the average of the target angles between all adjacent pixel points in the motion trajectory point sequence corresponding to the marked feature point. m is the average of the target angles between the marked feature point and its adjacent pixel points in the motion trajectory point sequence corresponding to the marked feature point. is the maximum value of the angle range between different directions, which can be 180°, so can be 360°. Since has a value range of [0°, 360°], therefore, can be used for normalization. n is the number of pixel points in the motion trajectory point sequence corresponding to the marked feature point. b is the serial number of the pixel point in the motion trajectory point sequence corresponding to the marked feature point. is the target angle between the b th pixel point and the b +1th pixel point in the motion trajectory point sequence corresponding to the marked feature point; that is, the angle between and . is the target direction between the b th pixel point and the b +1th pixel point in the motion trajectory point sequence corresponding to the marked feature point. is the reference direction between the b th pixel point and the b +1th pixel point in the motion trajectory point sequence corresponding to the marked feature point. is the target angle between the marked feature point and its previous pixel point in the motion trajectory point sequence corresponding to the marked feature point. is the target angle between the marker feature point and the next pixel point in the motion trajectory point sequence corresponding to the marker feature point.

[0162] It should be noted that, under normal circumstances, blood flow should flow smoothly along the direction of the blood vessel. If the motion trajectory point shows that the direction of the blood flow changes suddenly, it may be a sign of blood vessel malformation or thrombosis. Therefore, when Q The greater the value of is, the more likely it is that the motion trajectory point sequence corresponding to the marker feature point represents the blood flow direction of the part, and the more likely it is that the blood flow direction near the marker feature point is inconsistent with the direction of the blood vessel; the more likely it is that the marker feature point has a lesion.

[0163] In the sixth step, the blood flow velocity fluctuation corresponding to each feature point is determined according to the target distance between adjacent pixel points in the motion trajectory point sequence corresponding to each feature point.

[0164] For example, any one image block is determined as a marker image block, and any one feature point in the marker image block is determined as a marker feature point. The formula for determining the blood flow velocity fluctuation corresponding to the marker feature point can be:

[0165] ;

[0166] ;

[0167] ;

[0168] wherein, L is the blood flow velocity fluctuation corresponding to the marker feature point. U is the blood flow velocity deviation corresponding to the marker feature point. is a normalization function. is the variance of the target distance between all adjacent pixel points in the motion trajectory point sequence corresponding to the marker feature point. n is the number of pixel points in the motion trajectory point sequence corresponding to the marker feature point. b is the serial number of the pixel point in the motion trajectory point sequence corresponding to the marker feature point. is an absolute value function. w is the mean value of the target distance between the marker feature point and its adjacent pixel points in the motion trajectory point sequence corresponding to the marker feature point. is the target distance between the b th pixel point and the b +1th pixel point in the motion trajectory point sequence corresponding to the marker feature point. is the target distance between the marker feature point and the previous pixel point in the motion trajectory point sequence corresponding to the marker feature point. is a target distance between the marked feature point and a next pixel point in the motion trajectory point sequence corresponding to the marked feature point pair.

[0169] It should be noted that the normal blood flow velocity will have certain fluctuations, but if the fluctuations are too severe or irregular, it may reflect abnormal heart function or reduced blood vessel elasticity. Since the time length between adjacent pixel points in the motion trajectory point sequence is the same, the target distance between adjacent pixel points in the motion trajectory point sequence can to some extent represent the motion speed between the adjacent pixel points. When L is larger, it often indicates that the motion speed of the marked feature point is relatively larger, and the fluctuation degree of the motion speed of the marked feature point is relatively larger, which often indicates that the marked feature point is more likely to have a lesion.

[0170] Step 7: Determine the overall abnormal motion index corresponding to each image block according to the blood flow direction inconsistency and blood flow speed fluctuation of all feature points corresponding to each image block.

[0171] For example, the formula corresponding to the overall abnormal motion index corresponding to the image block can be:

[0172] ;

[0173] Wherein, R is the overall abnormal motion index corresponding to the image block. F is the number of feature points in the image block. f is the serial number of the feature point in the image block. is the blood flow direction inconsistency corresponding to the f th feature point in the image block. is a normalization function. is the blood flow speed fluctuation corresponding to the f th feature point in the image block.

[0174] It should be noted that when R is larger, it often indicates that the blood flow direction in the image block is more likely to be inconsistent with the blood vessel direction, and the blood flow speed and its fluctuation degree in the image block are relatively larger, which often indicates that the image block is more likely to have a lesion and needs to be paid attention to.

[0175] The region identification module 105 is configured to identify a target cardiovascular region from each frame of cardiovascular image by the cardiovascular identification network according to the overall edge irregularity, the edge information loss degree and the overall abnormal motion index corresponding to the image block in each frame of cardiovascular image.

[0176] Wherein, the target cardiovascular region can be a relatively accurate cardiovascular region finally identified.

[0177] As an example, identifying the target cardiovascular region from each frame of cardiovascular image can include the following steps:

[0178] Firstly, determine the target weight corresponding to each image block according to the overall edge irregularity, edge information loss degree and overall abnormal motion index corresponding to each image block.

[0179] For example, the formula for determining the target weight corresponding to the image block can be:

[0180] ;

[0181] ;

[0182] wherein, T is the target weight corresponding to the image block. Y is the blood vessel abnormality risk factor corresponding to the image block. R is the overall abnormal motion index corresponding to the image block. S is the overall edge irregularity corresponding to the image block. G is the edge information loss degree corresponding to the image block. is the cumulative value of the blood vessel abnormality risk factor corresponding to all image blocks in the cardiovascular image to which the image block belongs.

[0183] It should be noted that when G is larger, it often means that the edge information loss in the down-sampling process of the image block is relatively more. When S is larger, it often means that the edge of the initial cardiovascular region in the image block is more likely to be irregular, it often means that the initial cardiovascular region in the image block is more likely to have lesions or artifacts, and it often means that the cardiovascular edge in the image block is more likely to be disturbed. When R is larger, it often means that the blood flow direction in the image block is more likely to be inconsistent with the blood vessel direction, and the blood flow speed and its fluctuation degree in the image block are relatively larger, and it often means that the image block is more likely to have lesions. Therefore, T can represent the weight that the image block needs to be concerned about.

[0184] Secondly, construct the weighted mean square error loss function corresponding to each frame of cardiovascular image according to the target weight corresponding to all image blocks in each frame of cardiovascular image.

[0185] Wherein, the target weight corresponding to the image block can be the weight given in the weighted mean square error loss function corresponding to the cardiovascular image to which it belongs.

[0186] Thirdly, identify the target cardiovascular region in each frame of cardiovascular image by the cardiovascular recognition network trained in advance, taking the weighted mean square error loss function corresponding to each frame of cardiovascular image as its corresponding loss function.

[0187] The cardiovascular recognition network can be a U-Net-based deep learning network, which can be used for cardiovascular region recognition. The U-Net-based deep learning network has shown excellent performance in the field of medical image segmentation, especially in cardiovascular image processing. The weighted cross-entropy loss function is a commonly used loss function in deep learning, which can help the model better learn the features of the minority class and improve the segmentation effect in the image segmentation task.

[0188] Optionally, the training process of the cardiovascular recognition network can include the following sub-steps:

[0189] The first sub-step is to build a U-Net-based deep learning network as the pre-training cardiovascular recognition network.

[0190] The second sub-step is to obtain the target weight corresponding to different image blocks in different historical cardiovascular images of different historical patients.

[0191] The historical patient can be a patient who has undergone cardiovascular region annotation, specifically in the corresponding historical cardiovascular image. The cardiovascular region annotation method can be a manual annotation method. The historical cardiovascular image can be a cardiovascular image that has been annotated with a cardiovascular region.

[0192] It should be noted that the method of obtaining the target weight corresponding to different image blocks in the historical cardiovascular image is the same as the method of obtaining the target weight corresponding to different image blocks in the corresponding cardiovascular image of the patient to be detected, which will not be repeated here.

[0193] Optionally, the cardiovascular region can be sent to the doctor's mobile device for remote review, which can include the following steps:

[0194] The first step is data transmission. Specifically, the segmentation result can be sent to the doctor through a secure electronic medical record system or a dedicated medical image transmission software. And ensure that the patient privacy protection regulations such as HIPAA (Health Insurance Portability and Accountability Act) are followed throughout the transmission process.

[0195] The second step is for the doctor to review. Specifically, the doctor first conducts a preliminary evaluation of the segmentation result to check for obvious errors or omissions. Then the doctor can carefully analyze the shape, size, location and other characteristics of the cardiovascular region, as well as its relationship with the surrounding tissues. If the patient has previous image records, the doctor will compare the current segmentation result with the historical images to assess the changes in the patient's condition.

[0196] Step 3, record the evaluation results, specifically, the doctor will write a detailed diagnosis report according to the evaluation results, record the observed abnormal conditions and possible pathological changes. The report will include the interpretation of the segmentation results, as well as the preliminary diagnosis and recommendations for the patient's condition.

[0197] Step 4, follow-up action, specifically, based on the evaluation results, the doctor may develop or adjust the treatment plan. If necessary, the doctor may recommend further examination or testing to obtain more information.

[0198] Step 5, patient communication, specifically, the doctor will discuss the evaluation results with the patient, explain their meaning, and discuss the subsequent treatment plan. The doctor may provide education about the condition and treatment options to help the patient make informed decisions.

[0199] Reference Figure 2 Based on the same inventive concept as the above method embodiment, the present application provides a mobile device remote evaluation of cardiovascular image auxiliary processing method, comprising the following steps:

[0200] Step S1, obtain a preset number of frames of cardiovascular images corresponding to the patient to be detected, and divide each frame of cardiovascular image into equal parts to obtain image blocks.

[0201] Step S2, screen out the initial cardiovascular region from each image block, and determine the overall edge irregularity corresponding to each image block based on the difference between the initial cardiovascular region in each image block before and after the morphological opening operation, and the included angle between the gradient direction and the normal direction corresponding to the edge pixel points of the initial cardiovascular region in each image block.

[0202] Step S3, analyze the edge information loss condition according to the down-sampling result of each image block to obtain the edge information loss degree corresponding to each image block.

[0203] Step S4, obtain feature points in each image block, and determine the overall abnormal motion index corresponding to each image block based on the pre-acquired cardiovascular suspected skeleton corresponding to each image block and the motion condition corresponding to the feature points in each image block.

[0204] Step S5, according to the overall edge irregularity, edge information loss degree and overall abnormal motion index corresponding to the image blocks in each frame of cardiovascular image, the target cardiovascular region is identified from each frame of cardiovascular image through the cardiovascular recognition network.

[0205] Figure 3 is a structural schematic diagram of a computer device provided by an embodiment of the present application. For example, Figure 3As shown, the computer device 300 comprises a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and running on the processor 302, wherein the processor 302 executes the computer program 303, so that the computer device can execute the aforementioned mobile device remote reading cardiovascular image auxiliary processing method.

[0206] Based on the same inventive concept as the above method embodiments, the present application provides a server comprising a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes the aforementioned mobile device remote reading cardiovascular image auxiliary processing method.

[0207] Based on the same inventive concept as the above method embodiments, the present application provides a computer program product comprising computer program code, which, when executed on a computer, causes the computer to execute the aforementioned mobile device remote reading cardiovascular image auxiliary processing method.

[0208] Based on the same inventive concept as the above method embodiments, the present application provides a computer readable storage medium storing computer program code, which, when executed on a computer, causes the computer to execute the aforementioned mobile device remote reading cardiovascular image auxiliary processing method.

[0209] In summary, the embodiments of the present application determine the weight of the image block by analyzing the degree of interference of the center blood vessel edge and the degree of loss of cardiovascular edge information in each image block, combined with the performance of the lesion characteristics in the image block, to construct a weighted mean square error loss function, which ensures that the network model pays more attention to the edge, texture and other detailed information of the blood vessels and lesion areas in the image when segmenting the cardiovascular region. This helps to improve the overall performance of the model, especially in the case where the evaluation indicators are closely related to these important areas, and the adaptive loss function. This flexibility allows the model to adapt to various image segmentation tasks and challenges. It not only improves the attention of the model to important features, but also enhances the robustness and accuracy of the model in processing complex images, so as to obtain accurate cardiovascular regions, which can assist doctors in remote reading on mobile devices.

[0210] The above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some technical features can be replaced by equivalent features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A cardiovascular image-assisted processing system for remote image interpretation using mobile devices, characterized in that, The system includes: The acquisition and division module is used to acquire a preset number of cardiovascular images corresponding to the patient to be tested, and to divide each cardiovascular image into equal parts to obtain image blocks; The filtering and determination module is used to filter out the initial cardiovascular region from each image patch, and determine the overall edge irregularity of each image patch based on the difference between the initial cardiovascular region before and after the morphological opening operation, and the angle between the gradient direction and the normal direction of the edge pixels of the initial cardiovascular region in each image patch. This includes: using the Otsu algorithm to perform threshold segmentation on each image patch to obtain two sub-regions for each image patch, and determining the sub-region with the smaller gray value among the two sub-regions as the initial cardiovascular region; performing a morphological opening operation on each initial cardiovascular region to obtain a reference region for each initial cardiovascular region; and determining the edge smoothing factor for each image patch based on the overlap difference between the initial cardiovascular region and its corresponding reference region, including the following sub-steps: determining the number of pixels in the intersection of the initial cardiovascular region and its corresponding reference region for each image patch as the target edge smoothing factor for each image patch. The number of intersections is determined by the number of pixels in the union of the initial cardiovascular region and its corresponding reference region in each image block, which is then used to determine the target union number for each image block. The ratio of the target intersection number to the target union number for each image block is used to determine the edge smoothing factor for each image block. The directional irregularity for each image block is determined based on the angle between the gradient direction and the normal direction of all edge pixels in the initial cardiovascular region of each image block. This involves the following sub-steps: normalizing the angle between the gradient direction and the normal direction of each edge pixel in the initial cardiovascular region of each image block to obtain the local directional difference of each edge pixel in the initial cardiovascular region of each image block; determining the mean of the local directional differences of all edge pixels in the initial cardiovascular region of each image block as the directional irregularity for each image block; and finally, determining the overall edge irregularity for each image block based on the edge smoothing factor and the directional irregularity. The edge information loss analysis module is used to analyze the edge information loss based on the downsampling results of each image block, and to obtain the degree of edge information loss for each image block. This includes: using a pyramid downsampling method, downsampling each image block a preset number of times to obtain a downsampled image block sequence for each image block; determining the proportion of edge pixels in the initial cardiovascular region of each downsampled image block as the edge point proportion, thus obtaining an edge point proportion sequence for each image block; determining the mean of the gradient values ​​of all edge pixels in the initial cardiovascular region of each downsampled image block as the edge gradient feature value, thus obtaining an edge gradient feature value sequence for each image block; performing linear fitting on the edge point proportion sequence and the edge gradient feature value sequence for each image block to obtain a linear curve showing the change in the edge point proportion and the edge gradient feature value for each image block; and determining the degree of edge information loss for each image block based on the slopes of the linear curves showing the change in the edge point proportion and the edge gradient feature value for each image block. The acquisition and determination module is used to acquire feature points in each image block and, based on the pre-acquired suspected cardiovascular skeleton corresponding to each image block and the motion status of the feature points in each image block, determine the overall abnormal motion index corresponding to each image block. This includes: using the Harris corner detection algorithm to acquire the corner points of the initial cardiovascular region in each image block, which are recorded as feature points in each image block; selecting pixels matching the same feature point from all cardiovascular images to form the motion trajectory point sequence corresponding to that feature point; performing morphological thinning on the initial cardiovascular region in each image block, and recording the resulting skeleton as the suspected cardiovascular skeleton corresponding to each image block; and determining the motion trajectory point corresponding to each feature point based on the position of each adjacent pixel in the motion trajectory point sequence. The target distance and target direction between each adjacent pixel in the sequence include the following sub-steps: any image block is designated as a marked image block, and any feature point in the marked image block is designated as a marked feature point; any two adjacent pixels in the motion trajectory point sequence corresponding to the marked feature point are designated as the first detection point and the second detection point, respectively; pixels with the same position as the first detection point are selected from the cardiovascular image to which the second detection point belongs, and used as reference pixels; the distance between the second detection point and the reference pixels is determined as the target distance between the first detection point and the second detection point; the direction from the reference pixels to the second detection point is determined as the target direction between the first detection point and the second detection point.Based on the target direction between adjacent pixels in the motion trajectory point sequence corresponding to each feature point, and the suspected cardiovascular skeleton corresponding to the image block to which each feature point belongs, the inconsistency in blood flow direction corresponding to each feature point is determined, including the following sub-steps: Any image block is designated as a marked image block, and any feature point in the marked image block is designated as a marked feature point. From the suspected cardiovascular skeleton corresponding to the image block to which the marked feature point belongs, the pixel closest to each pixel in the motion trajectory point sequence corresponding to the marked feature point is selected as the reference skeleton point corresponding to each pixel in the motion trajectory point sequence corresponding to the marked feature point. The direction between the reference skeleton points corresponding to each adjacent pixel in the motion trajectory point sequence corresponding to the marked feature point is determined as the direction between each adjacent pixel in the motion trajectory point sequence corresponding to the marked feature point. The reference direction is defined as the angle between the target direction and the reference direction between each adjacent pixel in the motion trajectory point sequence corresponding to the marked feature point. Based on the average of the target angles between all adjacent pixels in the motion trajectory point sequence corresponding to the marked feature point, and the average of the target angles between the marked feature point and its adjacent pixels in the motion trajectory point sequence corresponding to the marked feature point, the blood flow direction inconsistency corresponding to the marked feature point is determined. Based on the target distance between adjacent pixels in the motion trajectory point sequence corresponding to each feature point, the blood flow velocity fluctuation corresponding to each feature point is determined. Based on the blood flow direction inconsistency and blood flow velocity fluctuation corresponding to all feature points in each image block, the overall abnormal motion index corresponding to each image block is determined. The region recognition module is used to identify the target cardiovascular region from each frame of cardiovascular image based on the overall edge irregularity, edge information loss degree and overall abnormal motion index of the image block in each frame of cardiovascular image through the cardiovascular recognition network.

2. The cardiovascular image-assisted processing system for remote image interpretation using a mobile device according to claim 1, characterized in that, The process of identifying the target cardiovascular region from each frame of cardiovascular images using a cardiovascular recognition network, based on the overall edge irregularity, edge information loss degree, and overall abnormal motion indicators of image blocks in each frame of cardiovascular images, includes: The target weight for each image block is determined based on the overall edge irregularity, the degree of edge information loss, and the overall abnormal motion index corresponding to each image block. Based on the target weights corresponding to all image blocks in each frame of cardiovascular image, a weighted mean square error loss function is constructed for each frame of cardiovascular image. Using the weighted mean square error loss function corresponding to each frame of cardiovascular image as its corresponding loss function, the target cardiovascular region in each frame of cardiovascular image is identified through a pre-trained cardiovascular recognition network.

Images