Osteoporotic vertebral compression fracture-oriented MRI image edema recognition method

Abstract

The present application relates to the technical field of image analysis, in particular to an MRI image edema recognition method for osteoporotic vertebral compression fractures, which comprises: acquiring a target MRI image corresponding to an osteoporotic vertebral compression fracture of a target patient, and dividing the target MRI image; determining a characteristic performance index; determining a merging possibility index between each two fragment regions according to the characteristic performance index difference, the gray difference and the position difference between each two fragment regions; adaptively merging the fragment regions in the target MRI image; based on all the seed points screened out, performing region segmentation on the target MRI image through a watershed algorithm; and screening out an edema region. The present application adaptively performs region segmentation on the image, comprehensively considers the gray distribution, gradient distribution and shape rule condition when screening the edema region, realizes edema region recognition, and improves the accuracy of edema region recognition.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology, specifically to a method for identifying edema in MRI images of osteoporotic vertebral compression fractures. Background Technology

[0002] With the development of technology, the application of region recognition technology is becoming increasingly widespread. For example, it can be applied to the identification of edema regions in MRI images for osteoporotic vertebral compression fractures. Currently, the common method for region recognition is to identify the required region from the image based on differences in grayscale values.

[0003] However, when identifying edematous areas from MRI images of osteoporotic vertebral compression fractures based on different gray values, the following technical problems often arise: In reality, the grayscale difference between the edema area and its surrounding tissues is often not significant. Therefore, when identifying the edema area from MRI images of osteoporotic vertebral compression fractures, considering only the difference in grayscale values ​​may lead to misjudgment of edema pixels, resulting in poor accuracy in identifying the edema area from MRI images of osteoporotic vertebral compression fractures. Summary of the Invention

[0004] To address the technical problem of poor accuracy in identifying edema regions in MRI images of osteoporotic vertebral compression fractures, this invention proposes a method for identifying edema in MRI images of osteoporotic vertebral compression fractures.

[0005] In a first aspect, the present invention provides a method for identifying edema in MRI images of osteoporotic vertebral compression fractures, the method comprising: Obtain the target MRI image corresponding to the osteoporotic vertebral compression fracture of the target patient, and divide the target MRI image into fragment regions based on the gradient value and position of the pixels in the target MRI image; Based on the number of pixels in each fragment region, the grayscale distribution in each fragment region, and the number of pixels in the preset neighborhood corresponding to each pixel in each fragment region that belong to the same fragment region, the feature performance index corresponding to each fragment region is determined. Based on the differences in the characteristic performance indicators corresponding to each pair of fragment regions, as well as the differences in grayscale and position between each pair of fragment regions, the possible indicators for merging between each pair of fragment regions are determined. Based on the merging probability index between different fragmented regions, the fragmented regions in the target MRI image are adaptively merged to obtain candidate regions. Based on the distance between pixels within a candidate region and its boundary, seed points are selected from all candidate regions. Then, based on all the selected seed points, the target MRI image is segmented using the watershed algorithm to obtain the target region. Based on the grayscale distribution, gradient distribution, and shape regularity of the target area, edema areas are selected from all target areas.

[0006] In conjunction with the first aspect above, in one possible implementation, the step of dividing the target MRI image into fragmented regions based on the gradient values ​​and positions corresponding to pixels in the target MRI image includes: Calculate the Euclidean distance between every two pixels in the target MRI image based on the positions of every two pixels in the target MRI image; The target distance metric between each two pixels in the target MRI image is determined based on the Euclidean distance between each two pixels in the target MRI image and the absolute value of the difference between the gradient values ​​corresponding to each two pixels in the target MRI image. Based on the target distance metric between different pixels in the target MRI image, the pixels in the target MRI image are clustered, and the region where each cluster is located is determined as a fragment region.

[0007] In conjunction with the first aspect above, in one possible implementation, determining the feature performance index corresponding to each fragment region based on the number of pixels in each fragment region, the grayscale distribution within each fragment region, and the number of pixels in the preset neighborhood corresponding to each pixel in each fragment region belonging to the same fragment region includes: The regional representation level of each fragment region is determined based on the number of pixels in each fragment region, the average gray value of all pixels in each fragment region, and the average gray value of all pixels in the target MRI image. The continuity index for each fragment region is determined based on the number of pixels in the preset neighborhood that belong to the same fragment region. The product of the regional performance level and the continuity index corresponding to each fragment region is used to determine the feature performance index corresponding to each fragment region.

[0008] In conjunction with the first aspect above, in one possible implementation, determining the regional representation degree corresponding to each fragment region based on the number of pixels in each fragment region, the average grayscale value corresponding to all pixels in each fragment region, and the average grayscale value corresponding to all pixels in the target MRI image includes: Any fragment region in the target MRI image is identified as a marked fragment region, and the average gray value of all pixels in the marked fragment region is identified as the gray value factor representing the marker. The mean value of the gray values ​​corresponding to all pixels in the target MRI image is determined as the overall representative gray factor. The degree of regional representation corresponding to the marked fragment region is determined based on the number of pixels within the marked fragment region and the absolute value of the difference between the grayscale factor represented by the marker and the overall grayscale factor.

[0009] In conjunction with the first aspect above, in one possible implementation, determining the continuity index corresponding to each fragment region based on the number of pixels in a preset neighborhood belonging to the same fragment region for each pixel within each fragment region includes: Any fragment region in the target MRI image is identified as a marked fragment region, and each pixel within the marked fragment region is identified as a marked pixel. Pixels belonging to the marked fragment region are selected from the preset neighborhood of each marked pixel and used as consecutive pixels corresponding to each marked pixel. The percentage of all consecutive pixels corresponding to each marked pixel within the preset neighborhood of that marked pixel is determined as the percentage of consecutive pixels corresponding to each marked pixel. The average percentage of consecutive pixels corresponding to all marked pixels is determined as the continuity index corresponding to the marked fragment region.

[0010] In conjunction with the first aspect above, in one possible implementation, determining the possible merging indicators between two fragment regions based on the differences in feature performance indicators corresponding to each pair of fragment regions, as well as the grayscale differences and positional differences between each pair of fragment regions, includes: The Euclidean distance between the center points of every two fragment regions is defined as the positional difference between every two fragment regions; The grayscale difference between each pair of fragment regions is determined based on the grayscale values ​​of all pixels within each pair of fragment regions. Based on the positional and grayscale differences between each pair of fragment regions, the target difference representative index between each pair of fragment regions is determined; The overall similarity between each pair of fragment regions is determined based on the target difference representative index. Based on the ratio between the minimum and maximum values ​​of the characteristic performance indicators corresponding to each pair of fragment regions, and the overall similarity between each pair of fragment regions, the possible merging indicators between each pair of fragment regions are determined.

[0011] In conjunction with the first aspect above, in one possible implementation, determining the grayscale difference between two fragment regions based on the grayscale values ​​corresponding to all pixels within each two fragment regions includes: The mean of the gray values ​​of all pixels in each fragment region is determined as the gray-level representative factor of each fragment region. The standard deviation of the gray values ​​corresponding to all pixels in each fragment region is determined as the region gray-level dispersion factor for each fragment region. The grayscale difference between each pair of fragment regions is determined by the absolute value of the difference between the grayscale representative factors corresponding to each pair of fragment regions and the absolute value of the difference between the grayscale discrete factors corresponding to each pair of fragment regions.

[0012] In conjunction with the first aspect above, in one possible implementation, the adaptive merging of fragmented regions in the target MRI image based on merging probability indices between different fragmented regions to obtain candidate regions includes: Based on the possible merging indicators and location differences between different fragment regions, the fragment regions in the target MRI image are clustered, and the resulting clusters are recorded as candidate regions.

[0013] In conjunction with the first aspect above, in one possible implementation, the step of selecting seed points from all candidate regions based on the distance between pixels within the candidate region and the region boundary includes: Any candidate region is determined as a labeled candidate region, and a preset number of pixels with the largest distance from the boundary of the labeled candidate region are selected from the labeled candidate region as a preset number of seed points corresponding to the labeled candidate region, thereby obtaining the seed points corresponding to all candidate regions.

[0014] In conjunction with the first aspect above, in one possible implementation, the step of filtering out edema regions from all target regions based on the grayscale distribution, gradient distribution, and shape regularity of the target region includes: Based on the mean gray value of all pixels in each target area, the mean gradient value of all pixels in each target area, and the roundness of each target area, determine the possible edema index for each target area. If the edema potential index corresponding to the target area is greater than the preset edema threshold, then the target area is determined as an edema area.

[0015] In a second aspect, the present invention provides an MRI imaging edema recognition system for osteoporotic vertebral compression fractures, the system comprising: The image acquisition and region segmentation module is used to acquire the target MRI image corresponding to the osteoporotic vertebral compression fracture of the target patient, and to segment the target MRI image based on the gradient value and position of the pixel in the target MRI image to obtain the fragment region. The feature performance index determination module is used to determine the feature performance index corresponding to each fragment region based on the number of pixels in each fragment region, the grayscale distribution in each fragment region, and the number of pixels in the preset neighborhood corresponding to each pixel in each fragment region that belong to the same fragment region. The module for determining possible merging indicators is used to determine possible merging indicators between two fragment regions based on the differences in the characteristic performance indicators corresponding to each two fragment regions, as well as the differences in grayscale and position between each two fragment regions. The adaptive merging module is used to adaptively merge fragmented regions in the target MRI image based on the merging probability index between different fragmented regions to obtain candidate regions. The seed point selection and region segmentation module is used to select seed points from all candidate regions based on the distance between the pixels in the candidate region and the region boundary, and to perform region segmentation on the target MRI image based on all selected seed points using the watershed algorithm to obtain the target region. The edema region filtering module is used to filter out edema regions from all target regions based on the grayscale distribution, gradient distribution, and shape regularity of the target region.

[0016] Thirdly, a server is provided, including 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, causing the device to perform the methods of the first aspect or any possible implementation thereof.

[0017] Fourthly, a computer program product is provided, comprising: computer program code, which, when run on a computer, causes the computer to perform the methods described in the first aspect or any possible implementation thereof.

[0018] Fifthly, a computer-readable storage medium is provided that stores computer program code, which, when executed on a computer, causes the computer to perform the methods described in the first aspect or any possible implementation thereof.

[0019] The present invention has the following beneficial effects: The present invention provides a method for identifying edema in MRI images of osteoporotic vertebral compression fractures. This method adaptively segments the image and comprehensively considers grayscale distribution, gradient distribution, and shape regularity when screening edema regions, thereby achieving edema region identification. This solves the technical problem of poor accuracy in identifying edema regions in MRI images of osteoporotic vertebral compression fractures and improves the accuracy of edema region identification. Specifically, this invention obtains gradient-similar fragmented regions from target MRI images, and quantifies the feature performance indicators characterizing the fragmented regions by analyzing the pixel distribution and gray-level distribution within the fragmented regions. Based on the differences in feature performance indicators, gray-level differences, and positional differences between two fragmented regions, it quantifies the merging probability indicators characterizing the possibility that the two fragmented regions belong to the same structure, thereby adaptively merging fragmented regions in the target MRI image, and then selecting seed points. Based on all the selected seed points, the target MRI image is segmented using a watershed algorithm. Based on the gray-level distribution, gradient distribution, and shape regularity of the regions, edema regions are selected from the segmented regions, realizing the identification of edema regions and improving the accuracy of edema region identification. Attached Figure Description

[0020] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of an MRI imaging edema identification method for osteoporotic vertebral compression fractures according to the present invention; Figure 2 This is a schematic diagram of the composition and structure of an MRI imaging edema recognition system for osteoporotic vertebral compression fractures according to the present invention. Figure 3 This is a schematic diagram of the structure of a computer device according to the present invention. Detailed Implementation

[0022] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solution proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0023] 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 invention pertains.

[0024] Osteoporotic vertebral compression fractures (OVCF) are a common disease among middle-aged and elderly people. Accurately identifying the bone marrow edema area within the vertebral body corresponding to an acute fracture is crucial for differentiating between fresh and old fractures, developing interventional treatment plans (such as vertebroplasty), and evaluating treatment efficacy. Magnetic resonance imaging (MRI), especially STIR (Short Tau Inversion Recovery) images, has often become the preferred imaging method for clinical assessment of vertebral edema due to its high sensitivity to tissue edema. Traditional identification methods mainly rely on visual assessment and manual delineation by radiologists. This process is highly subjective, time-consuming, and laborious, and diagnostic consistency is easily affected by physician experience. Automatic identification and segmentation methods based on image processing can identify tissue areas relatively objectively. However, due to the complex and variable patterns and deep semantic information of bone marrow edema in MRI images, the identification accuracy and generalization ability are often insufficient when the boundaries are blurred, the edema is uneven, or other pathological changes are present.

[0025] The watershed algorithm often performs well in segmenting images with blurred edges or irregular shapes. Theoretically, it can delineate complex and continuous lesion boundaries by simulating the flooding process, and it does not rely on fixed shape templates, making it highly adaptable. However, the segmentation effect is often related to the selection of the initial seed point. If the seed point is not properly located, the segmentation result will often be under-segmented or the region merging will be incorrect. It may misclassify other high-signal structures in the vertebral body (such as blood vessels and some degenerated tissues) as edema, and it is difficult to distinguish the difference in edema between fresh and old fractures. Therefore, it is crucial to accurately obtain the seed point when using the watershed algorithm for region segmentation.

[0026] Edema is essentially a pathological response of the vertebral bone marrow tissue to microfractures, inflammation, or local blood supply disorders, manifesting as abnormal accumulation of fluid in the intertrabecular spaces. Identifying edema areas involves not only locating the injury itself but also accurately depicting the extent of secondary pathological changes caused by the injury.

[0027] The boundary between edema areas and surrounding normal vertebral bone marrow tissue is usually a gradual transition rather than a clear, sharp one, and edema often presents as multifocal, irregular island-like, or finger-like spread. Watershed algorithms typically assume that each connected region is grown from one or a few seed points; however, for complex edema morphologies, the number, location, and distribution of seed points directly affect the segmentation results. If seed points are placed in every suspected region, minute noise or texture variations may be misjudged as independent regions, directly leading to catastrophic oversegmentation; if seed points are insufficient or unevenly distributed, some edema areas may not be recognized at all (undersegmentation). Therefore, accurate seed point location is crucial when using the watershed algorithm for edema region identification.

[0028] refer to Figure 1 This document illustrates the flowchart of some embodiments of an MRI imaging edema identification method for osteoporotic vertebral compression fractures according to the present invention. The method includes the following steps: Step S1: Obtain the target MRI image corresponding to the osteoporotic vertebral compression fracture of the target patient, and divide the target MRI image into fragment regions based on the gradient value and position of the pixels in the target MRI image.

[0029] The target patient can be one undergoing edema identification. The target MRI image can be a STIR (Short Tau Inversion Recovery) image of an osteoporotic vertebral compression fracture in the target patient. STIR images are a type of MRI (Magnetic Resonance Imaging) image.

[0030] As an example, this step may include the following steps: The first step is to acquire T2-weighted images of the osteoporotic vertebral compression fractures of the target patient using a magnetic resonance imaging scanner, which are recorded as the target MRI images.

[0031] The second step is to calculate the Euclidean distance between every two pixels in the target MRI image based on their corresponding positions.

[0032] The third step is to determine the target distance metric between each two pixels in the target MRI image based on the Euclidean distance between each two pixels in the target MRI image and the absolute value of the difference between the gradient values ​​corresponding to each two pixels in the target MRI image.

[0033] For example, the formula for determining the target distance metric between different pixels in a target MRI image can be: ; in, It is the target distance metric between the i-th and j-th pixels in the target MRI image. i and j are the indexes of the different pixels in the target MRI image. It is the Euclidean distance between the i-th pixel and the j-th pixel in the target MRI image. It is a pre-set adjustment factor, mainly used to adjust the function range. It can be used to prevent the corresponding factor from taking the value of 0. It can be adjusted according to the actual situation. For example, its preferred range can be 0.5-1.5, and its typical value can be 1. In the embodiment of the present invention, it can take the value of 1. It is an absolute value function. It is the gradient value corresponding to the i-th pixel in the target MRI image. It is the gradient value corresponding to the j-th pixel in the target MRI image.

[0034] It should be noted that all variables in the embodiments of the present invention, especially the denominator, can be set with corresponding adjustment factors according to the actual situation to adjust their value range or prevent the denominator from being 0. The preset parameters in the embodiments of the present invention, such as... and what will be used later , , and The method for obtaining this information can be pre-set based on human experience.

[0035] Optionally, , , , and Another method for obtaining this information is to acquire the target MRI images corresponding to osteoporotic vertebral compression fractures in historical patients, and then have the edema areas marked by the doctor on these images. , , , and The value range is {0.1, 0.2, 0.3, ..., 1.9, 2}, meaning each can have 20 possible values, and... , , , and The value ranges are combined to obtain a total of 20×20×20×20×20 combinations. Based on each combination, steps S1 to S6 are executed to output the edema area under each combination, which is recorded as the assessed edema area corresponding to each combination. The combination whose assessed edema area is closest to the labeled edema area is recorded as the target combination for historical patients. The target combination can be further analyzed. , , , and The value of is used as an example in the embodiments of the present invention. , , , and The final value of . Among them, historical patients can be patients with known edema areas.

[0036] Optionally, , , , and Another method for obtaining this information is to acquire multiple target combinations from historical patients and then extract the relevant data from all target combinations. The average value is denoted as the mean value in this embodiment of the invention. The final value; in the combination of all targets The average value is denoted as the mean value in this embodiment of the invention. The final value; in the combination of all targets The average value is denoted as the mean value in this embodiment of the invention. The final value; in the combination of all targets The average value is denoted as the mean value in this embodiment of the invention. The final value; in the combination of all targets The average value is denoted as the mean value in this embodiment of the invention. The final value of .

[0037] It should be noted that when The smaller the value, the more similar the gradients of the i-th and j-th pixels are, and the more likely they are to be adjacent in position.

[0038] Optionally, the method for obtaining the target distance metric between the i-th pixel and the j-th pixel can also be: ... and Perform a weighted summation, and use the resulting weighted summation value as the target distance metric between the i-th pixel and the j-th pixel.

[0039] The fourth step involves using the target distance metric between different pixels in the target MRI image to cluster the pixels. This is done using the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm. Pixels with adjacent locations and similar gradients are grouped into the same cluster, and the region containing each resulting cluster is designated as a fragmented region. It should be noted that, depending on the specific circumstances, the DBSCAN algorithm can be replaced with other clustering algorithms to achieve the desired clustering.

[0040] In this context, a fragmented area often does not represent a complete edema or a complete bone, but rather a small piece of flesh with the same texture characteristics. For example, a fragmented area may represent the brightest part of the center of the edema, the slightly darker ring around the edge of the edema, or a piece of normal bone marrow.

[0041] In practice, the fragmented regions of this invention are like tiny "puzzle pieces". The significance of dividing the fragmented regions lies in constructing an "atomic unit" that is more stable than a single pixel but more refined than a complete organ. It is the foundation for subsequent precise merging, calculation of complex features and generation of high-quality seed points.

[0042] Step S2: Determine the feature performance index corresponding to each fragment region based on the number of pixels in each fragment region, the grayscale distribution in each fragment region, and the number of pixels in the preset neighborhood corresponding to each pixel in each fragment region that belong to the same fragment region.

[0043] The preset neighborhood can be a neighborhood that is set in advance according to the actual situation, such as an eight-neighbor neighborhood.

[0044] As an example, this step may include the following steps: The first step, determining the regional representation level of each fragment region based on the number of pixels in each fragment region, the average grayscale value of all pixels in each fragment region, and the average grayscale value of all pixels in the target MRI image, may include the following sub-steps: The first sub-step involves identifying any fragment region in the target MRI image as the marked fragment region, and determining the average gray value of all pixels within the marked fragment region as the representative gray factor.

[0045] The second sub-step involves determining the average gray value of all pixels in the target MRI image as the overall representative gray factor.

[0046] The third sub-step involves determining the regional representation level corresponding to the marked fragment region based on the number of pixels within the marked fragment region and the absolute value of the difference between the grayscale factor represented by the marked fragment and the grayscale factor represented by the whole.

[0047] For example, the formula for determining the degree of regional performance corresponding to the marked fragment region can be: ; Where q represents the representation level of the marked fragment region, and N represents the number of pixels within the marked fragment region. It is a pre-set adjustment factor, mainly used to adjust the function range. It can be used to prevent the corresponding factor from taking the value of 0. It can be adjusted according to the actual situation. For example, its preferred range can be 0.5-1.5, and its typical value can be 1. In the embodiment of the present invention, it can take the value of 1. It is an absolute value function. G is a marker representing the grayscale factor. It represents the overall grayscale factor.

[0048] It should be noted that when q is larger, it often indicates that there are more pixels in the marked fragment region, and the grayscale difference between the marked fragment region and the whole image may be greater. This often indicates that the marked fragment region may be more prominent in the image, and the marked fragment region may have a greater degree of representation.

[0049] The second step, determining the continuity index for each fragment region based on the number of pixels within a preset neighborhood that belong to the same fragment region, may include the following sub-steps: The first sub-step involves identifying any fragment region in the target MRI image as a labeled fragment region, and identifying each pixel within the labeled fragment region as a labeled pixel.

[0050] The second sub-step involves selecting pixels belonging to the aforementioned marked fragment region from the preset neighborhood of each marked pixel, and using these pixels as consecutive pixels corresponding to each marked pixel.

[0051] The third sub-step is to determine the percentage of all consecutive pixels corresponding to each marked pixel within the preset neighborhood of that marked pixel, thus defining the percentage of consecutive pixels corresponding to each marked pixel.

[0052] For example, the formula for determining the proportion of consecutive pixels corresponding to a marked pixel can be: ; in, It represents the percentage of consecutive pixels corresponding to the a-th marked pixel. 'a' is the index of the marked pixel. It is the number of consecutive pixels corresponding to the a-th marked pixel, which is the number of pixels belonging to the marked fragment region within the preset neighborhood corresponding to the a-th marked pixel. It is the number of pixels in the preset neighborhood corresponding to the a-th marked pixel.

[0053] It should be noted that, To a certain extent, this value can characterize whether the a-th marked pixel is surrounded by pixels within the marked fragment region. A larger value generally indicates that the a-th marked pixel is more likely to be surrounded by pixels within its fragment region. Conversely, when... The smaller the value, the more likely the a-th marked pixel is not surrounded by pixels in its own fragment region. Conversely, if it is surrounded by other fragment regions, the more likely the a-th marked pixel is to be interspersed with other fragment regions.

[0054] The fourth sub-step is to determine the average percentage of consecutive pixels corresponding to all marked pixels as the continuity index corresponding to the marked fragment region.

[0055] It should be noted that the smaller the continuity index corresponding to the marked fragment region, the more likely the marked fragment region is to overlap with other fragment regions.

[0056] The third step is to determine the characteristic performance index corresponding to each fragment region by multiplying the regional performance degree and the continuity index corresponding to each fragment region.

[0057] It should be noted that the characteristic performance index corresponding to the fragment region can characterize the performance level of the fragment region and its cross-distribution with other fragment regions.

[0058] Step S3: Based on the differences in the feature performance indicators corresponding to each pair of fragment regions, as well as the differences in grayscale and position between each pair of fragment regions, determine the possible merging indicators between each pair of fragment regions.

[0059] As an example, this step may include the following steps: The first step is to determine the Euclidean distance between the center points of every two fragment regions as the positional difference between the two fragment regions.

[0060] It should be noted that, since the fragmented regions in the embodiments of the present invention are clusters, the center point of the fragmented regions can be the cluster center of the clusters.

[0061] It should be noted that the smaller the positional difference between two fragment regions, the more likely the two fragment regions are to be adjacent.

[0062] The second step, determining the grayscale difference between two fragment regions based on the grayscale values ​​of all pixels within each fragment region, may include the following sub-steps: The first sub-step is to determine the mean of the gray values ​​of all pixels within each fragment region as the region gray-level representative factor for each fragment region.

[0063] The second sub-step involves determining the standard deviation of the gray values ​​corresponding to all pixels within each fragment region as the region gray-level dispersion factor for each fragment region.

[0064] The third sub-step is to determine the grayscale difference between each pair of fragment regions based on the absolute value of the difference between the grayscale representative factors corresponding to each pair of fragment regions and the absolute value of the difference between the grayscale discrete factors corresponding to each pair of fragment regions.

[0065] For example, the formula for determining the grayscale difference between different fragment regions can be: ; in, It represents the grayscale difference between the m-th fragment region and the n-th fragment region. m and n are the indices of the different fragment regions. and These are all pre-set adjustment factors, mainly used to adjust the function range and prevent the corresponding factor from taking the value of 0. They can all be adjusted according to the actual situation. For example, their preferred range can be 0.5-1.5, and their typical value can be 1. In the embodiments of the present invention, they can all take the value of 1. It is an absolute value function. It is the gray-level representative factor corresponding to the m-th fragment region. It is the gray-level representative factor corresponding to the nth fragment region. It is the gray-level discretization factor corresponding to the m-th fragment region. It is the gray-level dispersion factor corresponding to the nth fragment region.

[0066] It should be noted that when The smaller the value, the more similar the grayscale distribution is between the m-th fragment region and the n-th fragment region.

[0067] Alternatively, the method for obtaining the grayscale difference between the m-th fragment region and the n-th fragment region can also be: [The method can be described in the original text]. and Perform a weighted summation, and determine the resulting weighted summation value as the grayscale difference between the m-th fragment region and the n-th fragment region.

[0068] The third step is to determine the target difference representative index between each pair of fragment regions based on the positional and grayscale differences between each pair of fragment regions.

[0069] For example, the positional difference and grayscale difference between every two fragment regions can be weighted and summed to obtain the target difference representative index between every two fragment regions. The weights can be set according to the actual situation. For example, the weight of positional difference can be 0.4 and the weight of grayscale difference can be 0.6.

[0070] Optionally, the target difference representative index can also be obtained by multiplying the positional difference and grayscale difference between each two fragment regions as the target difference representative index between each two fragment regions.

[0071] It should be noted that the smaller the target difference index between two fragment regions, the more likely the two fragment regions are to be adjacent and have similar gray-scale distributions.

[0072] The fourth step is to determine the overall similarity between each pair of fragment regions based on the target difference representative index.

[0073] For example, the formula for determining the overall similarity between different fragment regions can be: ; in, It represents the overall similarity between the m-th fragment region and the n-th fragment region. m and n are the indices of the different fragment regions. It is a pre-set adjustment factor, mainly used to adjust the function range. It can be used to prevent the denominator from being 0. It can be adjusted according to the actual situation. For example, its preferred range can be 0.5-1.5, and its typical value can be 1. In the embodiments of the present invention, it can be 1. It is a representative index of the target difference between the m-th fragment region and the n-th fragment region.

[0074] It should be noted that when The larger the value, the more likely the m-th fragment region and the n-th fragment region are to be adjacent, and the more similar their grayscale distributions are.

[0075] The fifth step is to determine the possible merging indicators between each pair of fragmented regions based on the ratio between the minimum and maximum values ​​of the characteristic performance indicators corresponding to each pair of fragmented regions, as well as the overall similarity between each pair of fragmented regions.

[0076] In practice, image features representing adjacent positions of the same structure often exhibit a certain degree of transition and similarity. Therefore, based on this characteristic, different fragment regions representing the same structure can be merged.

[0077] For example, the formula for determining the merging potential index between different fragmented regions can be: ; in, It is a possible merging index between the m-th fragment region and the n-th fragment region. m and n are the indices of the different fragment regions. It is a normalization function, mainly used to achieve normalization, but it can also be replaced with other normalization functions depending on the actual situation. It represents the overall similarity between the m-th fragment region and the n-th fragment region. It is the minimum value between the feature performance index corresponding to the m-th fragment region and the feature performance index corresponding to the n-th fragment region. It is the maximum value between the feature performance index corresponding to the m-th fragment region and the feature performance index corresponding to the n-th fragment region.

[0078] It should be noted that when A larger value often indicates that the m-th fragment region and the n-th fragment region are likely to be adjacent and have similar grayscale distributions. The feature performance index corresponding to a fragment region can characterize the performance level of that fragment region and its cross-distribution with other fragment regions. To some extent, this can balance the differences in regional performance and avoid the erroneous merging of highly significant and low-significance regions. Furthermore... A larger value often indicates a greater similarity in the performance and cross-distribution between the m-th fragment region and the n-th fragment region. Therefore, when The larger the value, the more likely the m-th fragment region and the n-th fragment region are to represent the same structure.

[0079] Step S4: Based on the merging probability index between different fragment regions, adaptive merging of fragment regions in the target MRI image is performed to obtain candidate regions.

[0080] One candidate region often represents a potential edema area or a normal tissue area.

[0081] As an example, the DBSCAN algorithm can be used to cluster fragmented regions in the target MRI image based on the merging probability index and location differences between different fragmented regions. In this case, fragmented regions that are adjacent in location and have a high merging probability index can be grouped into the same cluster, and the resulting clusters are recorded as candidate regions. It should be noted that, depending on the specific circumstances, the DBSCAN algorithm can be replaced with other clustering algorithms to achieve the desired clustering.

[0082] Optionally, the method for obtaining candidate regions can also be as follows: if the merging probability index between two adjacent fragment regions is greater than a preset merging threshold, then these two adjacent fragment regions are spliced ​​together, and so on, to achieve the corresponding splicing between all fragment regions, and each region obtained by the final splicing is recorded as a candidate region. The preset merging threshold can be a threshold pre-set according to the actual situation, such as 0.7.

[0083] It should be noted that by adaptively merging fragmented regions, we can obtain edema regions and normal tissue regions with high confidence in 'coarse localization,' i.e., candidate regions, thereby selecting high-quality seed points. This effectively avoids, to some extent, the over-segmentation or under-segmentation problems caused by improper seed point selection (such as noise interference) when directly using the watershed algorithm. Subsequently, using these optimized seed points, combined with gradient information, we can perform accurate boundary segmentation, thus achieving a step-by-step optimization from coarse localization to fine segmentation.

[0084] Step S5: Based on the distance between the pixels in the candidate region and the region boundary, seed points are selected from all candidate regions. Based on all the selected seed points, the target MRI image is segmented using the watershed algorithm to obtain the target region.

[0085] As an example, this step may include the following steps: The first step is to identify any candidate region as a labeled candidate region, and then select a preset number of pixels from the labeled candidate region that have the largest distance from the boundary of the labeled candidate region as a preset number of seed points corresponding to the labeled candidate region, thereby obtaining the seed points corresponding to all candidate regions.

[0086] The preset quantity can be a quantity set in advance based on the actual situation, such as 1.

[0087] The second step is to segment the target MRI image based on all the selected seed points using the watershed algorithm, and then denote the segmented region as the target region.

[0088] Step S6: Based on the grayscale distribution, gradient distribution, and shape regularity of the target area, filter out the edema area from all target areas.

[0089] In this embodiment of the invention, the edema area often refers to the bone marrow edema area within the vertebral body.

[0090] As an example, this step may include the following steps: The first step is to determine the possible edema indicators for each target region based on the mean gray value of all pixels in each target region, the mean gradient value of all pixels in each target region, and the roundness of each target region.

[0091] The value of circularity typically ranges from 0 to 1, and the closer it is to 1, the more likely the region is to be a regular circle; the closer the circularity is to 0, the more likely the region is to be an irregular shape. Circularity can be characterized by the ratio between the area and the perimeter of the region.

[0092] For example, the formula for determining the possible edema indicators corresponding to the target area can be: ; in, This represents the possible edema index corresponding to the t-th target region. t is the index of the target region. , and They are , and The preset weights can be set according to the actual situation, such as... , and The values ​​can be 0.5, 0.3, and 0.2 respectively. , and It is normalized data. It is the average grayscale value of all pixels in the t-th target region. It is the maximum gray value corresponding to all pixels in the target MRI image. It is the mean of the gradient values ​​corresponding to all pixels in the t-th target region. It is the maximum value of the gradient corresponding to all pixels in the target MRI image. It is the circularity of the t-th target region.

[0093] It should be noted that, in this embodiment of the invention, the target MRI image is taken as an example, specifically a STIR image. Edema areas typically exhibit high signal intensity in STIR images, and due to the gradual transition between edema and normal tissue, their boundaries are often blurred. Finally, the morphology of the edema area is often irregular, frequently appearing as finger-like or multifocal diffusion. Therefore, when... A larger value often indicates that the gray level of the t-th target region may be higher, the gradient may be smaller, and its shape may be more irregular; it also often indicates that the t-th target region is more likely to be an edema region.

[0094] The second step is to determine the target area as an edema area if the edema potential index of the target area is greater than the preset edema threshold.

[0095] The preset edema threshold can be a threshold set in advance based on the actual situation, such as 0.65.

[0096] refer to Figure 2 Based on the same inventive concept as the above-described method embodiments, this invention provides an MRI image edema recognition system for osteoporotic vertebral compression fractures. The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements the steps of an MRI image edema recognition method for osteoporotic vertebral compression fractures, specifically including: The image acquisition and region segmentation module 201 is used to acquire the target MRI image corresponding to the osteoporotic vertebral compression fracture of the target patient, and to segment the target MRI image based on the gradient value and position of the pixel in the target MRI image to obtain the fragment region. The feature performance index determination module 202 is used to determine the feature performance index corresponding to each fragment region based on the number of pixels in each fragment region, the grayscale distribution in each fragment region, and the number of pixels in the preset neighborhood corresponding to each pixel in each fragment region that belong to the same fragment region. The merging possibility index determination module 203 is used to determine the merging possibility index between each two fragment regions based on the differences between the feature performance indicators corresponding to each two fragment regions, as well as the gray level differences and position differences between each two fragment regions. The adaptive merging module 204 is used to adaptively merge fragmented regions in the target MRI image based on the merging possibility index between different fragmented regions to obtain candidate regions. The seed point selection and region segmentation module 205 is used to select seed points from all candidate regions based on the distance between the pixels in the candidate region and the region boundary, and to perform region segmentation on the target MRI image based on all selected seed points using the watershed algorithm to obtain the target region. The edema region filtering module 206 is used to filter out edema regions from all target regions based on the grayscale distribution, gradient distribution, and shape regularity of the target region.

[0097] Figure 3This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. For example, as shown... Figure 3 As shown, the computer device 300 includes: a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and running on the processor 302, wherein when the processor 302 executes the computer program 303, the computer device can execute any of the aforementioned MRI image edema identification methods for osteoporotic vertebral compression fractures.

[0098] Based on the same inventive concept as the above-described method embodiments, the present invention provides a server, including 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, causing the device to execute any of the above-described methods for identifying edema in MRI images of osteoporotic vertebral compression fractures.

[0099] Based on the same inventive concept as the above-described method embodiments, the present invention provides a computer program product comprising: computer program code, which, when executed on a computer, causes the computer to perform any of the above-described methods for identifying edema in MRI images of osteoporotic vertebral compression fractures.

[0100] Based on the same inventive concept as the above-described method embodiments, the present invention provides a computer-readable storage medium storing computer program code, which, when executed on a computer, causes the computer to perform any of the above-described methods for identifying edema in MRI images of osteoporotic vertebral compression fractures.

[0101] In summary, this invention obtains gradient-similar fragmented regions from target MRI images. By analyzing the pixel and gray-level distribution within these fragmented regions, it quantifies the feature performance indicators characterizing the fragmented regions. Based on the differences in feature performance indicators, gray-level differences, and positional differences between two fragmented regions, it quantifies the merging probability indicators representing the likelihood that these two fragmented regions belong to the same structure. This allows for adaptive merging of fragmented regions in the target MRI image, thereby selecting seed points. Based on all selected seed points, the watershed algorithm is used to segment the target MRI image into regions. Based on the gray-level distribution, gradient distribution, and shape regularity of the regions, edema regions are selected from the segmented regions, achieving the identification of edema regions and improving the accuracy of edema region identification.

[0102] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for identifying edema in MRI images of osteoporotic vertebral compression fractures, characterized in that, Includes the following steps: Obtain the target MRI image corresponding to the osteoporotic vertebral compression fracture of the target patient, and divide the target MRI image into fragment regions based on the gradient value and position of the pixels in the target MRI image; Based on the number of pixels in each fragment region, the grayscale distribution in each fragment region, and the number of pixels in the preset neighborhood corresponding to each pixel in each fragment region that belong to the same fragment region, the feature performance index corresponding to each fragment region is determined. Based on the differences in the characteristic performance indicators corresponding to each pair of fragment regions, as well as the differences in grayscale and position between each pair of fragment regions, the possible indicators for merging between each pair of fragment regions are determined. Based on the merging probability index between different fragmented regions, the fragmented regions in the target MRI image are adaptively merged to obtain candidate regions. Based on the distance between pixels within a candidate region and its boundary, seed points are selected from all candidate regions. Then, based on all the selected seed points, the target MRI image is segmented using the watershed algorithm to obtain the target region. Based on the grayscale distribution, gradient distribution, and shape regularity of the target area, edema areas are selected from all target areas.

2. The method for identifying edema in MRI images of osteoporotic vertebral compression fractures according to claim 1, characterized in that, The method involves dividing the target MRI image based on the gradient values ​​and positions of pixels in the target MRI image to obtain fragmented regions, including: Calculate the Euclidean distance between every two pixels in the target MRI image based on the positions of every two pixels in the target MRI image; The target distance metric between each two pixels in the target MRI image is determined based on the Euclidean distance between each two pixels in the target MRI image and the absolute value of the difference between the gradient values ​​corresponding to each two pixels in the target MRI image. Based on the target distance metric between different pixels in the target MRI image, the pixels in the target MRI image are clustered, and the region where each cluster is located is determined as a fragment region.

3. The method for identifying edema in MRI images of osteoporotic vertebral compression fractures according to claim 1, characterized in that, The feature performance index corresponding to each fragment region is determined based on the number of pixels in each fragment region, the grayscale distribution in each fragment region, and the number of pixels in the preset neighborhood corresponding to each pixel in each fragment region that belong to the same fragment region. This includes: The regional representation level of each fragment region is determined based on the number of pixels in each fragment region, the average gray value of all pixels in each fragment region, and the average gray value of all pixels in the target MRI image. The continuity index for each fragment region is determined based on the number of pixels in the preset neighborhood that belong to the same fragment region. The product of the regional performance level and the continuity index corresponding to each fragment region is used to determine the feature performance index corresponding to each fragment region.

4. The method for identifying edema in MRI images of osteoporotic vertebral compression fractures according to claim 3, characterized in that, The determination of the regional representation degree corresponding to each fragment region based on the number of pixels in each fragment region, the average gray value of all pixels in each fragment region, and the average gray value of all pixels in the target MRI image includes: Any fragment region in the target MRI image is identified as a marked fragment region, and the average gray value of all pixels in the marked fragment region is identified as the gray value factor representing the marker. The mean value of the gray values ​​corresponding to all pixels in the target MRI image is determined as the overall representative gray factor. The degree of regional representation corresponding to the marked fragment region is determined based on the number of pixels within the marked fragment region and the absolute value of the difference between the grayscale factor represented by the marker and the overall grayscale factor.

5. The method for identifying edema in MRI images of osteoporotic vertebral compression fractures according to claim 3, characterized in that, The step of determining the continuity index corresponding to each fragment region based on the number of pixels in the preset neighborhood corresponding to a pixel within each fragment region that belong to the same fragment region includes: Any fragment region in the target MRI image is identified as a marked fragment region, and each pixel within the marked fragment region is identified as a marked pixel. Pixels belonging to the marked fragment region are selected from the preset neighborhood of each marked pixel and used as consecutive pixels corresponding to each marked pixel. The percentage of all consecutive pixels corresponding to each marked pixel within the preset neighborhood of that marked pixel is determined as the percentage of consecutive pixels corresponding to each marked pixel. The average percentage of consecutive pixels corresponding to all marked pixels is determined as the continuity index corresponding to the marked fragment region.

6. The method for identifying edema in MRI images of osteoporotic vertebral compression fractures according to claim 1, characterized in that, The step of determining possible merging indicators between two fragment regions based on the differences in feature performance indicators corresponding to each pair of fragment regions, as well as the differences in grayscale and position between each pair of fragment regions, includes: The Euclidean distance between the center points of every two fragment regions is defined as the positional difference between every two fragment regions; The grayscale difference between each two fragment regions is determined based on the grayscale values ​​of all pixels within each fragment region. Based on the positional and grayscale differences between each pair of fragment regions, the target difference representative index between each pair of fragment regions is determined; The overall similarity between each pair of fragment regions is determined based on the target difference representative index. Based on the ratio between the minimum and maximum values ​​of the characteristic performance indicators corresponding to each pair of fragment regions, and the overall similarity between each pair of fragment regions, the possible merging indicators between each pair of fragment regions are determined.

7. The method for identifying edema in MRI images of osteoporotic vertebral compression fractures according to claim 6, characterized in that, The step of determining the grayscale difference between two fragment regions based on the grayscale values ​​of all pixels within each two fragment regions includes: The mean of the gray values ​​of all pixels in each fragment region is determined as the gray-level representative factor of each fragment region. The standard deviation of the gray values ​​corresponding to all pixels in each fragment region is determined as the region gray-level dispersion factor for each fragment region. The grayscale difference between each pair of fragment regions is determined by the absolute value of the difference between the grayscale representative factors corresponding to each pair of fragment regions and the absolute value of the difference between the grayscale discrete factors corresponding to each pair of fragment regions.

8. The method for identifying edema in MRI images of osteoporotic vertebral compression fractures according to claim 1, characterized in that, The process of adaptively merging fragmented regions in the target MRI image based on merging probability indices between different fragmented regions to obtain candidate regions includes: Based on the merging potential indicators and location differences between different fragment regions, the fragment regions in the target MRI image are clustered, and the resulting clusters are denoted as candidate regions.

9. The method for identifying edema in MRI images of osteoporotic vertebral compression fractures according to claim 1, characterized in that, The step of selecting seed points from all candidate regions based on the distance between pixels within the candidate region and the region boundary includes: Any candidate region is determined as a labeled candidate region, and a preset number of pixels with the largest distance from the boundary of the labeled candidate region are selected from the labeled candidate region as a preset number of seed points corresponding to the labeled candidate region, thereby obtaining the seed points corresponding to all candidate regions.

10. The method for identifying edema in MRI images of osteoporotic vertebral compression fractures according to claim 1, characterized in that, The process of filtering out edema regions from all target regions based on their grayscale distribution, gradient distribution, and shape regularity includes: Based on the mean gray value of all pixels in each target area, the mean gradient value of all pixels in each target area, and the roundness of each target area, determine the possible edema index for each target area. If the edema potential index corresponding to the target area is greater than the preset edema threshold, then the target area is determined as an edema area.

Images