WO2024083058A1 - 一种大脑纤维束异常区域精准定位系统 - Google Patents

一种大脑纤维束异常区域精准定位系统 Download PDF

Info

Publication number
WO2024083058A1
WO2024083058A1 PCT/CN2023/124640 CN2023124640W WO2024083058A1 WO 2024083058 A1 WO2024083058 A1 WO 2024083058A1 CN 2023124640 W CN2023124640 W CN 2023124640W WO 2024083058 A1 WO2024083058 A1 WO 2024083058A1
Authority
WO
WIPO (PCT)
Prior art keywords
fiber
fiber bundle
brain
diffusion
magnetic resonance
Prior art date
Application number
PCT/CN2023/124640
Other languages
English (en)
French (fr)
Inventor
张瑜
孙超良
王志超
张欢
钱浩天
蒋田仔
Original Assignee
之江实验室
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 之江实验室 filed Critical 之江实验室
Publication of WO2024083058A1 publication Critical patent/WO2024083058A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
    • A61B5/004—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part
    • A61B5/0042—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part for the brain
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves 
    • A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/40—Detecting, measuring or recording for evaluating the nervous system
    • A61B5/4058—Detecting, measuring or recording for evaluating the nervous system for evaluating the central nervous system
    • A61B5/4064—Evaluating the brain
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30016—Brain

Definitions

  • the present invention relates to the field of neuroimaging data analysis, and in particular to a system for accurately locating abnormal areas of brain fiber bundles.
  • Diffusion-weighted MRI is a quantitative MRI imaging method.
  • Conventional MRI applies a radio frequency pulse of a certain frequency to the human body in a magnetic field, which excites the hydrogen protons in the human body and produces resonance. After the pulse stops, the protons generate magnetic resonance signals during the relaxation process. After encoding, receiving and reconstructing the magnetic resonance signals, a static structural image is generated.
  • Diffusion-weighted MRI measures the diffusion movement of water molecules in the human body, that is, it measures the displacement of water within a predetermined diffusion time.
  • the diffusion movement of water molecules is a three-dimensional random movement, and the degree of diffusion in all directions is the same. This phenomenon is called diffusion isotropy.
  • the diffusion movement of water molecules in three-dimensional space is subject to various restrictions. If the diffusion movement of water molecules is hindered by cell membranes, macromolecules, etc., the displacement of water molecules will be reduced, and the water molecules will be hindered. For example, in a myelinated nerve fiber bundle, the diffusion movement of water molecules along the fiber direction will be much greater than the diffusion in the direction of the myelin sheath. This phenomenon is called diffusion anisotropy.
  • the diffusion tensor imaging (DTI) model is the development and improvement of the diffusion weighted magnetic resonance imaging technology.
  • the diffusion tensor model is a three-dimensional model that quantifies the anisotropic signal of water molecule diffusion. It uses diffusion-sensitive gradient pulses to expand the diffusion effect of water molecules to study the differences in the diffusion movement of water molecules in different tissues, so that the microstructure of brain tissue can be displayed more finely.
  • the Neurite orientation dispersion and density imaging (NODDI) model divides the microenvironment into three situations: intracellular, extracellular, and cerebrospinal fluid. Each microenvironment affects the diffusion of water molecules in different ways, which can be used to evaluate the complexity of the microstructure of axons and dendrites, and can reflect the information of different tissues in the brain.
  • Diffusion MRI detects the microstructural characteristics of the brain and the direction of fiber bundles by measuring the difference in the diffusion of water molecules. Using these diffusion characteristics, some commonly used parameters can be obtained. For example, through diffusion tensor imaging, we can calculate indicators such as fractional anisotropy (FA) and mean diffusivity (MD). We can also calculate the intra-neurite volume fraction (ICVF) and orientation dispersion index (ODI) through neurite directional dispersion and density imaging. And the corresponding fiber bundle pathways can be extracted through the fiber tracking results at the whole brain level.
  • FA fractional anisotropy
  • MD mean diffusivity
  • ICVF intra-neurite volume fraction
  • ODI orientation dispersion index
  • FA reflects the ratio of the anisotropic part of the diffusion to the total value of the diffusion tensor, which can be used to show the direction of the nerve conduction bundles in the white matter of the brain, and can observe the integrity and connectivity of the tissue structure, which is conducive to the judgment of the degree and range of damage to the white matter fiber bundles caused by various diseases.
  • MD reflects the range of diffusion movement of water molecules per unit time, such as diffusion. The number of water molecules with limited diffusion increases, which will cause a decrease in MD in the area.
  • ICVF can reflect nerve density, such as the density of axons.
  • ODI can quantify the discreteness of neurite directions, such as fan-shaped fibers and crossing fibers. These imaging indicators provide multiple angles of interpretation for the study of various diseases.
  • Fiber tracking is an important method for fiber bundle pathway analysis.
  • the traditional fiber tracking method based on diffusion tensor imaging constructs the whole-brain fiber bundle according to the main direction of the tensor and the preset deflection angle threshold, FA threshold and other parameters, and then tracks the specific fiber bundle by setting the ROI, and calculates the number of fiber bundles, average FA and other related indicators, but it cannot solve the problem of crossing fibers.
  • Support vector machines are a commonly used machine learning classification model.
  • the basic idea of SVM learning is to solve the separation hyperplane that can correctly divide the training data set and has the largest geometric interval.
  • the weight of the feature obtained by SVM can reflect which nodes of which fiber bundles have significant differences between the disease group and the healthy group.
  • the purpose of the present invention is to propose a precise positioning system for abnormal brain fiber bundles in view of the deficiencies of the prior art.
  • the present invention estimates the response function of the diffusion magnetic resonance data and reconstructs the fiber direction distribution diffusion model through spherical constrained deconvolution.
  • the spherical deconvolution model can estimate the diffusion distribution of each voxel.
  • the probabilistic fiber tracking method is used to randomly set seed points in the region of interest, and fiber bundle tracking is performed from each seed point until the specified number of fiber bundles is completed.
  • the fiber bundles are screened by spherical deconvolution filtering of the fiber bundles, and the fibers with physiological significance are retained, so that the number of local fibers tracked is proportional to the real local nerve fiber density.
  • the fiber bundle pathway that meets the region of interest can be obtained based on seed point fiber tracking.
  • the extracted fiber bundles of interest can be further segmented (such as 100 equal divisions), and the imaging indicators on each small segment are counted to achieve a more accurate analysis of the fiber bundle pathway.
  • the disease group and the healthy group are classified by the SVM method to locate the abnormal fiber bundle nodes.
  • the present invention can obtain the fiber bundle pathway of interest, divide it finely, and effectively utilize the imaging indicators of each diffusion model.
  • the abnormal fiber bundle segment can be accurately located through the machine learning method, and the lesions of brain white matter related diseases can be accurately located.
  • the present invention is realized by the following technical scheme: a system for accurately locating abnormal areas of brain fiber bundles, the system comprising the following modules:
  • a diffusion magnetic resonance data acquisition module is used to obtain diffusion magnetic resonance data of a disease group and corresponding diffusion magnetic resonance data of a healthy group;
  • a diffusion magnetic resonance data preprocessing module is used to perform noise reduction and correction processing on the diffusion magnetic resonance data collected by the diffusion magnetic resonance data collection module;
  • Whole-brain fiber tracking module used to extract fiber connections of the whole brain based on preprocessed diffusion MRI data
  • the fiber bundle pathway module of interest which is used to customize the fiber bundle pathway or extract the fiber bundle pathway based on the brain fiber bundle template
  • the fiber bundle pathway projection and segmentation module is used to project the fiber bundle pathway onto the fiber connection results of the whole brain and segment it, and define each segment as a node;
  • the fiber bundle node imaging index extraction module is used to calculate the anisotropy fraction, average diffusion rate, neurite volume ratio and directional dispersion of the diffusion magnetic resonance data, so as to obtain the imaging index of each node of each fiber bundle pathway;
  • the machine learning classification and abnormal node location module is used to classify disease groups and healthy groups using imaging indicators through machine learning methods, and to locate which nodes on which fiber bundle pathways have undergone abnormal changes under different diseases.
  • the diffusion magnetic resonance data preprocessing module is used to perform denoising on the diffusion magnetic resonance data based on the PCA method, perform distortion correction based on the inverse phase encoding image, perform head motion correction on the diffusion magnetic resonance data, and perform eddy current correction on the diffusion magnetic resonance data.
  • the whole-brain fiber tracking module is used to estimate the response function of the preprocessed diffusion magnetic resonance data and reconstruct the fiber direction distribution diffusion model through spherical constrained deconvolution, and perform whole-brain fiber tracking based on the reconstructed model; the fiber bundles are screened using the spherical deconvolution filtering method of the fiber bundles, and only the fiber bundles with physiological significance are retained.
  • the module for defining fiber bundle pathways of interest is used to define a starting region of interest, an ending region of interest, a region of interest through, and a region of interest avoidance on a standard brain template, and to perform seed point-based fiber tracking to obtain a fiber bundle pathway that satisfies the region of interest.
  • the module for defining fiber bundle pathways of interest can use fiber bundle pathways pre-defined on a fiber bundle atlas.
  • the fiber bundle pathway projection and segmentation module is used to nonlinearly align the fiber bundle region of interest defined on the standard brain template to the structural image of each subject, and then linearly align it to the individual space of the diffusion image of each subject; seed point-based fiber tracking is performed in the individual space of the subject to obtain the fiber bundle pathway that satisfies the region of interest.
  • the fiber bundle can be directly linearly aligned to the subject's individual space, the fiber bundle of interest can be extracted from the whole-brain fiber tracking results, and the fiber bundle can be evenly divided into several small segments according to the length, and each small segment is defined as a node.
  • the fiber bundle node image index extraction module is used to fit the diffusion tensor imaging DTI model to the diffusion magnetic resonance data, calculate the anisotropy fraction FA and the mean diffusion coefficient MD value of the whole brain, fit the neurite directional diffusion and density imaging model NODDI model to the diffusion magnetic resonance data, and calculate the intracellular volume fraction ICVF and directional diffusion fraction ODI value of the whole brain; the average value of the above indicators is calculated for each node of each fiber bundle.
  • the machine learning classification and abnormal node location module is used to classify each neural fiber based on the SVM classifier.
  • the dimensional node features were used as the input of the SVM classifier, and the group to which the subject belonged was used as the label as the output of the SVM classifier.
  • the SVM classifier training set used 10-fold cross-validation to obtain the feature weights of 10 models respectively, sorted the features from large to small according to the weights, took the top 10% features, and counted the node features that appeared repeatedly in the top 10% features, so as to determine which nodes on which fiber bundle pathways had abnormal changes under different diseases.
  • the method proposed in the present invention can obtain the fiber bundle pathway of interest through diffusion magnetic resonance data, divide it finely, and effectively utilize the imaging indicators of each diffusion model. These brain imaging indicators have been confirmed to be closely related to various diseases in previous studies. Through machine learning methods, abnormal fiber bundle segments can be accurately located to achieve accurate positioning of lesions of brain white matter-related diseases.
  • the present invention adopts a spherical constrained deconvolution reconstruction method to estimate the fiber orientation function on each voxel to reconstruct the fiber distribution on each voxel, which can effectively solve the problem of fiber crossing.
  • FIG1 is a schematic diagram of the structure of a system for accurately locating abnormal areas of brain fiber bundles provided by the present invention.
  • FIG. 2 is a schematic diagram of the construction of the fiber bundle pathway of the present invention.
  • FIG3 is a schematic diagram of the AUC value of the classification model test of the present invention.
  • the present invention proposes a precise positioning system for abnormal areas of brain fiber bundles.
  • the system can obtain the fiber bundle pathways of interest, divide them finely, and effectively utilize the imaging indicators of each diffusion model.
  • the abnormal fiber bundle segments can be accurately positioned through machine learning methods, thereby achieving precise positioning of lesions of white matter-related diseases.
  • the overall system structure schematic diagram is shown in Figure 1, including a diffusion magnetic resonance data acquisition module, a diffusion magnetic resonance data preprocessing module, a whole-brain fiber tracking module, a module for defining fiber bundle pathways of interest, a fiber bundle pathway projection and segmentation module, a fiber bundle node image indicator extraction module, and a machine learning classification and abnormal node positioning module;
  • the diffusion magnetic resonance data acquisition module is used to obtain the disease group diffusion magnetic resonance data and the corresponding healthy group diffusion magnetic resonance data;
  • the diffusion magnetic resonance data preprocessing module is used to preprocess the diffusion magnetic resonance data collected by the diffusion magnetic resonance data acquisition module, and the preprocessing includes: image denoising, distortion correction, extraction of b0 images for signal normalization, head movement and eddy current correction.
  • the whole-brain fiber tracking module is used Based on the distribution of the data b vector, the image is reconstructed by a diffusion model, the response function and the fiber direction distribution function are calculated, and the fiber tracking of the whole brain is performed.
  • the module for defining the fiber bundle pathway of interest is used to define the starting region of interest and the ending region of interest on the standard brain template MNI152NLinin2009cAsym, and to perform fiber tracking based on seed points through the region of interest and avoiding the region of interest, so as to obtain the fiber bundle pathway that satisfies the region of interest.
  • the fiber bundle pathway predefined on the fiber bundle atlas can also be used. As shown in Figure 2.
  • the fiber bundle pathway projection and segmentation module is used to nonlinearly register the fiber bundle region of interest defined on the standard brain template MNI152NLinin2009cAsym to the structural image of each subject, and then linearly register it to the diffusion image individual space of each subject; then, the fiber tracking based on the seed point is performed in the individual space of the subject to obtain the fiber bundle pathway that satisfies the region of interest.
  • the fiber bundle pathway can be directly linearly registered to the individual space of the subject, and then the fiber bundle is evenly divided into one hundred small segments according to the length, and each small segment is regarded as a node.
  • the fiber bundle node image index extraction module is used to calculate the anisotropy fraction FA, mean diffusion coefficient MD, intracellular volume fraction ICVF and directional diffusion fraction ODI value of each brain region through diffusion magnetic resonance, and calculate the average value of the above indicators on the small segment of fiber bundle for each node of each fiber bundle.
  • the machine learning classification and abnormal node positioning module is used to classify the disease group and the healthy group based on the SVM classifier.
  • the weight of the feature is obtained by SVM, which can reflect which nodes of which fiber bundles have significant differences between the disease group and the healthy group.
  • the diffusion magnetic resonance data acquisition module is used to obtain the diffusion magnetic resonance data of the disease group and the corresponding diffusion magnetic resonance data of the healthy group; the example of the present invention obtains the clinical data collected by the hospital, and the data are divided into type 0 patients, type 1 patients and healthy people according to clinical manifestations. The data are sorted and quality controlled, and the final subjects included 121 type 0 patients, 107 type 1 patients, and 109 healthy people.
  • the diffusion magnetic resonance data preprocessing module is used to perform noise reduction and correction processing on the diffusion magnetic resonance data collected by the diffusion magnetic resonance data acquisition module: specifically: the preprocessing of the diffusion magnetic resonance image data is completed based on the QSIPrep software package, and specifically includes the following processes: dicom to BIDS format conversion, and the data is organized into a standard format. Since the diffusion magnetic resonance image has a low signal-to-noise ratio, the image is denoised by the principal component analysis method (MP-PCA). Due to the inhomogeneity of the B0 field, the diffusion magnetic resonance image will be distorted in the phase gradient direction, so the image needs to be distorted (N4algorithm).
  • MP-PCA principal component analysis method Due to the inhomogeneity of the B0 field, the diffusion magnetic resonance image will be distorted in the phase gradient direction, so the image needs to be distorted (N4algorithm).
  • the eddy current effect generated by the switching of the gradient magnetic field will produce eddy currents that hinder the gradient change.
  • This additional disturbance will affect the change of the gradient field, causing its waveform to be severely distorted, causing the image to have geometric deformation, artifacts and other distortions.
  • This effect is called the eddy current effect, so the data needs to be eddy corrected (eddy).
  • the breathing and head movement of the subject during the scanning process will also affect the spatial position of the data, so the data needs to be corrected for head movement and eddy currents (eddy).
  • the whole-brain fiber tracking module is used to extract the fiber connections of the whole brain based on the preprocessed diffusion magnetic resonance data; specifically, the response function of the preprocessed diffusion magnetic resonance data is estimated using MRtrix3 software and the spherical constraint is used to extract the fiber connections of the whole brain.
  • the spherical deconvolution model can estimate the diffusion distribution of each voxel.
  • the fiber tracking uses the probabilistic fiber tracking method, sets seed points in the region of interest, and starts fiber bundle tracking from each seed point until the specified number of fiber bundles is completed.
  • the present invention sets 10 million fiber bundles; finally, the fiber bundle spherical deconvolution filtering method is used to screen the fiber tracking results to retain the fiber bundles with physiological significance.
  • the module for defining the fiber bundle pathway of interest is used to customize the fiber bundle pathway or extract the fiber bundle pathway based on the brain fiber bundle template; specifically: define the starting region of interest, the ending region of interest, the passing region of interest and the avoiding region of interest on the standard brain template MNI152NLinin2009cAsym, and modify the region of interest, such as segmenting the left and right brains, corrosion, and removing the outside brain.
  • the selected region of interest is subjected to fiber tracking based on the seed point, starting from the starting region of interest, passing through or avoiding the corresponding region of interest, and reaching the ending region of interest, and finally obtaining the fiber bundle pathway that satisfies the region of interest.
  • the corresponding fiber bundle pathway can also be directly extracted from the public fiber bundle template.
  • the fiber bundles of the left and right hemispheres are preferably separately counted.
  • the present invention extracts a total of 80 different fiber bundle pathways. Two of the 80 pathways were not proposed from any subject, so the remaining 78 pathways were statistically analyzed (most pathways had more than 100 subjects in each group, and a small number of pathways had results for less than 10 subjects in each group).
  • the fiber bundle pathway projection and segmentation module is used to project the fiber bundle pathway onto the fiber connection results of the whole brain and segment, and each segmentation is defined as a node; Specifically: Based on the characteristics of diffusion magnetic resonance data, the calculation of diffusion magnetic resonance index needs to be carried out in the individual space of the subject.
  • the present invention needs to nonlinearly register the fiber bundle region of interest defined on the standard brain template MNI152NLinin2009cAsym to the structural image of each subject, and then linearly register to the diffusion image individual space of each subject; Then, fiber tracking based on seed points is carried out in the individual space of the subject, and the fiber bundle pathway that meets the region of interest is obtained, especially, in the case of having obtained the fiber bundle pathway, the fiber bundle can be directly linearly registered to the individual space of the subject, and the fiber bundle of interest is extracted from the fiber tracking result of the whole brain.
  • the extracted fiber bundle of interest can be further segmented (such as 100 equal divisions), and each small segment is recognized as a node.
  • the fiber bundle node imaging index extraction module is used to calculate the anisotropy fraction, average diffusion rate, neurite volume ratio and directional dispersion of the diffusion magnetic resonance data, so as to obtain the imaging index at each node of each fiber bundle pathway; specifically: the diffusion magnetic resonance data are fitted with the diffusion tensor imaging DTI model to calculate the anisotropy fraction FA and the average diffusion coefficient MD value of the whole brain; and the diffusion magnetic resonance data are fitted with the neurite directional diffusion and density imaging model NODDI model to calculate the intracellular volume fraction ICVF and the directional diffusion fraction ODI value of the whole brain; finally, for each node of each fiber bundle, the average value of the above indicators at the node is calculated.
  • the machine learning classification and abnormal node location module is used to classify the abnormal nodes between the disease group and the healthy group through machine learning.
  • Use imaging indicators for classification to locate which nodes on which fiber bundle pathways have undergone abnormal changes under different diseases; specifically: count the fiber bundles tracked by the healthy group and the disease group, use the 100 node values of these fiber bundles as features, construct a feature set, use 0 as the healthy group label and 1 as the disease group label, construct a label set, and the entire data set consists of a feature set and a corresponding label set; secondly, randomly divide the data set into a training set and a test set in an 8:2 ratio based on the subject; then, use the SVM classifier for binary classification prediction, the SVM classifier kernel function is the Linear kernel, and use the feature of each nerve fiber node as the input of the SVM classifier, and the group to which the subject belongs is the test set.
  • the labels are used as the output of the SVM classifier, wherein the training set uses 10-fold cross validation, and the coef_ parameter of the SVM model is used to obtain the weights of all node features of the 10 models after training, respectively, and the features are sorted from large to small according to the weights, and the features whose weight values of the 10 models account for the top 10% are taken, and the node features that appear repeatedly in the top 10% of the features are counted, so as to determine which nodes on which fiber bundle pathways have abnormal changes under different diseases; finally, the test set is used to test the model to obtain the classification accuracy and AUC value, and the AUC value of the classification model test of type 0 patients and healthy people is shown in FIG3 , and the AUC in the embodiment of the present invention is 0.64.
  • fiber bundle extraction and segmentation, image index calculation and machine learning abnormal node localization are performed on the clinical data set, and the FA values of the middle and posterior halves of the left and right parietopontine bundles and the front and rear ends of the corticothalamic pathway are node features that appear repeatedly in the top 10% of the features in the classification of type 0 patients and healthy people, as shown in Table 1 below.
  • the headers in Table 1 are the names of the fiber pathways, and the numbers are the locations of the nodes with large weights during classification.
  • the abnormalities of these pathways are closely related to the clinical manifestations of the disease, such as inconvenience in movement and increased body fat percentage, indicating that the results obtained by the system have good interpretability.
  • the system can accurately locate the abnormal nodes of the fiber bundle pathways and obtain the fiber bundle segments with intergroup differences between the disease group and the healthy group.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Neurology (AREA)
  • Radiology & Medical Imaging (AREA)
  • Pathology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Biomedical Technology (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Biophysics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Neurosurgery (AREA)
  • Psychology (AREA)
  • Physiology (AREA)
  • Quality & Reliability (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

本发明公开了一种大脑纤维束异常区域精准定位系统,该系统从扩散磁共振数据中提取全脑的纤维连接，通过自定义纤维束通路或者基于大脑纤维束模板提取纤维束通路。将选定的纤维束通路投射到全脑的纤维连接结果上并进行精细地分段。用扩散磁共振数据计算各向异性分数，平均扩散率，神经突内容积比以及方向分散度等影像学指标，从而得到每条纤维束通路每个节点上的影像学指标，用机器学习的方法在疾病组和健康组之间用这些影像学指标做分类，可以精准定位不同疾病下哪些纤维束通路上的哪些节点发生了异常变化。本发明采用球面约束反卷积重建方法，估计每个体素上的纤维走向函数，来重建每个体素上的纤维分布，可以有效解决纤维交叉的问题。

Description

一种大脑纤维束异常区域精准定位系统 技术领域
本发明涉及神经影像数据分析领域，尤其涉及一种大脑纤维束异常区域精准定位系统。
背景技术
扩散加权磁共振是一种定量的磁共振成像方法。传统磁共振是通过对磁场中的人体施加某种特定频率的射频脉冲，使人体内的氢质子受到激励而产生共振现象。脉冲停止后，质子在弛豫过程中产生磁共振信号，通过对磁共振信号的编码、接收和重建后，生成静态的结构图像。而扩散加权磁共振测量的是人体内水分子的扩散运动，也就是测量预定扩散时间内水的位移。
在匀质的水中，水分子的扩散运动是一个三维的随机运动，在各个方向的扩散程度相同，这种现象称为扩散的各向同性，而在人体中，水分子在三维空间中的扩散运动会受到各种限制，如果水分子的扩散运动被细胞膜、大分子等阻碍，那么水分子的位移就会减少，水分子的受阻扩散。比如在有髓鞘的神经纤维束中，水分子沿着纤维走向的扩散运动会远大于向髓鞘方向的扩散，这种现象称为扩散的各向异性。
基于这个原理，扩散张量成像(diffusion tensor imaging,DTI)模型是对磁共振扩散加权成像技术的发展与改进。扩散张量模型是一个三维的立体模型，量化了水分子扩散的各向异性的信号，利用扩散敏感梯度脉冲将水分子扩散效应扩大，来研究不同组织中水分子扩散运动的差异，使脑组织的微结构能够更加精细地显示。神经突方向分散度和密度成像(Neurite orientation dispersion and density imaging,NODDI)模型把微环境分为了三种情况：细胞内、细胞外、脑脊液。每一种微环境影响水分子弥散的方式都不同，可用于评估轴突和树突微结构的复杂性，可以反映脑组织中不同组织的信息。
扩散磁共振成像通过测量水分子扩散差异来检测大脑的微结构特性以及纤维束走向。利用这些扩散特性，可以得到一些常用的参数。比如可以通过扩散张量成像，计算各向异性分数(Fractional anisotropy,FA)、平均扩散率(Mean diffusivity,MD)等指标，也可通过神经突方向分散度和密度成像，计算神经突内容积比(Intra-neurite volume fraction,ICVF)以及方向分散度(Orientation dispersion index,ODI)等。并且可以通过全脑层面的纤维追踪结果，提取出相应的纤维束通路。其中，FA反映了扩散的各项异性部分与扩散张量总值的比值，可用于显示脑白质内神经传导束的走行方向，可以观察组织结构的完整性和连通性，利于对各种疾病引起的白质纤维束损害程度及范围的判断。MD反映了水分子单位时间内扩散运动的范围，比如扩 散受限的水分子增加，这就会引起该区域MD的下降，ICVF可以反映神经密度，比如轴突的密度。ODI可以量化神经突方向的离散度，比如扇形纤维和交叉纤维。这些影像学指标为研究各类疾病提供了多种角度的解读。
纤维追踪是进行纤维束通路分析的重要方法。传统基于弥散张量成像的纤维追踪方法根据张量主方向以及预设的偏转角阈值、FA阈值等参数对全脑纤维束进行构建，而后通过设定ROI的方式追踪特定纤维束，并计算纤维束的数目、平均FA等相关指标，但是无法解决交叉纤维的问题。
支持向量机(Support vector machines,SVM)是一种常用的机器学习分类模型，SVM学习的基本想法是求解能够正确划分训练数据集并且几何间隔最大的分离超平面。通过SVM得到特征的权重，可以反映哪些纤维束的哪些节点在疾病组和健康组之间有显著的差异。
以往的纤维束成像分析方法，局限于全脑或是纤维通路的整体分析，且只进行简单的统计比较组间差异。
发明内容
本发明的目的在于针对现有技术的不足，提出一种大脑纤维束异常区域精准定位系统。本发明通过对扩散磁共振数据进行响应函数的估计以及通过球面约束反卷积重建纤维方向分布扩散模型，球面反卷积模型可以估计出每个体素的弥散分布。然后用概率性纤维追踪法，随机的在感兴趣区域内设定种子点，从每个种子点开始进行纤维束追踪，直到完成指定数目的纤维束，然后用纤维束的球面反卷积滤波对纤维束进行筛选，保留有生理意义的纤维，使得追出的局部纤维数目正比于真实的局部神经纤维密度。通过定义起始感兴趣区域，终止感兴趣区域，通过感兴趣区域以及避开感兴趣区域，进行基于种子点的纤维追踪，可以获得满足感兴趣区域的纤维束通路。通过自动定量方法，可对提取出来的感兴趣的纤维束进行进一步分段(如100等分)，统计每一小段上的影像学指标，做到对纤维束通路更精确的分析。最后用SVM方法对疾病组和健康组进行分类，定位纤维束异常节点。本发明可以得到感兴趣的纤维束通路，对其进行精细的划分，并且有效利用了各个扩散模型的影像学指标。通过机器学习的方法可以精准定位到异常的纤维束段，实现脑白质相关疾病病灶的精准定位。
本发明是通过以下技术方案来实现的：一种大脑纤维束异常区域精准定位系统，该系统包括如下模块：
扩散磁共振数据采集模块，用于获取疾病组扩散磁共振数据以及相应的健康组扩散磁共振数据；
扩散磁共振数据预处理模块，用于对扩散磁共振数据采集模块采集到的扩散磁共振数据进行降噪和矫正处理；
全脑纤维追踪模块，用于基于预处理后的扩散磁共振数据提取全脑的纤维连接；
定义感兴趣纤维束通路模块，用于自定义纤维束通路或者基于大脑纤维束模板提取纤维束通路；
纤维束通路投射与分段模块，用于将纤维束通路投射到全脑的纤维连接结果上并进行分段，将每个分段定义为节点；
纤维束节点影像指标提取模块，用于计算扩散磁共振数据的各向异性分数、平均扩散率、神经突内容积比以及方向分散度，从而得到每条纤维束通路每个节点上的影像学指标；
机器学习分类与异常节点定位模块，用于通过机器学习的方法在疾病组和健康组之间用影像学指标做分类，定位不同疾病下哪些纤维束通路上的哪些节点发生了异常变化。
进一步地，所述扩散磁共振数据预处理模块用于对扩散磁共振数据进行基于PCA方法的去噪，并进行基于反相位编码图像的畸变矫正以及对扩散磁共振数据进行头动矫正和对扩散磁共振数据进行涡流矫正。
进一步地，所述全脑纤维追踪模块用于对预处理后的扩散磁共振数据进行响应函数的估计以及通过球面约束反卷积重建纤维方向分布扩散模型，基于重建的模型进行全脑纤维追踪；用纤维束的球面反卷积滤波方法对纤维束进行筛选，仅保留有生理意义的纤维束。
进一步地，所述定义感兴趣纤维束通路模块用于在标准脑模板上定义起始感兴趣区域、终止感兴趣区域、通过感兴趣区域以及避开感兴趣区域，并进行基于种子点的纤维追踪，获得满足感兴趣区域的纤维束通路。
进一步地，所述定义感兴趣纤维束通路模块能够使用纤维束图谱上预先定义的纤维束通路。
进一步地，所述纤维束通路投射与分段模块用于将标准脑模板上的定义的纤维束感兴趣区域非线性配准到每个被试的结构像上，进而线性配准到每个被试的扩散像个体空间；在被试的个体空间进行基于种子点的纤维追踪，获得满足感兴趣区域的纤维束通路。
进一步地，在已经获得纤维束通路的情况下，能够直接对纤维束进行线性配准至被试的个体空间，从全脑的纤维追踪结果中提取出感兴趣的纤维束，将纤维束根据长度平均分成若干小段，每小段定义为一个节点。
进一步地，所述纤维束节点影像指标提取模块用于对扩散磁共振数据进行弥散张量成像DTI模型拟合，计算全脑的各向异性分数FA和平均弥散系数MD值，对扩散磁共振数据进行神经突定向弥散和密度成像模型NODDI模型拟合，计算全脑的细胞内体积分数ICVF和方向扩散分数ODI值；对每条纤维束的每个节点计算上述指标的平均值。
进一步地，所述机器学习分类与异常节点定位模块用于基于SVM分类器，以每条神经纤 维节点特征作为SVM分类器的输入，以被试所在的组别为标签，作为SVM分类器的输出，SVM分类器训练集使用10折交叉验证，分别获取10个模型的特征权重，根据权重对特征进行从大到小排序，取前10％特征，并统计前10％特征中重复出现的节点特征，从而确定不同疾病下哪些纤维束通路上的哪些节点发生了异常变化。
本发明的有益效果：本发明提出的方法能够通过扩散磁共振数据，得到感兴趣的纤维束通路，对其进行精细的划分，并且有效利用了各个扩散模型的影像学指标，这些脑影像指标在以往研究中被证实与各类疾病密切相关，通过机器学习的方法可以精准定位到异常的纤维束段，实现脑白质相关疾病病灶的精准定位。本发明采用球面约束反卷积重建方法，估计每个体素上的纤维走向函数，来重建每个体素上的纤维分布，可以有效解决纤维交叉的问题。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图做简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动前提下，还可以根据这些附图获得其他附图。
图1是本发明提供的一种大脑纤维束异常区域精准定位系统结构示意图。
图2是本发明纤维束通路构建示意图。
图3是本发明分类模型测试的AUC值示意图。
具体实施方式
下面将结合附图对本发明作进一步的说明。为了使本领域的人员更好地理解本申请中的技术方案，下面将结合附图对本发明作进一步的说明。但这仅仅是本申请的一部分实施例，而不是全部的实施例。基于本申请所述的具体实施例，本领域的其他人员在没有做出创造性劳动的前提下所获得的其他实施例，都应当落在本发明的构思范围之内。
总体而言，本发明提出一种大脑纤维束异常区域精准定位系统。该系统可以得到感兴趣的纤维束通路，对其进行精细的划分，并且有效利用了各个扩散模型的影像学指标。通过机器学习的方法可以精准定位到异常的纤维束段，实现脑白质相关疾病病灶的精准定位。整体系统结构示意图如图1所示，包括扩散磁共振数据采集模块、扩散磁共振数据预处理模块、全脑纤维追踪模块、定义感兴趣纤维束通路模块、纤维束通路投射与分段模块、纤维束节点影像指标提取模块和机器学习分类与异常节点定位模块；所述扩散磁共振数据采集模块用于获取疾病组扩散磁共振数据以及相应的健康组扩散磁共振数据；所述扩散磁共振数据预处理模块用于对扩散磁共振数据采集模块采集到的扩散磁共振数据进行预处理，预处理包括：图像去噪、畸变矫正、提取b0图像做信号归一化、头动和涡流矫正。所述全脑纤维追踪模块用 于基于数据b向量分布情况对图像进行弥散模型重建，计算响应函数和纤维方向分布函数，进行全脑的纤维追踪。所述定义感兴趣纤维束通路模块用于在标准脑模板MNI152NLinin2009cAsym上定义起始感兴趣区域，终止感兴趣区域，通过感兴趣区域以及避开感兴趣区域，进行基于种子点的纤维追踪，获得满足感兴趣区域的纤维束通路，特别地，也可以使用纤维束图谱上预先定义的纤维束通路。如图2所示。所述纤维束通路投射与分段模块用于将标准脑模板MNI152NLinin2009cAsym上的定义的纤维束感兴趣区域非线性配准到每个被试的结构像上，进而线性配准到每个被试的扩散像个体空间；然后在被试的个体空间进行基于种子点的纤维追踪，获得满足感兴趣区域的纤维束通路，特别地，在已经获得纤维束通路的情况下，可以直接对纤维束进行线性配准至被试的个体空间，接下来将纤维束根据长度平均分成一百小段，每小段认作一个节点。所述纤维束节点影像指标提取模块用于通过扩散磁共振计算得到各个脑区的各向异性分数FA，平均弥散系数MD，细胞内体积分数ICVF和方向扩散分数ODI值等，对每条纤维束的每个节点计算上述指标在该小段纤维束上的平均值。所述机器学习分类与异常节点定位模块用于基于SVM分类器进行疾病组和健康组的分类，通过SVM得到特征的权重，可以反映哪些纤维束的哪些节点在疾病组和健康组之间有显著的差异。
本发明系统的具体实施过程如下：
所述扩散磁共振数据采集模块用于获取疾病组扩散磁共振数据以及相应的健康组扩散磁共振数据；本发明实例获取的是医院采集得到的临床数据，根据临床表现将数据分为0型病人，1型病人以及健康人。对数据进行整理和质量控制，最后入组的被试有0型病人121例，1型病人107例，健康人109例。
所述扩散磁共振数据预处理模块用于对扩散磁共振数据采集模块采集到的扩散磁共振数据进行降噪和矫正处理：具体为：扩散磁共振影像数据的预处理整体基于QSIPrep软件包完成，具体包括以下过程：dicom到BIDS格式转换，把数据整理成符合标准的格式。由于扩散磁共振图像存在信噪比较低的问题，所以用主成分分析的方法对图像进行去噪(MP-PCA)。由于B0场的不均匀性，扩散磁共振图像会在相位梯度方向产生畸变，所以要对图像进行畸变矫正(N4algorithm)。由于梯度磁场切换产生的涡流效应会产生阻碍梯度变化的涡流。这种额外的扰动将影响梯度场的变化使其波形严重畸变使图像存在几何形变、伪影等失真，这种效应被称为涡流效应，所以要对数据进行涡流矫正(eddy)。以及扫描过程中被试的呼吸和头动也会对数据的空间位置产生影响了影像，所以要对数据进行头动和涡流矫正(eddy)。
所述全脑纤维追踪模块用于基于预处理后的扩散磁共振数据提取全脑的纤维连接；具体为：使用MRtrix3软件对预处理后的扩散磁共振数据进行响应函数的估计以及通过球面约束 反卷积重建纤维方向分布模型，因为该数据包含了b＝0,1000,2000的b向量，所以本发明采用球面约束反卷积重建纤维方向分布扩散模型multi-shell-multi-tissue的方法进行模型重建；其次，基于重建的模型进行全脑纤维追踪，球面反卷积模型可以估计出每个体素的弥散分布，纤维追踪使用概率性纤维追踪法，在感兴趣区域内设定种子点，从每个种子点开始进行纤维束追踪，直到完成指定数目的纤维束。本发明设定是1千万条纤维束；最后，用纤维束的球面反卷积滤波方法对纤维追踪的结果做筛选，保留有生理学意义的纤维束。
所述定义感兴趣纤维束通路模块用于自定义纤维束通路或者基于大脑纤维束模板提取纤维束通路；具体为：在标准脑模板MNI152NLinin2009cAsym上定义起始感兴趣区域，终止感兴趣区域，通过感兴趣区域以及避开感兴趣区域，并对感兴趣区域进行修正，比如分割左右脑，腐蚀，脑外部分去除等操作。对选定的感兴趣区域进行基于种子点的纤维追踪，从起始感兴趣区域出发，通过或避开相应的感兴趣区域，到达终止感兴趣区域，最后获得满足感兴趣区域的纤维束通路。特别地，相应的纤维束通路也可以直接从公共的纤维束模板上提取。需要注意的是，左右半脑的纤维束最好分开统计。在本实例中，本发明共提取了80条不同的纤维束通路。80条通路中有2条没有从任何被试中提出，所以对剩下78条通路进行统计分析(大部分通路每组都有100多名被试，小部分通路每组只有不到10名被试有结果)。
所述纤维束通路投射与分段模块用于将纤维束通路投射到全脑的纤维连接结果上并进行分段，将每个分段定义为节点；具体为：基于扩散磁共振数据的特性，扩散磁共振指标的计算需要在被试的个体空间进行。所以本发明需要将标准脑模板MNI152NLinin2009cAsym上定义的纤维束感兴趣区域非线性配准到每个被试的结构像上，进而线性配准到每个被试的扩散像个体空间；然后在被试的个体空间进行基于种子点的纤维追踪，获得满足感兴趣区域的纤维束通路，特别地，在已经获得纤维束通路的情况下，可以直接对纤维束进行线性配准至被试的个体空间，从全脑的纤维追踪结果中提取出感兴趣的纤维束。接下来将通过pyAFQ工具包，通过自动定量方法，可对提取出来的感兴趣的纤维束进行进一步分段(如100等分)，每小段认作一个节点。
所述纤维束节点影像指标提取模块用于计算扩散磁共振数据的各向异性分数、平均扩散率、神经突内容积比以及方向分散度，从而得到每条纤维束通路每个节点上的影像学指标；具体为：对扩散磁共振数据进行弥散张量成像DTI模型拟合，计算全脑的各向异性分数FA和平均弥散系数MD值；并对扩散磁共振数据进行神经突定向弥散和密度成像模型NODDI模型拟合，计算全脑的细胞内体积分数ICVF和方向扩散分数ODI值；最后，对每条纤维束的每个节点计算上述指标在该节点的平均值。
所述机器学习分类与异常节点定位模块用于通过机器学习的方法在疾病组和健康组之间 用影像学指标做分类，定位不同疾病下哪些纤维束通路上的哪些节点发生了异常变化；具体为：统计健康组和疾病组共同追踪出的纤维束，以这些纤维束的100个节点值作为特征，构造特征集，以0为健康组标签，1为疾病组标签，构造标签集，整个数据集由特征集和相对应的标签集组成；其次，将数据集以被试为单位按8:2的比例随机分成训练集和测试集；然后，使用SVM分类器进行二分类预测，SVM分类器核函数为Linear核，以每条神经纤维节点特征作为SVM分类器的输入，以被试所在的组别为标签，作为SVM分类器的输出，其中训练集使用10折交叉验证，使用SVM模型的coef_参数，分别获取训练之后的10个模型的所有节点特征权重，根据权重对特征进行从大到小排序，取这个10个模型权重值占前10％的特征，并统计在前10％的特征中重复出现的节点特征，从而确定不同疾病下哪些纤维束通路上的哪些节点发生了异常变化；最后使用测试集进行模型测试，获取分类的准确率和AUC值，0型病人和健康人分类模型测试的AUC值如图3所示，本发明实施例中的AUC＝0.64。
在本发明实例中，针对该临床数据集进行纤维束提取分段，影像指标计算以及机器学习异常节点定位，定位到左右侧顶桥束的中段和后半段，皮质丘脑通路的前端和后端的FA值是在0型病人和健康人分类中前10％的特征中重复出现的节点特征，如下面表1所示。
表1
表1中表头为纤维通路的名称，数字为分类时权重大的节点位置。这些通路的异常与该疾病在临床上表现为行动不便，体脂率上升密切相关，说明该系统得到的结果具有较好的可解释性。并且该系统实现了纤维束通路异常节点的精准定位，得到了疾病组与健康组之间具有组间差异的纤维束段。
上述实施例用来解释说明本发明，而不是对本发明进行限制，在本发明的精神和权利要求的保护范围内，对本发明作出的任何修改和改变，都落入本发明的保护范围。

Claims (6)

  1. 一种大脑纤维束异常区域精准定位系统，其特征在于，该系统包括如下模块：
    扩散磁共振数据采集模块，用于获取疾病组扩散磁共振数据以及相应的健康组扩散磁共振数据；
    扩散磁共振数据预处理模块，用于对扩散磁共振数据采集模块采集到的扩散磁共振数据进行降噪和矫正处理；
    全脑纤维追踪模块，用于基于预处理后的扩散磁共振数据提取全脑的纤维连接；
    定义感兴趣纤维束通路模块，用于自定义纤维束通路或者基于大脑纤维束模板提取纤维束通路；具体为：在标准脑模板上定义起始感兴趣区域、终止感兴趣区域、通过感兴趣区域以及避开感兴趣区域，并进行基于种子点的纤维追踪，获得满足感兴趣区域的纤维束通路；
    纤维束通路投射与分段模块，用于将纤维束通路投射到全脑的纤维连接结果上，在已经获得纤维束通路的情况下，能够直接对纤维束进行线性配准至被试的个体空间，从全脑的纤维追踪结果中提取出感兴趣的纤维束，将纤维束根据长度平均分成若干小段，每小段定义为一个节点；纤维束节点影像指标提取模块，用于对扩散磁共振数据进行弥散张量成像DTI模型拟合，计算全脑的各向异性分数FA和平均弥散系数MD值，对扩散磁共振数据进行神经突定向弥散和密度成像模型NODDI模型拟合，计算全脑的细胞内体积分数ICVF和方向扩散分数ODI值；对每条纤维束的每个节点计算上述指标的平均值，从而得到每条纤维束通路每个节点上的影像学指标；
    机器学习分类与异常节点定位模块，用于通过机器学习的方法在疾病组和健康组之间用影像学指标做分类，以每条神经纤维节点特征作为分类器的输入，以被试所在的组别为标签作为分类器的输出，定位不同疾病下哪些纤维束通路上的哪些节点发生了异常变化。
  2. 根据权利要求1所述的一种大脑纤维束异常区域精准定位系统，其特征在于，所述扩散磁共振数据预处理模块用于对扩散磁共振数据进行基于PCA方法的去噪，并进行基于反相位编码图像的畸变矫正以及对扩散磁共振数据进行头动矫正和对扩散磁共振数据进行涡流矫正。
  3. 根据权利要求1所述的一种大脑纤维束异常区域精准定位系统，其特征在于，所述全脑纤维追踪模块用于对预处理后的扩散磁共振数据进行响应函数的估计以及通过球面约束反卷积重建纤维方向分布扩散模型，基于重建的模型进行全脑纤维追踪；用纤维束的球面反卷积滤波方法对纤维束进行筛选，仅保留有生理意义的纤维束。
  4. 根据权利要求1所述的一种大脑纤维束异常区域精准定位系统，其特征在于，所述定义感兴趣纤维束通路模块能够使用纤维束图谱上预先定义的纤维束通路。
  5. 根据权利要求1所述的一种大脑纤维束异常区域精准定位系统，其特征在于，所述纤维束通路投射与分段模块用于将标准脑模板上的定义的纤维束感兴趣区域非线性配准到每个被试的结构像上，进而线性配准到每个被试的扩散像个体空间；在被试的个体空间进行基于种子点的纤维追踪，获得满足感兴趣区域的纤维束通路。
  6. 根据权利要求1所述的一种大脑纤维束异常区域精准定位系统，其特征在于，所述机器学习分类与异常节点定位模块用于基于SVM分类器，以每条神经纤维节点特征作为SVM分类器的输入，以被试所在的组别为标签，作为SVM分类器的输出，SVM分类器训练集使用10折交叉验证，分别获取10个模型的特征权重，根据权重对特征进行从大到小排序，取前10％特征，并统计前10％特征中重复出现的节点特征，从而确定不同疾病下哪些纤维束通路上的哪些节点发生了异常变化。
PCT/CN2023/124640 2022-10-19 2023-10-16 一种大脑纤维束异常区域精准定位系统 WO2024083058A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211276171.8 2022-10-19
CN202211276171.8A CN115359305B (zh) 2022-10-19 2022-10-19 一种大脑纤维束异常区域精准定位系统

Publications (1)

Publication Number Publication Date
WO2024083058A1 true WO2024083058A1 (zh) 2024-04-25

Family

ID=84009003

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/124640 WO2024083058A1 (zh) 2022-10-19 2023-10-16 一种大脑纤维束异常区域精准定位系统

Country Status (2)

Country Link
CN (1) CN115359305B (zh)
WO (1) WO2024083058A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115359305B (zh) * 2022-10-19 2023-01-10 之江实验室 一种大脑纤维束异常区域精准定位系统
CN116542997B (zh) * 2023-07-04 2023-11-17 首都医科大学附属北京朝阳医院 磁共振图像的处理方法、装置以及计算机设备
CN117438054B (zh) * 2023-12-15 2024-03-26 之江实验室 一种脑影像数据的bids格式自动转换方法和装置

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080205733A1 (en) * 2006-10-19 2008-08-28 Brown Universisty Quantitative tract-of-interest metrics for white matter integrity based on diffusion tensor MRI data
US20170052241A1 (en) * 2015-08-17 2017-02-23 Siemens Healthcare Gmbh Tractography Framework With Magnetic Resonance Imaging For Brain Connectivity Analysis
CN110827282A (zh) * 2020-01-13 2020-02-21 南京慧脑云计算有限公司 一种基于磁共振成像的脑白质纤维束示踪分析方法及系统
CN114120024A (zh) * 2020-09-01 2022-03-01 盐城市第三人民医院 用于阿尔兹海默病分类预测的mr自动纤维定量分析方法
CN114842969A (zh) * 2022-03-23 2022-08-02 中国电子科技集团公司第十四研究所 一种基于关键纤维束的轻度认知障碍症评估方法
CN114983389A (zh) * 2022-06-15 2022-09-02 浙江大学 基于磁共振扩散张量成像的人脑轴突密度定量评估方法
CN115170540A (zh) * 2022-07-26 2022-10-11 浙江工业大学 一种基于多模态影像特征融合的轻度创伤性脑损伤分类方法
CN115359305A (zh) * 2022-10-19 2022-11-18 之江实验室 一种大脑纤维束异常区域精准定位系统

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7672790B2 (en) * 2006-11-02 2010-03-02 Siemens Medical Solutions Usa, Inc. System and method for stochastic DT-MRI connectivity mapping on the GPU
WO2011011554A1 (en) * 2009-07-21 2011-01-27 The Regents Of The University Of California Methods for the identification and targeting of brain regions and structures and treatments related thereto
CN103049901A (zh) * 2012-08-03 2013-04-17 上海理工大学 磁共振弥散张量成像纤维束追踪装置
TWI509534B (zh) * 2014-05-12 2015-11-21 Univ Nat Taiwan 自動化計算大腦纖維連結強度的方法
CN104899884A (zh) * 2015-06-03 2015-09-09 浙江工业大学 一种用于早期帕金森症预测的综合分析方法
CN106667490B (zh) * 2017-01-09 2017-12-29 北京师范大学 一种基于磁共振脑影像的被试对象间个体差异的数据间关系分析方法
EP3407295A1 (en) * 2017-05-22 2018-11-28 Koninklijke Philips N.V. Fibre-tracking from a diffusion-weighted magnetic resonance image
CN108734163B (zh) * 2018-05-04 2021-12-14 北京雅森科技发展有限公司 确定弥散张量成像感兴趣区的方法
CN110942489B (zh) * 2018-09-25 2023-04-25 西门子医疗系统有限公司 磁共振弥散张量成像方法、装置和纤维束追踪方法、装置
EP3734550A1 (en) * 2019-05-02 2020-11-04 Koninklijke Philips N.V. Removal of false positives from white matter fiber tracts
CN110811622A (zh) * 2019-11-12 2020-02-21 北京大学 一种基于扩散磁共振成像纤维束追踪技术的个体化结构连接脑图谱绘制方法
CN110992439B (zh) * 2019-12-02 2023-09-26 上海联影智能医疗科技有限公司 纤维束追踪方法、计算机设备和存储介质
CN113221952B (zh) * 2021-04-13 2023-09-15 山东师范大学 多中心大脑弥散张量成像图分类方法及系统
CN114187258A (zh) * 2021-12-09 2022-03-15 深圳先进技术研究院 基于人脑功能磁共振影像的自闭症分类器构建方法及系统
CN114494132A (zh) * 2021-12-24 2022-05-13 山东师范大学 基于深度学习和纤维束空间统计分析的疾病分类系统
CN114627283B (zh) * 2022-03-16 2023-04-28 西安市儿童医院 基于聚类去噪的感兴趣脑区纤维束提取系统及方法

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080205733A1 (en) * 2006-10-19 2008-08-28 Brown Universisty Quantitative tract-of-interest metrics for white matter integrity based on diffusion tensor MRI data
US20170052241A1 (en) * 2015-08-17 2017-02-23 Siemens Healthcare Gmbh Tractography Framework With Magnetic Resonance Imaging For Brain Connectivity Analysis
CN110827282A (zh) * 2020-01-13 2020-02-21 南京慧脑云计算有限公司 一种基于磁共振成像的脑白质纤维束示踪分析方法及系统
CN114120024A (zh) * 2020-09-01 2022-03-01 盐城市第三人民医院 用于阿尔兹海默病分类预测的mr自动纤维定量分析方法
CN114842969A (zh) * 2022-03-23 2022-08-02 中国电子科技集团公司第十四研究所 一种基于关键纤维束的轻度认知障碍症评估方法
CN114983389A (zh) * 2022-06-15 2022-09-02 浙江大学 基于磁共振扩散张量成像的人脑轴突密度定量评估方法
CN115170540A (zh) * 2022-07-26 2022-10-11 浙江工业大学 一种基于多模态影像特征融合的轻度创伤性脑损伤分类方法
CN115359305A (zh) * 2022-10-19 2022-11-18 之江实验室 一种大脑纤维束异常区域精准定位系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YUE QING: "A Fiber Tracking Algorithm Based on Non-Local Constrained Spherical Deconvolution", BOPUXUE ZAZHI = CHINESE JOURNAL OF MAGNETIC RESONANCE, ZHONGGUO KEXUEYUAN WUHAN WULI YANJIUSUO, WUHAN,, CN, vol. 37, no. 4, 10 December 2020 (2020-12-10), CN , pages 422 - 433, XP093159932, ISSN: 1000-4556, DOI: 10.11938/cjmr20192798 *

Also Published As

Publication number Publication date
CN115359305B (zh) 2023-01-10
CN115359305A (zh) 2022-11-18

Similar Documents

Publication Publication Date Title
WO2024083058A1 (zh) 一种大脑纤维束异常区域精准定位系统
Khan et al. Fetal brain growth portrayed by a spatiotemporal diffusion tensor MRI atlas computed from in utero images
Zhang et al. Mapping population-based structural connectomes
Rheault et al. Bundle-specific tractography with incorporated anatomical and orientational priors
WO2024083057A1 (zh) 基于多模态磁共振成像的图卷积神经网络疾病预测系统
O’Donnell et al. White matter tract clustering and correspondence in populations
Li et al. Wavelet-based segmentation of renal compartments in DCE-MRI of human kidney: initial results in patients and healthy volunteers
CN111047589A (zh) 一种注意力增强的脑肿瘤辅助智能检测识别方法
CN108898135B (zh) 一种大脑边缘系统图谱构建方法
Dhollander et al. Track orientation density imaging (TODI) and track orientation distribution (TOD) based tractography
Legarreta et al. Filtering in tractography using autoencoders (FINTA)
Li et al. Quantitative assessment of a framework for creating anatomical brain networks via global tractography
Zeng et al. FOD-Net: A deep learning method for fiber orientation distribution angular super resolution
CN115170540A (zh) 一种基于多模态影像特征融合的轻度创伤性脑损伤分类方法
Jha et al. VRfRNet: Volumetric ROI fODF reconstruction network for estimation of multi-tissue constrained spherical deconvolution with only single shell dMRI
Cover et al. Data-driven corpus callosum parcellation method through diffusion tensor imaging
Rathi et al. Biomarkers for identifying first-episode schizophrenia patients using diffusion weighted imaging
Zhang et al. A Bayesian approach to the creation of a study-customized neonatal brain atlas
Liu et al. Unsupervised automatic white matter fiber clustering using a Gaussian mixture model
CN114596306A (zh) 基于机器学习诊断帕金森疾病的系统
Legarreta et al. Tractography filtering using autoencoders.
CN114494132A (zh) 基于深度学习和纤维束空间统计分析的疾病分类系统
Liang et al. Shape modeling and clustering of white matter fiber tracts using fourier descriptors
Robinson et al. Multivariate statistical analysis of whole brain structural networks obtained using probabilistic tractography
CN112837807A (zh) 一种t2dm脑衰老认知障碍早期智能高精度辅诊方法