Fiber bundle division method and system based on fiber similarity and storage medium

Abstract

The invention provides a fiber bundle division method and system based on fiber similarity and a storage medium, and relates to the technical field of image processing. The method comprises the following steps: performing interval division on each fiber of a target fiber bundle, performing clustering point sampling on the fibers in each interval, calculating the similarity between the fibers based on the clustering points, generating an initial similarity matrix, performing hierarchical clustering on the fiber bundle by using a hierarchical clustering algorithm based on the similarity matrix, generating a preliminary clustering result, and performing clustering on the fiber bundle according to the preliminary clustering result. And calculating the similarity of the fiber clusters between the intervals based on the clustering points, generating a clustering similarity matrix, carrying out hierarchical clustering on the fiber clusters based on the clustering similarity matrix, and carrying out hierarchical clustering on a clustered result to generate a fiber clustering result of the target fiber bundle. The fiber bundles are clustered for multiple times, so that fibers in the clusters have tightness, the separation degree exists between the clusters, and the fiber bundle division precision and efficiency are improved.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of image processing, and in particular to a fiber bundle segmentation method, system and storage medium based on fiber similarity. Background Art

[0002] White matter fiber bundles are large fiber clusters composed of a series of adjacent nerve fibers with similar structures. They are key neural structures that play a connecting role in the brain, responsible for connecting different brain areas and promoting information transmission. Diffusion Tensor Imaging (DTI) technology is a non-invasive in vivo detection technology that estimates the structure of white matter (WM) by measuring the diffusion of water molecules. Based on diffusion tensor imaging (DTI), the local directional field of white matter is estimated to generate a three-dimensional representation of white matter fiber bundles, which can non-invasively study the organizational structure and morphological characteristics of white matter fiber bundles, so as to accurately characterize the fiber bundles, thereby better understanding brain structural connections, brain functional networks, pathological changes of neurological diseases, and formulating personalized treatment plans.

[0003] Although diffusion tensor imaging (DTI) technology provides rich fiber tracking information, the precise division of fiber bundles is still challenging due to the complex structure of fiber bundles. Due to the complex fiber crossing and overlapping in the white matter region of the brain, the traditional division method based on spatial position or simple geometric features cannot accurately divide the morphology of fiber bundles. The length, shape and spatial distribution of fiber bundles vary, which requires adaptive division of fibers of different scales. In addition, the noise in diffusion tensor imaging (DTI) data will affect the accuracy of fiber tracking, making subsequent division more difficult. Therefore, there is an urgent need for a fine division method for fiber bundles suitable for studying the structure and function of white matter fiber bundles in the brain. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a fiber bundle division method, system and storage medium based on fiber similarity in view of the deficiencies in the prior art.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] A fiber bundle division method based on fiber similarity comprises the following steps:

[0007] Importing a diffusion tensor image of the brain, performing probabilistic fiber tracking on the diffusion tensor image to obtain a fiber data set, and selecting a target fiber bundle from the fiber data set, wherein the target fiber bundle includes a plurality of fibers;

[0008] Determine the longest fiber length among multiple fibers, divide the longest fiber length into multiple length intervals in an increasing manner, match multiple fibers with multiple length intervals, divide them into multiple initial fiber groups corresponding to the length intervals, calculate the similarity between each fiber in the multiple initial fiber groups by a fiber similarity algorithm, and obtain an initial similarity matrix of the multiple initial fiber groups;

[0009] Merging the fibers in the corresponding initial fiber groups according to the first clustering threshold and the multiple initial similarity matrices by a hierarchical clustering algorithm to obtain multiple fiber groups, each fiber group including multiple different fiber clusters and multiple unmerged remaining fibers;

[0010] The similarities between the plurality of fiber clusters and the plurality of remaining fibers among the plurality of fiber groups are calculated respectively by a fiber similarity algorithm to obtain an initial clustering similarity matrix, and the plurality of fiber clusters of all fiber groups are merged with the plurality of remaining fibers by a hierarchical clustering algorithm according to a second clustering threshold and the initial clustering similarity matrix to obtain a plurality of large fiber clusters and a plurality of small fiber clusters;

[0011] The similarities between the multiple large fiber clusters and the multiple small fiber clusters are calculated respectively by a fiber similarity algorithm to obtain a cluster similarity matrix, and the multiple large fiber clusters and the multiple small fiber clusters are merged according to a third clustering threshold and the cluster similarity matrix by a hierarchical clustering algorithm to obtain a fiber clustering result consisting of multiple clusters.

[0012] Another technical solution of the present invention to solve the above technical problems is as follows:

[0013] A fiber bundle partitioning system based on fiber similarity, comprising:

[0014] A preprocessing module, used for importing a diffusion tensor image of a brain, performing probabilistic fiber tracking on the diffusion tensor image to obtain a fiber data set, and selecting a target fiber bundle from the fiber data set, wherein the target fiber bundle includes a plurality of fibers;

[0015] A similarity calculation module is used to determine the longest fiber length among multiple fibers, divide the longest fiber length into multiple length intervals in an increasing manner, match multiple fibers with multiple length intervals, divide them into multiple initial fiber groups corresponding to the length intervals, and calculate the similarity between each fiber in the multiple initial fiber groups by a fiber similarity algorithm to obtain an initial similarity matrix of the multiple initial fiber groups;

[0016] An intra-group clustering module, used for merging the fibers in the corresponding initial fiber groups according to the first clustering threshold and the multiple initial similarity matrices by a hierarchical clustering algorithm to obtain multiple fiber groups, each fiber group including multiple different fiber clusters and multiple unmerged remaining fibers;

[0017] An inter-group clustering module is used to calculate the similarity between the plurality of fiber clusters and the plurality of remaining fibers between the plurality of fiber groups respectively by using a fiber similarity algorithm to obtain an initial clustering similarity matrix, and to merge the plurality of fiber clusters of all fiber groups with the plurality of remaining fibers by using a hierarchical clustering algorithm according to a second clustering threshold and the initial clustering similarity matrix to obtain a plurality of large fiber clusters and a plurality of small fiber clusters;

[0018] The inter-cluster clustering module is used to calculate the similarity between multiple large fiber clusters and multiple small fiber clusters respectively through a fiber similarity algorithm to obtain a cluster similarity matrix, and merge the multiple large fiber clusters and the multiple small fiber clusters according to a third clustering threshold and the cluster similarity matrix through a hierarchical clustering algorithm to obtain a fiber clustering result composed of multiple clusters.

[0019] Another technical solution of the present invention to solve the above technical problems is as follows:

[0020] A fiber bundle division system based on fiber similarity comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the fiber bundle division method based on fiber similarity as described above is implemented.

[0021] Another technical solution of the present invention to solve the above technical problems is as follows:

[0022] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the fiber bundle division method based on fiber similarity is implemented.

[0023] The beneficial effects of the present invention are as follows: the present invention takes the distance similarity between fiber bundles as a feature, divides the fiber bundles into several small fiber clusters through a hierarchical clustering algorithm, and then merges the small clusters into a large cluster to form the final fiber bundle division result, that is, similar fibers of the fiber bundle are merged into one cluster, so that the fiber morphology and direction within the cluster of the division result are similar, ensuring the compactness of the fibers in the cluster, and improving the separation between clusters, thereby achieving high-precision and high-efficiency division of the fiber bundles. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A flowchart of a fiber bundle division method based on fiber similarity provided by an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of a target fiber bundle provided by an embodiment of the present invention;

[0026] Figure 3 A schematic diagram of fiber clustering results of a target fiber bundle provided by an embodiment of the present invention;

[0027] Figure 4 A schematic diagram of a fiber bundle division process provided by an embodiment of the present invention;

[0028] Figure 5 A module block diagram of a fiber bundle partitioning system based on fiber similarity provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0030] The fiber tracking algorithm is used to estimate the local direction field of the white matter (WM) from the diffusion tensor imaging (DTI), reconstruct the main trajectory of the white matter, and generate a data set containing a three-dimensional representation of the white matter fibers (i.e., the diffusion tensor image); the fiber tracking algorithm implements probabilistic streamline tracking, following the preferred direction of water molecule diffusion in each voxel to reconstruct the trajectory represented by a series of three-dimensional space midpoint coordinates. These trajectories are called "streamlines" or simply "fibers", although they do not represent real nerve fibers, but are estimates of the main trajectory of the white matter tracts in the brain. The white matter fiber tract is a large fiber cluster composed of a series of adjacent nerve fibers with similar structures. It is a key neural structure that plays a connecting role in the brain, responsible for connecting different brain areas and promoting information transmission. Through technologies such as DTI, the organizational structure and morphological characteristics of these fiber tracts can be studied non-invasively, and then the fiber tracts can be classified and divided by characteristics. The accurate division of fiber tracts helps to more accurately understand the brain structure connection, brain functional network, pathological changes of neurological diseases, and formulate personalized treatment plans.

[0031] like Figure 1-Figure 4 As shown, a fiber bundle division method based on fiber similarity provided by an embodiment of the present invention includes the following steps:

[0032] Importing a diffusion tensor image of the brain, performing probabilistic fiber tracking on the diffusion tensor image to obtain a fiber data set, and selecting a target fiber bundle from the fiber data set, wherein the target fiber bundle includes a plurality of fibers;

[0033] Determine the longest fiber length among multiple fibers, divide the longest fiber length into multiple length intervals in an increasing manner, match multiple fibers with multiple length intervals, divide them into multiple initial fiber groups corresponding to the length intervals, calculate the similarity between each fiber in the multiple initial fiber groups by a fiber similarity algorithm, and obtain an initial similarity matrix of the multiple initial fiber groups;

[0034] Merging the fibers in the corresponding initial fiber groups according to the first clustering threshold and the multiple initial similarity matrices by a hierarchical clustering algorithm to obtain multiple fiber groups, each fiber group including multiple different fiber clusters and multiple unmerged remaining fibers;

[0035] The similarities between the plurality of fiber clusters and the plurality of remaining fibers among the plurality of fiber groups are calculated respectively by a fiber similarity algorithm to obtain an initial clustering similarity matrix, and the plurality of fiber clusters of all fiber groups are merged with the plurality of remaining fibers by a hierarchical clustering algorithm according to a second clustering threshold and the initial clustering similarity matrix to obtain a plurality of large fiber clusters and a plurality of small fiber clusters;

[0036] The similarities between the multiple large fiber clusters and the multiple small fiber clusters are calculated respectively by a fiber similarity algorithm to obtain a cluster similarity matrix, and the multiple large fiber clusters and the multiple small fiber clusters are merged according to a third clustering threshold and the cluster similarity matrix by a hierarchical clustering algorithm to obtain a fiber clustering result consisting of multiple clusters.

[0037] In the embodiment of the present invention, the sampling hierarchical clustering algorithm has important advantages in processing large-scale fiber tracking data. It can classify similar fibers into the same category according to the geometric or topological characteristics of the fiber bundles, thereby reducing the complexity of the data. After the fiber bundles are divided and clustered according to similarity, it can help accurately identify and segment the fiber bundles related to specific anatomical functions, while reducing the complexity of subsequent brain structure analysis, and thus better understand the structural and functional characteristics of the brain; through the differences in fiber bundle morphology or connectivity, it provides a basis for disease diagnosis and staging, and helps neurosurgery planning to avoid damage to important fibers, thereby specifying personalized surgical treatment plans.

[0038] Preferably, relevant image processing tools are used to pre-process the diffusion tensor image DTI of the brain in NIFTI format, the gradient encoding file (.bval file) of the DTI in the magnetic field gradient application intensity, and the gradient encoding file (.bvec file) of the DTI in the magnetic field gradient application direction, to generate a mask file (nodif_brain_mask.nii.gz), tensor model fractional anisotropy (FA), mean diffusivity (MD) and other parameters for whole-brain probabilistic fiber tracking, thereby obtaining a fiber data set of the whole brain, and generating the desired fiber bundle streamlines (i.e., target fiber bundles) according to the custom region of interest ROI, streamlines = {s1, s2, ..., s i ,…,s k}, k ≥ i, i ≥ 1, where s i Represents the i-th fiber in the fiber bundle to be divided.

[0039] It should be understood that the relevant image processing tools are FMRIB Software Library (FSL) software or mrtrix3 software. The mask file is used to define the area of ​​fiber tracking and limit the tracking range. FA is fractional anisotropy, and the FA value is used to guide the fiber tracking algorithm to track the fiber bundle along the main direction of water molecule diffusion; MD is mean diffusivity, which mainly reflects the overall characteristics of water molecule diffusion, assists in judging the tissue state and optimizes the tracking results. These files serve as input files for FSL software to provide a basis for fiber bundle tracking. The result of whole-brain probabilistic fiber tracking is a fiber data set containing multiple fibers generated based on a diffusion tensor imaging (DTI) image. The fiber bundles (i.e., target fiber bundles) that you want to divide are generated from the fiber data set, and the region of interest ROI of the individual target can be generated by registration through the region of interest ROI of the standard template.

[0040] Preferably, the longest fiber length is determined among the multiple fibers, and the longest fiber length is divided into multiple length intervals in an increasing manner, including: counting the length of each fiber, taking the length L of the longest fiber max , and the fiber length interval is [0,L max ], the fiber length interval is divided into c equal parts, and c initial fiber intervals (i.e., initial fiber group SI) are obtained. i )streamlineinterval={SI1,SI2,…,SI i ,…,SI c}, c≥1.

[0041] When the length of the longest fiber is L max=87.58mm, and the fiber length interval is [0,87.58]. To facilitate the interval division, the interval is set to [0,90], and the fiber length interval is divided into 6 equal parts. The length intervals of the preliminary fiber intervals are (0,15], (15,30], (30,45], (45,60], (60,75], (75,90], that is, divided into 6 preliminary fiber intervals streamlineinterval = {SI1, SI2, …, SI6}.

[0042] It should be understood that the fiber bundle is a spatial structure, and the data of each fiber is generally stored in the form of three-dimensional coordinates. The three-dimensional coordinates are calculated using matlab tools to obtain the length of the fiber.

[0043] In the embodiment of the present invention, a plurality of preliminary fiber intervals are first divided, and the range of each interval is set, and fibers with a fiber length within the range of the preliminary fiber interval are divided into the interval, so as to partition the target fiber bundle, which can reduce the amount of data for subsequent fiber clustering.

[0044] Preferably, the similarity between each fiber in the plurality of initial fiber groups is calculated by using a fiber similarity algorithm to obtain an initial similarity matrix of the plurality of initial fiber groups, including:

[0045] S11, acquiring the same number of clustering points on any two fibers respectively according to the set sampling order, and obtaining a plurality of positive-sequence clustering points corresponding to any two fibers;

[0046] S12, calculating the distances between the positive order clustering points of the same sampling order in any two fibers respectively by using the Manhattan distance expression, and obtaining multiple positive Manhattan distances of any two fibers;

[0047] S13, calculating the plurality of forward Manhattan distances using a similarity expression to obtain a forward similarity;

[0048] S14, performing reverse order processing on the plurality of positive order clustering points of one fiber among any two fibers to obtain a plurality of reverse order clustering points;

[0049] S15, calculating the distances between the positive order clustering points and the corresponding reverse order clustering points of the same sampling order in any two fibers respectively by using the Manhattan distance expression, and obtaining multiple reverse Manhattan distances of any two fibers;

[0050] S16, calculating the multiple reverse Manhattan distances by a similarity expression to obtain reverse similarity;

[0051] S17, selecting a maximum value from the forward similarity and the reverse similarity to obtain an initial similarity between any two fibers;

[0052] S18, repeat S11-S17 until the similarities between the fibers in all the initial fiber groups are calculated, and an initial similarity matrix corresponding to the multiple initial fiber groups is obtained.

[0053] Specifically, according to the set sampling parameters, any fiber is sampled at equal distances to obtain multiple sampling points (such as collecting 13 sampling points), and multiple clustering points are selected from the multiple sampling points in the sampling order (such as selecting the first, fourth, seventh, tenth and thirteenth sampling points as clustering points). Each preliminary fiber interval selects the clustering points of the fiber (that is, multiple clustering points in positive order of the initial fiber group) to calculate the fiber similarity, and calculates the Manhattan distance of the corresponding clustering points between each fiber (such as the first point of two fibers corresponds to the first point, the seventh point corresponds to the seventh point, etc.), and obtains the distance set D of each fiber, D = {D1, D4, D7, D 10 ,D 13}, the distance calculation formula (that is, the Manhattan distance expression) is as follows:

[0054] D bj-aj =|x bj -x aj |+|y bj -y aj |+|z bj -z aj |,

[0055] Where a, b∈(1,k), k represents the number of fibers, j∈{1,4,7,10,13}, D bj-aj Indicates fiber b The jth sampling point and fiber s a The Manhattan distance between the j-th sampling points;

[0056] The Manhattan distances of the five sampling points (i.e., multiple positive order clustering points) corresponding to the two fibers are summed up and averaged, and then the reciprocal is taken to obtain the similarity. The calculation formula (i.e., similarity expression) is as follows:

[0057]

[0058] Among them, D E(b-a) As the similarity between the fibers in the positive direction (i.e., positive similarity), 0 <D E(b-a) <1;

[0059] Flip the fibers and calculate similarity, including:

[0060] For the two fibers after calculating the forward similarity, select one of them, reverse the order of its 5 sampling points, calculate the Manhattan distance between the 5 sampling points of one fiber and the 5 sampling points of the reversed fiber, and calculate the fiber flip similarity (i.e., reverse similarity) based on the obtained 5 Manhattan distances. The calculation formula is as follows:

[0061]

[0062] D F(b-a) =D Fb-a =D b-Fa ,

[0063] Among them, D′ bj-aj represents the Manhattan distance of the jth sampling point after reverse order, D F(b-a) Indicates the similarity of flipped fibers, D Fb-a To flip the fibers Fb With fiber a The similarity between b-Fa For fiber b With flipped fibers Fa The similarity between <D F(b-a) =D Fb-a =D b-Fa <1;

[0064] Take the positive fiber similarity D E(b-a) Similarity with flipped fiber D F(b-a) The maximum value of is taken as the similarity D of the two fibers. b-a , as the data of the similarity matrix, the calculation formula is as follows:

[0065] D b-a =max(D E(b-a) ,D F(b-a) ),

[0066] Among them, D b-a As the similarity between two fibers, 0 <D b-a <1;

[0067] Multiple similarities are obtained by calculating each fiber in the preliminary fiber interval two by two to construct an initial similarity matrix.

[0068] It should be understood that if the three-dimensional coordinates of the five positive-order clustering points of a fiber are P1=(10,15,20), P2=(12,18,25), P3=(14,20,30), P4=(16,23,35) and P5=(18,25,40), flipping the fiber is a simulated flipping, and the five positive-order clustering points are arranged in reverse order. The three-dimensional coordinates of the five reverse-order clustering points are P1′=(18,25,40), P2′=(16,23,35), P3′=(14,20,30), P4′=(12,18,25) and P5′=(10,15,20).

[0069] Since the spatial layout of the fibers themselves remains unchanged, flipping the fibers changes the direction of the fibers but does not change their geometry or length. Fiber bundles have directionality in three-dimensional space, and the direction of a fiber bundle may have the same biological significance as its reverse direction in some analyses, so flipping the fibers avoids the same fiber being mistakenly considered as a different cluster (i.e., avoiding low similarity).

[0070] In an embodiment of the present invention, the Manhattan distance between corresponding sampling points on each pair of fibers is calculated to further calculate the similarity of the two fibers. The fibers are flipped to avoid the same fiber being mistakenly regarded as a different cluster, and the initial similarity matrix of each initial fiber group is obtained to perform clustering based on similarity.

[0071] Preferably, the S11 includes:

[0072] The fiber length of any fiber is calculated according to the set sampling parameters to obtain the sampling distance, multiple sampling points are obtained on any fiber according to the sampling distance, cluster points are selected from the multiple sampling points according to the set sampling order to obtain multiple positive order cluster points, and the remaining fibers are processed in this process to obtain multiple positive order cluster points corresponding to any two fibers.

[0073] Specifically, the sampling parameter is set to 13, one end of the fiber is set as the starting point, and the other end is set as the end point. Each fiber of the target fiber bundle is sampled at equal distances in the order from the starting point to the end point, and 13 sampling points are obtained. i ={(x1,y1,z1), (x2,y2,z2),…,(x j ,y j ,z j ),…,(x 13 ,y 13 ,z 13 )}, 1≤j≤13, (x j ,y j ,z j ) represents the jth sampling point, x j ,y j and zb The voxel coordinates on the x, y and z axes corresponding to the fiber sampling point are respectively represented. Five of them are selected as clustering points in the order from the starting point to the end point of the fiber. These five points are the center point of the fiber (x7, y7, z7), the starting point of the fiber (x1, y1, z1) and the end point (x 13 ,y 13 ,z 13 ) two endpoints, the midpoint between the starting point and the center point (x4, y4, z4), and the midpoint between the end point and the center point (x 10 ,y 10 ,z 10 ) as multiple positive order clustering points.

[0074] In the embodiment of the present invention, a fixed number of cluster points are sampled for the fibers to reduce the data size, thereby significantly reducing the time complexity of calculating the distance or similarity matrix, and the equidistant sampling extracts the global shape of the fiber by evenly distributing points, reducing the impact of noise while retaining the key morphological features of the fiber. The same number of cluster points are collected for each fiber, so that the point coordinates of the two fibers are calculated point by point in the subsequent step, so as to measure the similarity between the fibers by the distance between the cluster points.

[0075] Preferably, the fibers in the corresponding initial fiber groups are merged according to the first clustering threshold and the multiple initial similarity matrices by using a hierarchical clustering algorithm to obtain multiple fiber groups, including:

[0076] S21, selecting any initial similarity reaching the first clustering threshold from any of the initial similarity matrices as the clustering similarity, merging two fibers corresponding to the clustering similarity to obtain a fiber cluster, and in this way, processing the fibers corresponding to all the initial similarities to obtain a plurality of different fiber clusters of any of the initial fiber groups and a plurality of unmerged remaining fiber clusters;

[0077] S22, calculating the similarities between the plurality of fiber clusters and the plurality of remaining fiber clusters respectively by using a fiber similarity algorithm to obtain a clustering similarity matrix;

[0078] S23, looping S21-S22, processing the clustering similarity matrix, the plurality of fiber clusters and the plurality of remaining fiber clusters generated in each loop, until no similarity reaching the first clustering threshold can be screened out, and obtaining a fiber group consisting of a plurality of different fiber clusters and a plurality of unmerged remaining fibers;

[0079] In this way, S21-S23 are performed to process the fibers in all the initial fiber groups to obtain multiple fiber groups.

[0080] Specifically, the clustering algorithm used is agglomerative hierarchical clustering, which is a bottom-up clustering algorithm, and the linkage criterion selected is average linkage. When starting the clustering, each fiber is initially regarded as an independent cluster, that is, there are initially k clusters, and each cluster contains only one fiber. At each step in the clustering process, the two clusters with the highest similarity in the similarity matrix (since similarity is the reciprocal of the Manhattan distance between two fibers, that is, the minimum Manhattan distance) are found and merged to form a new cluster. Then, the fiber similarity algorithm is used to calculate the similarity between the new cluster and other clusters (including other new clusters and unmerged clusters), and the similarity matrix is updated synchronously to reflect the Manhattan distance between the new cluster and other clusters.

[0081] Among them, when two clusters are merged into a new cluster, 5 clustering points on the centroid line of the new cluster are selected as the basis for calculating the Manhattan distance. The process of calculating the similarity between two clusters (i.e., the new cluster and other clusters) by the fiber similarity algorithm is the same as the process of calculating the similarity between each fiber in multiple said initial fiber groups by the fiber similarity algorithm.

[0082] In hierarchical clustering, the clustering threshold is used to control when to stop the clustering process. If no threshold is set, it will cause the clustering process to continuously merge similar clusters until all fibers are merged into one cluster, which is contrary to the clustering algorithm. Therefore, the first clustering threshold D1 (determined according to the actual situation, set to 0.15), 0 < D1 < 1, and the similarity between two clusters is where c m and c n are clusters, 1 ≤ m ≤ k, 1 ≤ n ≤ k, and it is judged whether holds. If so, then c m and c n are merged into a new cluster. Otherwise, the two are not merged. Then, the similarity between the new cluster and other clusters is calculated, and the similarity between the two clusters is compared with the first clustering threshold D1, and the merging process is repeated until all similarities are less than the first clustering threshold D1 and no clusters can be merged into new clusters of fibers, that is, the fiber clustering is completed.

[0083] Hierarchical clustering is performed on each preliminary fiber interval (i.e., the initial fiber group) respectively. Each preliminary fiber interval SI p respectively obtains clusters C p ={C p1 ,C p2 ,C p3 ,…,C pi}, where i ≥ 1, p ∈ {1, 6}, and i is the number of clusters in each preliminary fiber interval.

[0084] In the embodiment of the present invention, fibers that can be merged are screened out from the similarity matrix according to the first clustering threshold and merged, and the similarity between the merged fiber cluster and the remaining fibers is calculated, so that the fibers are continuously merged layer by layer according to the first clustering threshold until similar fibers are merged into one cluster, thereby completing the initial category division within the interval of the target fiber bundle.

[0085] Preferably, the similarities between the plurality of fiber clusters and the plurality of remaining fibers between the plurality of fiber groups are calculated respectively by a fiber similarity algorithm to obtain an initial clustering similarity matrix, and the plurality of fiber clusters of all fiber groups and the plurality of remaining fibers are merged by a hierarchical clustering algorithm according to a second clustering threshold and the initial clustering similarity matrix to obtain a plurality of large fiber clusters and a plurality of small fiber clusters, including:

[0086] Hierarchical clustering is performed on the merged and unmerged fibers in all fiber groups, that is, the similarities between multiple fiber clusters and multiple remaining fibers in one fiber group and multiple fiber clusters and multiple remaining fibers in another fiber group are calculated pairwise, and two clusters (i.e., fiber clusters and fiber clusters, fiber clusters and remaining fibers, or remaining fibers and remaining fibers) with similarities greater than or equal to the second clustering threshold are merged. The merged target fiber bundle includes multiple clusters (i.e., clusters containing multiple fibers after merging and clusters containing only one fiber before merging), and clusters with less than or equal to 9 fibers are regarded as small fiber clusters, while clusters with more than or equal to 10 fibers are regarded as large fiber clusters.

[0087] It should be understood that the process of calculating the similarity of two clusters by the fiber similarity algorithm is the same as the process of calculating the similarity between each fiber in the multiple initial fiber groups by the fiber similarity algorithm. The process of merging the multiple fiber clusters of all fiber groups with the multiple remaining fibers according to the second clustering threshold and the initial clustering similarity matrix by the hierarchical clustering algorithm is the same as the process of merging the fibers in the corresponding initial fiber groups according to the first clustering threshold and the multiple initial similarity matrices by the hierarchical clustering algorithm. The second clustering threshold D2 needs to be less than the first clustering threshold D1, and is set to 0.09, so that the fiber bundles are more compact and have a higher degree of separation.

[0088] In the embodiment of the present invention, all fiber groups, that is, all fibers, are clustered, and the target fiber bundles are classified into categories in a full range.

[0089] Preferably, the similarities between the multiple large fiber clusters and the multiple small fiber clusters are calculated respectively by a fiber similarity algorithm to obtain a cluster similarity matrix, and the multiple large fiber clusters and the multiple small fiber clusters are merged according to the third clustering threshold and the cluster similarity matrix by a hierarchical clustering algorithm to obtain multiple fiber clustering results.

[0090] The fiber clustering result consisting of a plurality of clusters is obtained, including:

[0091] It is determined whether the number of fibers in the plurality of clusters is less than or equal to a preset number. If so, the corresponding clusters are treated as noise and deleted.

[0092] Specifically, set the merge distance threshold D R (i.e. the third clustering threshold), 0 <D R <1, is set to 0.0625, as a basis for judging whether to merge small fiber clusters into large fiber clusters; the similarity between each small fiber cluster and each large fiber cluster is calculated by a fiber similarity algorithm, and the calculation process is the same as the process of calculating the similarity between each fiber in the multiple initial fiber groups by a fiber similarity algorithm.

[0093] Based on the similarity value, the small fiber cluster finds the large fiber cluster with the greatest similarity. If the similarity is greater than the merge distance threshold D R , the small clusters are merged into the large cluster, otherwise they are not merged, and the above operation is repeated until no small fiber clusters can be merged into the large fiber cluster, thereby obtaining the merged large fiber cluster, the merged small fiber clusters, the unmerged large fiber clusters and the unmerged small fiber clusters; the process is the same as the process of merging the fibers in the corresponding initial fiber groups according to the first clustering threshold and multiple initial similarity matrices through the hierarchical clustering algorithm.

[0094] The merged results are screened, and if the number of fibers in a small cluster is less than or equal to 3, the cluster is removed as a noise cluster.

[0095] In the embodiment of the present invention, after the target fiber bundle is clustered in the entire range, it is clustered again so that fibers with similar structures and morphologies can be merged into one category. The white matter fibers in the target range are clustered comprehensively, rather than just the fibers near one fiber. Similar structures are found in the entire brain to achieve high-precision category division.

[0096] like Figure 5 As shown, a fiber bundle division system based on fiber similarity provided by an embodiment of the present invention includes:

[0097] A preprocessing module, used for importing a diffusion tensor image of a brain, performing probabilistic fiber tracking on the diffusion tensor image to obtain a fiber data set, and selecting a target fiber bundle from the fiber data set, wherein the target fiber bundle includes a plurality of fibers;

[0098] A similarity calculation module is used to determine the longest fiber length among multiple fibers, divide the longest fiber length into multiple length intervals in an increasing manner, match multiple fibers with multiple length intervals, divide them into multiple initial fiber groups corresponding to the length intervals, and calculate the similarity between each fiber in the multiple initial fiber groups by a fiber similarity algorithm to obtain an initial similarity matrix of the multiple initial fiber groups;

[0099] An intra-group clustering module, used for merging the fibers in the corresponding initial fiber groups according to the first clustering threshold and the multiple initial similarity matrices by a hierarchical clustering algorithm to obtain multiple fiber groups, each fiber group including multiple different fiber clusters and multiple unmerged remaining fibers;

[0100] An inter-group clustering module is used to calculate the similarity between the plurality of fiber clusters and the plurality of remaining fibers between the plurality of fiber groups respectively by using a fiber similarity algorithm to obtain an initial clustering similarity matrix, and to merge the plurality of fiber clusters of all fiber groups with the plurality of remaining fibers by using a hierarchical clustering algorithm according to a second clustering threshold and the initial clustering similarity matrix to obtain a plurality of large fiber clusters and a plurality of small fiber clusters;

[0101] The inter-cluster clustering module is used to calculate the similarity between multiple large fiber clusters and multiple small fiber clusters respectively through a fiber similarity algorithm to obtain a cluster similarity matrix, and merge the multiple large fiber clusters and the multiple small fiber clusters according to a third clustering threshold and the cluster similarity matrix through a hierarchical clustering algorithm to obtain a fiber clustering result composed of multiple clusters.

[0102] Another technical solution of the present invention to solve the above technical problems is as follows:

[0103] A fiber bundle division system based on fiber similarity comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the fiber bundle division method based on fiber similarity as described above is implemented.

[0104] Another technical solution of the present invention to solve the above technical problems is as follows:

[0105] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the fiber bundle division method based on fiber similarity is implemented.

[0106] The above-mentioned fiber bundle division system and storage medium based on fiber similarity can refer to the implementation content and beneficial effects of the fiber bundle division method based on fiber similarity specifically described above, which will not be repeated here.

[0107] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0108] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0109] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0110] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0111] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A fiber bundle division method based on fiber similarity, characterized in that: The steps include: Importing a diffusion tensor image of the brain, performing probabilistic fiber tracking on the diffusion tensor image to obtain a fiber data set, and selecting a target fiber bundle from the fiber data set, wherein the target fiber bundle includes a plurality of fibers; Determine the longest fiber length among multiple fibers, divide the longest fiber length into multiple length intervals in an increasing manner, match multiple fibers with multiple length intervals, divide them into multiple initial fiber groups corresponding to the length intervals, calculate the similarity between each fiber in the multiple initial fiber groups by a fiber similarity algorithm, and obtain an initial similarity matrix of the multiple initial fiber groups; Merging the fibers in the corresponding initial fiber groups according to the first clustering threshold and the multiple initial similarity matrices by a hierarchical clustering algorithm to obtain multiple fiber groups, each fiber group including multiple different fiber clusters and multiple unmerged remaining fibers; The similarities between the plurality of fiber clusters and the plurality of remaining fibers among the plurality of fiber groups are calculated respectively by a fiber similarity algorithm to obtain an initial clustering similarity matrix, and the plurality of fiber clusters of all fiber groups are merged with the plurality of remaining fibers by a hierarchical clustering algorithm according to a second clustering threshold and the initial clustering similarity matrix to obtain a plurality of large fiber clusters and a plurality of small fiber clusters; The similarities between the multiple large fiber clusters and the multiple small fiber clusters are calculated respectively by a fiber similarity algorithm to obtain a cluster similarity matrix, and the multiple large fiber clusters and the multiple small fiber clusters are merged according to a third clustering threshold and the cluster similarity matrix by a hierarchical clustering algorithm to obtain a fiber clustering result consisting of multiple clusters.

2. The fiber bundle division method according to claim 1, characterized in that: The similarity between each fiber in the plurality of initial fiber groups is calculated by using a fiber similarity algorithm to obtain an initial similarity matrix of the plurality of initial fiber groups, including: S11, acquiring the same number of clustering points on any two fibers respectively according to the set sampling order, and obtaining a plurality of positive-sequence clustering points corresponding to any two fibers; S12, calculating the distances between the positive order clustering points of the same sampling order in any two fibers respectively by using the Manhattan distance expression, and obtaining multiple positive Manhattan distances of any two fibers; S13, calculating the plurality of forward Manhattan distances using a similarity expression to obtain a forward similarity; S14, performing reverse order processing on the plurality of positive order clustering points of one fiber among any two fibers to obtain a plurality of reverse order clustering points; S15, calculating the distances between the positive order clustering points and the corresponding reverse order clustering points of the same sampling order in any two fibers respectively by using the Manhattan distance expression, and obtaining multiple reverse Manhattan distances of any two fibers; S16, calculating the multiple reverse Manhattan distances by a similarity expression to obtain reverse similarity; S17, selecting a maximum value from the forward similarity and the reverse similarity to obtain an initial similarity between any two fibers; S18, repeat S11-S17 until the similarities between the fibers in all the initial fiber groups are calculated, and an initial similarity matrix corresponding to the multiple initial fiber groups is obtained.

3. The fiber bundle division method according to claim 2, characterized in that: The S11 includes: The fiber length of any fiber is calculated according to the set sampling parameters to obtain the sampling distance, multiple sampling points are obtained on any fiber according to the sampling distance, cluster points are selected from the multiple sampling points according to the set sampling order to obtain multiple positive order cluster points, and the remaining fibers are processed in this process to obtain multiple positive order cluster points corresponding to any two fibers.

4. The fiber bundle division method according to claim 1, characterized in that: The fibers in the corresponding initial fiber groups are merged according to the first clustering threshold and the multiple initial similarity matrices by using a hierarchical clustering algorithm to obtain multiple fiber groups, including: S21, selecting any initial similarity reaching the first clustering threshold from any of the initial similarity matrices as the clustering similarity, merging two fibers corresponding to the clustering similarity to obtain a fiber cluster, and in this way, processing the fibers corresponding to all the initial similarities to obtain a plurality of different fiber clusters of any of the initial fiber groups and a plurality of unmerged remaining fiber clusters; S22, calculating the similarities between the plurality of fiber clusters and the plurality of remaining fiber clusters respectively by using a fiber similarity algorithm to obtain a clustering similarity matrix; S23, looping S21-S22, processing the clustering similarity matrix generated each time, the plurality of fiber clusters and the plurality of remaining fiber clusters, until no similarity reaching the first clustering threshold can be screened out, and obtaining a fiber group consisting of a plurality of different fiber clusters and a plurality of unmerged remaining fibers; In this way, S21-S23 are performed to process the fibers in all the initial fiber groups to obtain multiple fiber groups.

5. The fiber bundle dividing method according to claim 1, characterized in that: The fiber clustering result consisting of a plurality of clusters is obtained, including: It is determined whether the number of fibers in the plurality of clusters is less than or equal to a preset number. If so, the corresponding clusters are treated as noise and deleted.

6. A fiber bundle partitioning system based on fiber similarity, characterized in that: include: A preprocessing module, used for importing a diffusion tensor image of a brain, performing probabilistic fiber tracking on the diffusion tensor image to obtain a fiber data set, and selecting a target fiber bundle from the fiber data set, wherein the target fiber bundle includes a plurality of fibers; A similarity calculation module is used to determine the longest fiber length among multiple fibers, divide the longest fiber length into multiple length intervals in an increasing manner, match multiple fibers with multiple length intervals, divide them into multiple initial fiber groups corresponding to the length intervals, and calculate the similarity between each fiber in the multiple initial fiber groups by a fiber similarity algorithm to obtain an initial similarity matrix of the multiple initial fiber groups; An intra-group clustering module, used for merging the fibers in the corresponding initial fiber groups according to the first clustering threshold and the multiple initial similarity matrices by a hierarchical clustering algorithm to obtain multiple fiber groups, each fiber group including multiple different fiber clusters and multiple unmerged remaining fibers; An inter-group clustering module is used to calculate the similarity between the plurality of fiber clusters and the plurality of remaining fibers between the plurality of fiber groups respectively by using a fiber similarity algorithm to obtain an initial clustering similarity matrix, and to merge the plurality of fiber clusters of all fiber groups with the plurality of remaining fibers by using a hierarchical clustering algorithm according to a second clustering threshold and the initial clustering similarity matrix to obtain a plurality of large fiber clusters and a plurality of small fiber clusters; The inter-cluster clustering module is used to calculate the similarity between multiple large fiber clusters and multiple small fiber clusters respectively through a fiber similarity algorithm to obtain a cluster similarity matrix, and merge the multiple large fiber clusters and the multiple small fiber clusters according to a third clustering threshold and the cluster similarity matrix through a hierarchical clustering algorithm to obtain a fiber clustering result composed of multiple clusters.

7. The fiber bundle dividing system according to claim 6, characterized in that: In the similarity calculation module, the similarity between each fiber in the multiple initial fiber groups is calculated by a fiber similarity algorithm to obtain an initial similarity matrix of the multiple initial fiber groups, including: The same number of cluster points are obtained on any two fibers according to the set sampling order, and multiple positive order cluster points corresponding to any two fibers are obtained; The distances between the positive order clustering points of the same sampling order in any two fibers are calculated by using the Manhattan distance expression, and multiple positive Manhattan distances of any two fibers are obtained; Calculating the plurality of forward Manhattan distances using a similarity expression to obtain a forward similarity; Reverse the order of the plurality of positive order clustering points of one of any two fibers to obtain a plurality of reverse order clustering points; The distances between the positive order clustering points and the corresponding reverse order clustering points of the same sampling order in any two fibers are calculated by using the Manhattan distance expression, and multiple reverse Manhattan distances of any two fibers are obtained; Calculating the multiple reverse Manhattan distances through a similarity expression to obtain reverse similarity; Selecting a maximum value from the forward similarity and the reverse similarity to obtain an initial similarity between any two fibers; The above process is repeated until the similarities between the fibers in all the initial fiber groups are calculated, and the initial similarity matrices corresponding to the multiple initial fiber groups are obtained.

8. The fiber bundle dividing system according to claim 7, characterized in that: In the similarity calculation module, the same number of cluster points are obtained on any two fibers according to the set sampling order to obtain multiple positive order cluster points corresponding to any two fibers, including: The fiber length of any fiber is calculated according to the set sampling parameters to obtain the sampling distance, multiple sampling points are obtained on any fiber according to the sampling distance, cluster points are selected from the multiple sampling points according to the set sampling order to obtain multiple positive order cluster points, and the remaining fibers are processed in this process to obtain multiple positive order cluster points corresponding to any two fibers.

9. A fiber bundle partitioning system based on fiber similarity, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the fiber bundle division method based on fiber similarity according to any one of claims 1 to 5 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the fiber bundle division method based on fiber similarity according to any one of claims 1 to 5 is implemented.

Images