| CPC A61B 5/055 (2013.01) [A61B 5/0042 (2013.01); A61B 5/7267 (2013.01); G06T 7/0012 (2013.01); G06T 2207/10088 (2013.01); G06T 2207/30016 (2013.01)] | 5 Claims |
|
1. A system for precisely locating abnormal areas of brain fiber bundles, comprising:
a diffusion magnetic resonance data acquisition module configured to acquire diffusion magnetic resonance data from a disease group and a healthy group;
a diffusion magnetic resonance data preprocessing module configured to denoise and correct the diffusion magnetic resonance data acquired by the diffusion magnetic resonance data acquisition module;
a whole brain fiber tracking module configured to extract fiber connections of a whole brain based on the preprocessed diffusion magnetic resonance data;
a fiber bundle pathway of interest defining module configured to self-define a fiber bundle pathway or extract a fiber bundle pathway based on a brain fiber bundle template, further comprising: defining initiate regions of interest, ending the regions of interest, passing the regions of interest and avoiding the regions of interest on a standard brain template, and tracking fibers based on seed points to obtain fiber bundle pathways satisfying the regions of interest;
a fiber bundle pathway projection and segmentation module configured to project the fiber bundle pathway onto a fiber connection result of the whole brain, directly linearly register a fiber bundle to an individual space of a subject when the fiber bundle pathway has been obtained, extract a fiber bundle of interest from a fiber tracking result of the whole brain, and segment the fiber bundle into a plurality of small segments on average according to a length, wherein each small segment is defined as a node;
a fiber bundle node image index extraction module configured to perform diffusion tensor imaging (DTI) model fitting on the diffusion magnetic resonance data, calculate anisotropy fraction (FA) and mean diffusivity (MD) values of the whole brain, perform neurite orientation dispersion and density imaging (NODDI) fitting on the diffusion magnetic resonance data, and calculate intracellular volume fraction (ICVF) and orientation dispersion index (ODI) values of the whole brain; and calculate an average value of above indexes for each node of each fiber bundle to obtain imaging indexes on each node of each fiber bundle pathway; and
a machine learning classification and abnormal node locating module configured to classify the disease group and the healthy group by using the imaging indexes through a machine learning method with features of each nerve fiber node as an input of a classifier and a group label of the subject as an output of the classifier, and locate which nodes on which fiber bundle pathways have abnormal changes with different diseases;
wherein the machine learning classification and abnormal node locating module is further configured to obtain feature weights from 10 models generated using a support vector machine (SVM) classifier, respectively, based on the SVM classifier with the features of each nerve fiber node as the input and the group label of the subject belongs as the output, wherein a training set of the SVM classifier uses a 10-fold cross validation, and sort the features from large to small depending on the feature weights, select the top 10% features, and count node features that appear repeatedly in the top 10% features to determine which nodes on which fiber bundles pathways have the abnormal changes with the different diseases.
|