White matter lesion area intelligent positioning method based on MRI (Magnetic Resonance Imaging) image
By constructing a white matter network and adjusting the cluster affiliation of edge nodes, the problem of incorrect identification of white matter lesion areas in existing technologies has been solved, achieving higher accuracy and efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-23
- Publication Date
- 2026-03-24
AI Technical Summary
In existing technologies, when determining the area of white matter lesions based on the comparison between the patient's white matter network and the normal white matter network, the results are easily affected by the significant differences in brain structure between individuals, leading to errors in judgment.
A brain white matter network was constructed based on MRI images. By clustering the nodes, edge nodes were identified, and their affiliation to the next cluster was adjusted based on the differences and affiliation degree between the edge nodes and their neighboring clusters until the preset conditions were met, and the final adjustment result was determined.
It improves the accuracy of identifying white matter lesion areas, reduces missed and false diagnoses, and enhances the precision and efficiency of identifying white matter lesion areas.
Smart Images
Figure CN121724972A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of image data processing or generation technology, specifically to a method for intelligent localization of white matter lesion regions in the brain based on MRI images. Background Technology
[0002] The white matter is where nerve fibers gather inside the brain. Due to poisoning, genetic degeneration, infection, hydrocephalus, and other reasons, abnormal changes may occur in the white matter, which can damage the myelin sheath of central nerve cells, resulting in white matter lesions.
[0003] In related technologies, the lesion area of the white matter can be determined by comparing the relative spatial location of the patient's white matter network with that of the normal white matter network, as well as the results of the node clustering coefficient and node degree.
[0004] However, due to the significant differences in brain structure between individuals, the white matter network can also vary considerably. Therefore, determining the area of white matter lesions based on a comparison between the patient's white matter network and the normal white matter network may lead to incorrect identification of the white matter lesion area. Summary of the Invention
[0005] To address the technical problem that determining the region of white matter lesions based on comparisons between the patient's white matter network and the normal white matter network may lead to misjudgments, this application aims to provide an intelligent localization method for white matter lesions based on nuclear magnetic resonance imaging (MRI) images. The specific technical solution adopted is as follows: This application provides a method for intelligent localization of white matter lesion regions based on MRI images, the method comprising: A white matter network is constructed based on brain MRI images. This white matter network includes multiple nodes and edges between nodes. Nodes correspond to white matter aggregation points, and edges between nodes correspond to white matter regions. The nodes in the white matter network are clustered to obtain multiple clusters, each of which includes at least one edge node. Based on the differences and affiliations of each edge node with its corresponding neighboring clusters, the affiliation of each edge node is adjusted. The neighboring clusters of an edge node include the current affiliation of that edge node. If the adjustment results meet preset conditions, the final adjustment result is determined, which includes the adjusted nodes. The white matter aggregation points corresponding to the adjusted nodes are identified as white matter lesion regions.
[0006] Optionally, the white matter network described above includes the grayscale value of each node. The method further includes: determining the distance between a first edge node and the cluster center of a first adjacent cluster, and the distance between the first edge node and the center node of the first adjacent cluster, wherein the first edge node is any one of the at least one edge node, and the first adjacent cluster is any one of the adjacent clusters corresponding to the first edge node; and determining the degree of difference between the first edge node and the first adjacent cluster based on the distance between the first edge node and the cluster center of the first adjacent cluster, the distance between the first edge node and the center node of the first adjacent cluster, the grayscale value of the first edge node, and the grayscale value of the center node of the first adjacent cluster.
[0007] Optionally, the method further includes: determining the number and length of edges between the first edge node and nodes in the first adjacent cluster; and determining the degree of belonging of the first edge node to the first adjacent cluster based on the number and length of edges between the first edge node and nodes in the first adjacent cluster.
[0008] Optionally, based on the differences and affiliations between the first edge node and its corresponding neighboring clusters, the affiliation of the first edge node is adjusted, including: determining the adjustment requirement of the first edge node for each neighboring cluster based on the differences and affiliations between the first edge node and its corresponding neighboring clusters, wherein the adjustment requirement of an edge node for a neighboring cluster is used to characterize the matching degree between the edge node and the neighboring cluster; and adjusting the affiliation of the first edge node to the target neighboring cluster when the adjustment requirement of the target neighboring cluster is greater than or equal to the adjustment requirement threshold, wherein the target neighboring cluster is the neighboring cluster corresponding to the maximum adjustment requirement.
[0009] Optionally, based on the differences and affiliations between the first edge node and its corresponding adjacent clusters, the adjustment requirement of the first edge node to the second adjacent cluster is determined, including: based on the differences between the first edge node and its affiliation cluster, the affiliation degree between the first edge node and its affiliation cluster, the differences between the first edge node and the second adjacent cluster, and the affiliation degree between the first edge node and the second adjacent cluster, the adjustment requirement of the first edge node to the second adjacent cluster is determined, wherein the second adjacent cluster is the adjacent cluster of the first edge node that is not the affiliation cluster of the first edge node.
[0010] Optionally, the adjustment result includes the validity of this adjustment, and the preset condition includes that the validity of this adjustment is less than or equal to the validity threshold of the adjustment. The validity of this adjustment is used to characterize the appropriateness of the cluster to which the edge node belongs after this adjustment compared with the cluster to which the edge node belongs before the adjustment.
[0011] Optionally, the method further includes: determining the maximum adjustment requirement of each edge node that needs to be adjusted before the current adjustment, and determining the maximum adjustment requirement of each edge node that needs to be adjusted after the current adjustment; and determining the validity of the current adjustment based on the maximum adjustment requirement of each edge node that needs to be adjusted before the current adjustment and the maximum adjustment requirement of each edge node that needs to be adjusted after the current adjustment.
[0012] Optionally, determining the validity of the current adjustment based on the maximum adjustment requirement of each edge node that needs adjustment before the current adjustment and the maximum adjustment requirement of each edge node that needs adjustment after the current adjustment includes: determining the validity of the current adjustment based on the sum of the maximum adjustment requirements of each edge node that needs adjustment before the current adjustment and the sum of the maximum adjustment requirements of each edge node that needs adjustment after the current adjustment.
[0013] Optionally, determining the final adjustment result includes: determining the adjusted nodes based on the cluster to which each edge node belonged before this adjustment and the original cluster to which each edge node belonged.
[0014] Optionally, the method further includes: if the adjustment demand degree of the target adjacent cluster is less than the adjustment demand degree threshold, determining the belonging cluster of the first edge node as the current belonging cluster of the first edge node.
[0015] This application has the following beneficial effects: In this embodiment, a white matter network is constructed based on brain MRI images, and the nodes in the white matter network are clustered to obtain multiple clusters. Based on the difference and belonging degree between each edge node and its corresponding adjacent clusters, the belonging cluster of each edge node is adjusted. If the adjustment result meets the preset conditions, it means that the belonging cluster of the edge node has been adjusted to the optimal value. At this time, the adjusted node is identified as an edge node that is significantly different from other surrounding nodes and causes clustering errors. Since white matter lesions weaken white matter fibers in the lesion area and cause the edges between nodes to become blurred, the sense of belonging to each brain region becomes blurred. Therefore, the adjusted node is very likely to be a node that produces lesions. Identifying the white matter cluster point corresponding to the adjusted node as the white matter lesion area can improve the accuracy of the determination of the white matter lesion area. Attached Figure Description
[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 A flowchart illustrating an intelligent localization method for white matter lesions in the brain based on MRI images, provided as an embodiment of this application; Figure 2 This is a flowchart of another intelligent localization method for brain white matter lesion regions based on MRI images, provided as an embodiment of this application. Detailed Implementation
[0018] To further illustrate the technical means and effects adopted by this application to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for intelligent localization of white matter lesions based on MRI images proposed in this application. 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.
[0019] 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 pertains.
[0020] Early symptoms of white matter lesions are generally not obvious. In the early stages of white matter lesions, MRI has high sensitivity and can detect asymptomatic white matter lesions in brain regions. Therefore, head MRI is now commonly used for early detection of white matter lesions and for further diagnosis and treatment.
[0021] With the development of magnetic resonance imaging (MRI) technology for brain examination, diffusion tensor imaging (DTI) emerged. DTI is a special form of MRI and a new method for describing brain structure. When using DTI to detect and diagnose white matter lesions, the patient's white matter network is often divided and constructed based on DTI images, and then the lesion area is analyzed and located based on the obtained white matter network.
[0022] When analyzing and screening for outliers in constructed white matter networks, existing techniques often compare the relative spatial location, node clustering coefficient, and node degree of the patient's white matter network with those of a normal white matter network to identify outliers that differ significantly from the normal network. However, this method of screening outliers by comparing node feature parameters with those of a normal white matter network fails to consider the complexity of the human brain structure. Screening solely based on the differences in feature parameters between patient and normal white matter nodes is too one-sided. White matter networks obtained from different individuals inherently differ, and the method does not consider the comparison of nodes within the patient's own white matter network, which can easily lead to missed or false positives, affecting the accuracy of subsequent localization of white matter lesions.
[0023] The following description, in conjunction with the accompanying drawings, details a specific scheme for an intelligent localization method for brain white matter lesions based on MRI images, as provided in this application.
[0024] Please see Figure 1 The diagram illustrates a flowchart of an intelligent localization method for white matter lesions in the brain based on MRI images, provided in one embodiment of this application.
[0025] like Figure 1 As shown, the intelligent localization method for brain white matter lesion regions based on MRI images includes S101-S105.
[0026] S101. Constructing a white matter network based on brain MRI images.
[0027] The brain white matter network includes multiple nodes and edges between nodes. Nodes correspond to white matter aggregation points, and the edges between nodes correspond to white matter regions.
[0028] In one alternative implementation, brain tissue segmentation can be performed based on a deformable surface model segmentation method to extract brain tissue regions, including gray matter and white matter, from brain MRI images to obtain brain tissue images. Then, a brain white matter network can be constructed based on the diffusion tensor model (DTI) technique.
[0029] Optionally, a three-dimensional spatial coordinate system can be deployed to obtain the three-dimensional coordinates of each node in the white matter network.
[0030] Optionally, the brain MRI images are acquired by taking images of the brain while the patient is lying in the correct position on the MRI machine.
[0031] S102. Cluster the nodes in the brain white matter network to obtain multiple clusters.
[0032] Each of these multiple clusters includes at least one edge node.
[0033] It should be understood that a cluster includes at least one node, and an edge node is a node located at the edge of the cluster.
[0034] Optionally, nodes in the brain white matter network can be clustered based on Mean-Shift clustering algorithm to obtain preliminary clustering results.
[0035] Optionally, the brain MRI image may also include the gray value of each pixel. Based on the correspondence between the brain MRI image and the nodes of the brain white matter network, the gray value of each node can be obtained, and then the nodes in the brain white matter network can be clustered based on the gray value of the nodes.
[0036] It should be understood that clustering based on the grayscale value of nodes can group nodes with similar grayscale values into a cluster, that is, filter out nodes that are similar to each other.
[0037] It is understandable that since clustering groups similar nodes together, edge nodes have a lower degree of similarity to other nodes and are therefore located at the edge.
[0038] S103. Based on the differences and affiliation degree between each edge node and its corresponding neighboring clusters, adjust the affiliation of each edge node to a specific cluster.
[0039] Among them, the neighboring clusters of an edge node include the current cluster to which the edge node belongs.
[0040] It should be understood that the adjacent clusters corresponding to an edge node are the clusters of the other nodes connected to that node. An edge node can have multiple adjacent clusters, including other connected clusters in addition to the current cluster to which the edge node belongs.
[0041] It is understandable that white matter lesions can damage and weaken white matter fibers within the lesion area. In the white matter network, the edges connecting each node are white matter regions. Therefore, when a lesion area is mapped onto the white matter network, it will affect the edges and nodes of that region, potentially causing the loss of edges of nodes and making them different from other normal nodes. Thus, the white matter lesion area is very likely to be a marginal node.
[0042] However, clustering based on the similarity between nodes does not consider the comparison between nodes within the white matter network itself, resulting in a more one-sided clustering result with lower accuracy. Therefore, further analysis can be conducted on the differences and affiliations between edge nodes and their corresponding neighboring clusters.
[0043] It is understandable that the difference between an edge node and its corresponding neighboring cluster is used to characterize the degree of difference between the edge node and the neighboring cluster, and the membership degree of an edge node and its corresponding neighboring cluster is used to characterize the degree of connection between the edge node and the neighboring cluster.
[0044] It should be understood that if an edge node has little difference from an adjacent cluster and a high degree of affiliation, then the edge node should be adjusted to the adjacent cluster.
[0045] It is understandable that when clustering based on gray values, the more similar the gray values of two regions are, the higher the similarity of the nodes corresponding to the two regions. Furthermore, the farther a node is from the cluster center, the higher the probability that the node is incorrectly classified into that cluster during the clustering process, and the lower its similarity to that cluster may be. Therefore, the difference between an edge node and its corresponding adjacent cluster can be determined based on the gray value of the edge node and the distance between the edge node and the cluster center of its corresponding adjacent cluster.
[0046] The following uses the first neighboring cluster corresponding to the first edge node as an example to illustrate the method for determining the difference and belonging degree between the first edge node and its corresponding first neighboring cluster.
[0047] In one implementation of this application, the distance between the first edge node and the cluster center of the first adjacent cluster, and the distance between the first edge node and the center node of the first adjacent cluster can be determined. Then, based on the distance between the first edge node and the cluster center of the first adjacent cluster, the distance between the first edge node and the center node of the first adjacent cluster, the gray value of the first edge node, and the gray value of the center node of the first adjacent cluster, the difference between the first edge node and the first adjacent cluster can be determined.
[0048] It should be understood that the first edge node is any one of the at least one edge node, and the first adjacent cluster is any one of the adjacent clusters corresponding to the first edge node.
[0049] Understandably, the greater the distance from a node to the cluster center, the higher the probability that the node is incorrectly classified into that cluster during the clustering process, and the lower its similarity to that cluster.
[0050] Optionally, the average coordinates of all nodes in the first adjacent cluster can be used as the coordinates of the cluster center of the first adjacent cluster. Then, based on the coordinates of the first edge node and the coordinates of the cluster center of the first adjacent cluster, the distance between the first edge node and the cluster center of the first adjacent cluster can be determined.
[0051] Optionally, the node closest to the cluster center of the first adjacent cluster can be determined as the center node of the first adjacent cluster, and then the distance between the first edge node and the center node of the first adjacent cluster can be determined based on the coordinates of the first edge node and the coordinates of the center node of the first adjacent cluster.
[0052] Optionally, the difference between an edge node and its corresponding neighboring cluster satisfies the following formula: in, Represents edge nodes with neighboring clusters Differences, Represents edge nodes with neighboring clusters The distance from the cluster center Represents adjacent clusters central node with neighboring clusters Cluster center distance, Represents edge nodes grayscale value, Represents adjacent clusters central node The grayscale value.
[0053] Based on the above formula, it should be understood that, The larger the value, the more likely it is to be an edge node. Neighbor clusters The greater the distance between its center location coordinates and its neighboring clusters, the closer it is to the center location coordinates. central node The greater the difference; For edge nodes The gray value and its neighboring clusters central node The difference in grayscale values specifically represents the edge nodes. The gray values of the corresponding regions in brain MRI images and the central nodes The difference in grayscale values between corresponding regions in a brain MRI image. The larger the value, the greater the difference between the regions in the corresponding brain MRI images, and the higher its correlation with neighboring clusters. central node The greater the difference.
[0054] Understandably, since grayscale value is the main factor for similarity between nodes, the difference between grayscale values and the ratio of distances can more accurately determine the difference between an edge node and its corresponding neighboring cluster.
[0055] Optionally, the parameters in the above formula can be normalized or the data standardized to eliminate the influence of dimensions.
[0056] It should be understood that an edge node and its corresponding neighboring clusters central node The greater the difference between the edge node and the center node of its cluster, the higher the difference between the edge node and the center node of its cluster, and the more likely it is to need to be adjusted for cluster affiliation.
[0057] In one implementation of this application, the number and length of edges between the first edge node and the nodes in the first adjacent cluster can be determined; based on the number and length of edges between the first edge node and the nodes in the first adjacent cluster, the degree of belonging of the first edge node to the first adjacent cluster can be determined.
[0058] It should be understood that the more edges there are between the first edge node and the nodes in the first adjacent cluster, and the shorter their length, the greater the likelihood that the first edge node belongs to the first adjacent cluster.
[0059] Optionally, the affiliation degree of an edge node with its corresponding neighboring cluster satisfies the following formula: in, Represents edge nodes with neighboring clusters degree of belonging Represents edge nodes with neighboring clusters The number of edges between them, Indicates the first The length of the strip.
[0060] Based on the above formula, it should be understood that, The larger the value, the more likely it is to be an edge node. with neighboring clusters The more edges there are between them, the more edge nodes there are. For adjacent clusters The higher the degree of belonging; The smaller the value, the more likely it is to be an edge node. with neighboring clusters The closer the nodes connected to each other, the better the edge node is. For adjacent clusters The higher the degree of belonging.
[0061] In one alternative implementation, an edge node can be reassigned to a neighboring cluster that is less dissimilar to it and has a higher degree of belonging.
[0062] It should be understood that the purpose of the adjustment is to place the areas of the white matter network affected by white matter lesions as close as possible to the periphery of the clusters, making them more clearly visible and facilitating subsequent diagnosis.
[0063] S104. If the adjustment result meets the preset conditions, determine the final adjustment result.
[0064] The final adjustment result includes the nodes that have been adjusted.
[0065] Understandably, if the adjustment results meet the preset conditions, it means that the edge nodes have been adjusted to the most appropriate cluster. At this point, it can be determined that the adjustment is complete and the final adjustment result is determined.
[0066] In one optional implementation, the adjustment result may include the validity of the current adjustment, and the preset condition may be that the validity of the current adjustment is less than or equal to the adjustment validity threshold.
[0067] For example, the effectiveness threshold for this adjustment can be 0.5.
[0068] It is understandable that the effectiveness of this adjustment is used to characterize the suitability of the edge node's assigned cluster after the adjustment compared to the edge node's assigned cluster before the adjustment. If the effectiveness of this adjustment is less than or equal to the effectiveness threshold, it means that the edge node's assigned cluster after the adjustment is not as suitable as the edge node's assigned cluster before the adjustment, and the edge node's assigned cluster cannot be adjusted to a better one. At this point, it can be determined that the adjustment is complete and the final adjustment result is determined.
[0069] It should be understood that since the clusters to which edge nodes are assigned after this adjustment are not as appropriate as those to which edge nodes were assigned before the adjustment, using the results before this adjustment as the final adjustment result can improve the accuracy of the cluster assignment of edge nodes.
[0070] Understandably, the adjusted nodes are edge nodes whose affiliation has changed compared to their original affiliation cluster.
[0071] Optionally, the adjusted nodes can be determined based on the cluster to which each edge node belonged before this adjustment and the original cluster to which each edge node belonged.
[0072] Specifically, edge nodes whose affiliation to a different cluster are identified as adjusted nodes.
[0073] In one alternative implementation, the maximum adjustment requirement of each edge node that needs adjustment before the current adjustment can be determined, and the maximum adjustment requirement of each edge node that needs to be adjusted after the current adjustment can be determined; based on the maximum adjustment requirement of each edge node that needs adjustment before the current adjustment and the maximum adjustment requirement of each edge node that needs to be adjusted after the current adjustment, the validity of the current adjustment can be determined.
[0074] It is understandable that there may still be edge nodes that need to be adjusted after this adjustment. Therefore, we can identify the edge nodes that need to be adjusted before this adjustment and the edge nodes that need to be adjusted after this adjustment, and compare the maximum adjustment requirements of the two to determine the effectiveness of this adjustment.
[0075] Optionally, the method for determining the maximum adjustment requirement of the edge nodes that need to be adjusted after this adjustment is similar to the method for determining the edge nodes that need to be adjusted before this adjustment.
[0076] Optionally, the number of edge nodes that need to be adjusted before this adjustment and the number of edge nodes that need to be adjusted after this adjustment can also be determined.
[0077] It is understandable that the maximum adjustment requirement of a node represents the maximum matching degree between the node and other adjacent clusters. When the maximum adjustment requirement of a node is higher, it means that the node is more matched with other adjacent clusters, and the cluster to which the node currently belongs is less reasonable.
[0078] Optionally, the validity of the current adjustment can be determined based on the sum of the maximum adjustment demand of each edge node that needs to be adjusted before the current adjustment, and the sum of the maximum adjustment demand of each edge node that needs to be adjusted after the current adjustment.
[0079] Optionally, the validity of this adjustment satisfies the following formula: Here, indicates the validity of this adjustment. This indicates the number of edge nodes that need to be adjusted after this adjustment. This indicates the number after this adjustment. The maximum adjustment requirement for each edge node that needs to be adjusted. This indicates the number of edge nodes that need to be adjusted before this adjustment. This indicates the number before this adjustment. The maximum adjustment requirement for each edge node that needs to be adjusted.
[0080] It should be understood that the above formula is derived from... Simplified This indicates the difference in demand for adjustment before and after the adjustment. The larger the value, the smaller the demand for adjustment after the adjustment compared to before the adjustment, and the higher the effectiveness of the adjustment. This represents the ratio of the reduction in adjustment demand after the adjustment to the total adjustment demand before the adjustment. It also represents the percentage of adjustment demand before the adjustment that is met after the adjustment. The higher this value, the more effective the adjustment is.
[0081] In one alternative implementation, the adjustment result may also include the number of adjustments, with the preset condition being that the number of adjustments is greater than or equal to a threshold number of adjustments.
[0082] For example, the threshold for the number of adjustments can be 5.
[0083] In one alternative implementation, if the adjustment result does not meet the preset conditions, the adjustment can be repeated based on the adjusted clusters and edge nodes until the preset conditions are met.
[0084] S105. The adjusted nodes are identified as the corresponding white matter aggregation points and the white matter lesion areas are determined.
[0085] It should be understood that if the cluster to which a node belongs is adjusted, it means that the clustering of that node is inaccurate. This may be because the edges connected to the node are weakened by the lesions, leading to the node being clustered incorrectly. Therefore, the adjusted node can be identified as an abnormal node, and the white matter aggregation points corresponding to the adjusted node can be identified as white matter lesion areas.
[0086] Optionally, the adjusted node can be mapped onto a brain MRI image and marked to obtain the location of local areas in the MRI image that may be affected by white matter lesions. Then, the marked MRI image can be uploaded to the system for subsequent diagnostic work.
[0087] In this embodiment, a white matter network is constructed based on brain MRI images, and the nodes in the white matter network are clustered to obtain multiple clusters. Then, based on the difference and belonging degree between each edge node and its corresponding neighboring clusters, the belonging cluster of each edge node is adjusted. If the adjustment result meets the preset conditions, it means that the belonging cluster of the edge node has been adjusted to the optimal value. At this time, the adjusted node is identified as an edge node that is significantly different from other surrounding nodes and causes clustering errors. Since white matter lesions weaken white matter fibers in the lesion area and cause the edges between nodes to become blurred, thus blurring the sense of belonging to each brain region, the adjusted node is very likely to be the node that caused the lesion. Identifying the white matter cluster point corresponding to the adjusted node as the white matter lesion area can improve the accuracy of the determination of the white matter lesion area.
[0088] Combination Figure 1 When adjusting the cluster affiliation of each edge node based on its differences and affiliation with its corresponding neighboring clusters, taking the first edge node as an example, as follows... Figure 2 As shown, the affiliation of the first edge node to its corresponding neighboring cluster is adjusted based on the difference and affiliation degree between the first edge node and its neighboring clusters, specifically including S201-S202.
[0089] S201. Based on the differences and affiliations between the first edge node and its corresponding adjacent clusters, determine the adjustment requirement of the first edge node for each adjacent cluster.
[0090] The adjustment requirement of an edge node to a neighboring cluster is used to characterize the degree of matching between the edge node and the neighboring cluster.
[0091] It should be understood that in a healthy human brain, brain regions with similar functions are closely connected, while the connections between different functional systems are relatively sparse. If white matter lesions occur, they will damage and weaken the white matter fibers connecting various brain regions within the lesion area. This is reflected in the white matter network as several edges in the white matter lesion area. The nodes that are the endpoints of these edges are also likely to be affected, making their sense of belonging to each brain region ambiguous, and they may be classified into the wrong cluster by clustering algorithms.
[0092] Understandably, the higher the adjustment requirement of an edge node to a neighboring cluster, the higher the matching degree between the edge node and the neighboring cluster. When the matching degree between the edge node and the neighboring cluster is greater than the matching degree between the edge node and its current cluster, it indicates that the edge node may be clustered incorrectly, and the edge node should be adjusted to the neighboring cluster.
[0093] It should be understood that the greater the difference, the smaller the need for adjustment; conversely, the greater the degree of belonging, the greater the need for adjustment.
[0094] In one alternative implementation, the ratio of the degree of belonging to a first edge node to the degree of dissimilarity between it and a certain neighboring cluster can be determined as the degree of adjustment requirement between the first edge node and a certain neighboring cluster.
[0095] In another alternative implementation, the difference and affiliation degree between the first edge node and a certain neighboring cluster can be compared with the difference and affiliation degree of the current affiliation cluster of the first edge node to obtain the adjustment requirement degree.
[0096] Specifically, taking the adjustment requirement of the first edge node to the second adjacent cluster as an example, the adjustment requirement of the first edge node to the second adjacent cluster can be determined based on the difference between the first edge node and its own cluster, the degree of belonging between the first edge node and its own cluster, the difference between the first edge node and the second adjacent cluster, and the degree of belonging between the first edge node and the second adjacent cluster.
[0097] The second adjacent cluster is the adjacent cluster of the first edge node, excluding the cluster to which the first edge node belongs.
[0098] Optionally, the adjustment requirement of an edge node to a neighboring cluster satisfies the following formula: in, Represents edge nodes For adjacent clusters Adjustment of demand Represents edge nodes With edge nodes The difference in the cluster to which they belong, Represents edge nodes with neighboring clusters Differences, Represents edge nodes with neighboring clusters degree of belonging Represents edge nodes With edge nodes The degree of belonging to the cluster to which it belongs.
[0099] Based on the above formula, it should be understood that, The larger the value, the more important the edge node is compared to its cluster. Its neighboring clusters The smaller the difference, the better the edge node. The more it needs to be adjusted to that adjacent cluster, the more likely it is to be affected. Inside; When the value is positive and larger, it indicates that the edge node... For adjacent clusters The higher the degree of belonging, the better the edge node. The more it needs to be adjusted to that adjacent cluster, the more likely it is to be affected. Inside.
[0100] Optionally, since the adjustment demand degree obtained from the above formula may be negative, the normalization function sigmoid() can be used to map the negative adjustment demand degree to [0,0.5] and the positive adjustment demand degree to (0.5,1).
[0101] S202. If the adjustment demand degree of the target adjacent cluster is greater than or equal to the adjustment demand degree threshold, adjust the belonging cluster of the first edge node to the target adjacent cluster.
[0102] Among them, the target adjacent cluster is the adjacent cluster corresponding to the maximum adjustment demand degree.
[0103] It is understandable that, since the target neighboring cluster is the neighboring cluster corresponding to the maximum adjustment demand degree, if the adjustment demand degree of the target neighboring cluster is greater than or equal to the adjustment demand degree threshold, it means that the matching degree of the target neighboring cluster is better than the current belonging cluster of the first edge node. In this case, the belonging cluster of the first edge node can be adjusted to the target neighboring cluster.
[0104] Optionally, if the adjustment demand degree of the target adjacent cluster is less than the adjustment demand degree threshold, it indicates that the matching degree between the first edge node and the target adjacent cluster is similar to or lower than the matching degree between the first edge node and the current belonging cluster. In this case, there is no need to adjust the belonging cluster of the first edge node, and the belonging cluster of the first edge node is determined to be the current belonging cluster of the first edge node.
[0105] Optionally, after the above-mentioned adjustment demand normalization process, the adjustment demand threshold can be 0.8, that is, when the matching degree of an edge node with its neighboring clusters is much better than the matching degree of the edge node with its own cluster, the belonging cluster of the edge node is then adjusted.
[0106] In this embodiment, the difference and affiliation degree between the first edge node and a certain neighboring cluster are compared with the difference and affiliation degree between the first edge node and the current affiliation cluster. The resulting adjustment demand degree can accurately reflect whether the first edge node needs to be adjusted to other neighboring clusters. At this time, adjusting the affiliation cluster of the first edge node to the neighboring cluster with the largest adjustment demand degree and greater than the adjustment demand degree threshold can improve the effectiveness of clustering result adjustment.
[0107] In summary, existing technologies that use comparisons with node feature parameters in normal white matter networks to screen for abnormal nodes fail to consider the complexity of the human brain structure. White matter networks obtained from different individuals inherently differ, and the methods do not consider the internal node comparisons within the patient's own white matter network, easily leading to missed or false positives, thus affecting the accuracy of subsequent localization of white matter lesions. In contrast, this application further combines the comparison of nodes within the obtained white matter network with their surrounding nodes to identify the locations of nodes with significant differences from their surroundings, accurately locating abnormal nodes in the white matter network and improving the accuracy and efficiency of identifying and locating white matter lesions.
[0108] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0109] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. A method for intelligent localization of white matter lesions in the brain based on MRI images, characterized in that, The method includes: A brain white matter network is constructed based on brain MRI images. The brain white matter network includes multiple nodes and edges between nodes. Nodes correspond to white matter aggregation points, and the edges between nodes correspond to white matter regions. The nodes in the white matter network are clustered to obtain multiple clusters, and each cluster includes at least one edge node. Based on the differences and affiliation degree between each edge node and its corresponding neighboring clusters, the affiliation cluster of each edge node is adjusted, and the neighboring clusters of an edge node include the current affiliation cluster of the edge node; If the adjustment results meet the preset conditions, the final adjustment result is determined, and the final adjustment result includes the adjusted nodes; The white matter aggregation points corresponding to the adjusted nodes were identified as areas of white matter lesions.
2. The intelligent localization method for brain white matter lesions based on MRI images according to claim 1, characterized in that, The white matter network includes grayscale values for each node, and the method further includes: Determine the distance between the first edge node and the cluster center of the first adjacent cluster, and the distance between the first edge node and the center node of the first adjacent cluster, wherein the first edge node is any one of the at least one edge node, and the first adjacent cluster is any one of the adjacent clusters corresponding to the first edge node; The degree of difference between the first edge node and the first adjacent cluster is determined based on the distance between the first edge node and the cluster center of the first adjacent cluster, the distance between the first edge node and the center node of the first adjacent cluster, the gray value of the first edge node, and the gray value of the center node of the first adjacent cluster.
3. The intelligent localization method for brain white matter lesion regions based on MRI images according to claim 1, characterized in that, The method further includes: Determine the number and length of the edges between the first edge node and the nodes in the first adjacent cluster; The degree of belonging of the first edge node to the first adjacent cluster is determined based on the number and length of the edges between the first edge node and the nodes in the first adjacent cluster.
4. The intelligent localization method for brain white matter lesion regions based on MRI images according to claim 1, characterized in that, Based on the differences and affiliations between the first edge node and its corresponding neighboring clusters, the affiliation of the first edge node is adjusted, including: Based on the differences and affiliations between the first edge node and its corresponding neighboring clusters, the adjustment requirement of the first edge node for each neighboring cluster is determined. The adjustment requirement of an edge node for a neighboring cluster is used to characterize the degree of matching between the edge node and the neighboring cluster. If the adjustment demand degree of the target neighboring cluster is greater than or equal to the adjustment demand degree threshold, the belonging cluster of the first edge node is adjusted to the target neighboring cluster, which is the neighboring cluster corresponding to the maximum adjustment demand degree.
5. The intelligent localization method for brain white matter lesion regions based on MRI images according to claim 4, characterized in that, Based on the differences and affiliations between the first edge node and its corresponding neighboring clusters, the adjustment requirement of the first edge node for the second neighboring cluster is determined, including: Based on the differences between the first edge node and its affiliated cluster, the degree of affiliation between the first edge node and its affiliated cluster, the differences between the first edge node and the second adjacent cluster, and the degree of affiliation between the first edge node and the second adjacent cluster, the adjustment requirement of the first edge node to the second adjacent cluster is determined. The second adjacent cluster is the adjacent cluster of the first edge node, excluding the first edge node's affiliated cluster.
6. The intelligent localization method for brain white matter lesion regions based on MRI images according to claim 1, characterized in that, The adjustment result includes the validity of this adjustment, and the preset condition includes that the validity of this adjustment is less than or equal to the validity threshold of the adjustment. The validity of this adjustment is used to characterize the appropriateness of the cluster affiliation of the edge node after this adjustment compared with the affiliation of the edge node before the adjustment.
7. The intelligent localization method for brain white matter lesion regions based on MRI images according to claim 6, characterized in that, The method further includes: Determine the maximum adjustment requirement for each edge node that needs adjustment before this adjustment, and determine the maximum adjustment requirement for each edge node that needs adjustment after this adjustment; The effectiveness of this adjustment is determined based on the maximum adjustment requirement of each edge node that needs adjustment before this adjustment, and the maximum adjustment requirement of each edge node that needs adjustment after this adjustment.
8. The intelligent localization method for brain white matter lesion regions based on MRI images according to claim 7, characterized in that, The determination of the effectiveness of the current adjustment based on the maximum adjustment requirement of each edge node that needs adjustment before the current adjustment and the maximum adjustment requirement of each edge node that needs adjustment after the current adjustment includes: The validity of this adjustment is determined based on the sum of the maximum adjustment requirements of each edge node that needs adjustment before this adjustment, and the sum of the maximum adjustment requirements of each edge node that needs adjustment after this adjustment.
9. The intelligent localization method for brain white matter lesion regions based on MRI images according to claim 1, characterized in that, Determining the final adjustment result includes: Based on the cluster to which each edge node belonged before this adjustment and the original cluster to which each edge node belonged, the adjusted nodes are determined.
10. The intelligent localization method for brain white matter lesion regions based on MRI images according to claim 2, characterized in that, The method further includes: If the adjustment demand of the target adjacent clusters is less than the adjustment demand threshold, the cluster to which the first edge node belongs is determined as the current cluster to which the first edge node belongs.