Structural magnetic resonance imaging analysis method, system and storage medium

Abstract

The present disclosure provides a structural magnetic resonance imaging analysis method, system and storage medium, including: obtaining sample data, wherein the sample data is a T1-weighted structural image obtained by structural magnetic resonance imaging, determining a map of morphological indicators of a region of interest in the original brain network space according to the sample data, resampling the map to a preset standard brain network space template, and performing Gaussian kernel smoothing to obtain standard brain network space template data, mapping the standard brain network space template data to a pre-constructed brain atlas template, determining the feature input of a machine learning model according to the mapped brain atlas template, and training the classification performance of the machine learning model based on the feature input to obtain a classification model. The effectiveness and reliability of the training can be improved, thereby improving the accuracy of classification based on the classification model obtained by training.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to an analysis method, system and storage medium for structural magnetic resonance imaging. Background Art

[0002] Structural MRI (sMRI) is a non-invasive neuroimaging technique used to perform three-dimensional imaging of the structure of the human brain. It can provide high-resolution brain anatomical information, showing the types of different structures and tissues in the human brain, such as gray matter, white matter, cerebrospinal fluid, and the spatial relationship between brain structures.

[0003] By classifying and analyzing the images obtained by the structural magnetic resonance imaging technology, the type of image can be determined, for example, whether the image corresponds to a disease symptom or a healthy state.

[0004] Classification analysis can adopt the traditional machine learning method of support vector machine (SVM), but how to improve the effectiveness and reliability of classification analysis has become an urgent problem to be solved.

[0005] The content of the background technology section is only the information known to the inventor personally, and does not mean that the above information has entered the public domain before the application date of this disclosure, nor does it mean that it can become the prior art of the present disclosure. Summary of the invention

[0006] The present disclosure provides a structural magnetic resonance imaging analysis method, system and storage medium to improve the effectiveness and reliability of classification analysis.

[0007] In a first aspect, the present disclosure provides a method for analyzing structural magnetic resonance imaging, comprising:

[0008] Obtaining sample data, wherein the sample data is a T1-weighted structural image obtained by structural magnetic resonance imaging;

[0009] Determine a map of morphological indicators of the region of interest in the original brain network space according to the sample data, wherein the morphological indicators are used to characterize the anatomical and structural properties of the cerebral cortex, and the morphological indicators include cortical thickness, sulcus depth, cortical folds and fractal dimension;

[0010] Resampling the map to a preset standard brain network space template and performing Gaussian kernel smoothing to obtain standard brain network space template data;

[0011] Mapping the standard brain network space template data to a pre-constructed brain atlas template, wherein the brain atlas template divides the cerebral cortex surface into 148 regions of interest, and the 148 regions of interest are divided into 7 major brain networks, the 7 major brain networks including: visual network, somatomotor network, dorsal attention network, ventral attention network, limbic network, frontoparietal network, and default mode network;

[0012] Determine the feature input of the machine learning model based on the mapped brain atlas template;

[0013] The classification performance of the machine learning model is trained based on the feature input to obtain a classification model, wherein the classification performance is the ability to distinguish the type corresponding to the T1 weighted structural image, and the classification model is used to classify the T1 weighted structural image to be classified to obtain a classification result.

[0014] In some embodiments, the sample data is preprocessed data, and the preprocessing includes at least one of image format conversion, image quality inspection, tissue segmentation, cortical reconstruction and smoothing, segmentation quality inspection and cortical reconstruction inspection.

[0015] In some embodiments, determining the feature input of the machine learning model according to the mapped brain atlas template includes:

[0016] For each of the 148 regions of interest, extract the morphological index of each vertex in each region of interest, and determine the probability density of each region of interest;

[0017] The feature input is determined according to each probability density.

[0018] In some embodiments, determining the feature input according to each probability density includes:

[0019] According to each probability density similarity measure, wherein the similarity measure includes Kullback-Leibler divergence KLD and Jensen-Shannon divergence JSD, and the KLD and the JSD are used to characterize the morphological similarity between brain regions;

[0020] The feature input is determined based on the KLD and the JSD.

[0021] In some embodiments, determining the feature input according to the KLD and the JSD includes:

[0022] Determine a similarity matrix based on the KLD and the JSD;

[0023] The upper triangular data of the similarity matrix is ​​extracted, and the extracted upper triangular data is constructed to obtain the feature input of the machine learning model.

[0024] In some embodiments, the machine learning model is an unsupervised Kmeans clustering model, or a hypergraph learning-based model;

[0025] In the case where the machine learning model is the model based on hypergraph learning, the hypergraph includes multiple nodes and multiple hyperedges, the nodes are used to represent elements or entities, and the hyperedges are used to represent the relationships between the elements or entities.

[0026] In some embodiments, the training of the classification performance of the machine learning model based on the feature input to obtain the classification model includes:

[0027] Based on a preset feature selection method, selecting a target feature from the feature input, wherein the preset feature selection method includes feature selection based on statistics and / or feature selection based on F-score, the feature selection based on statistics is used to determine features whose differences are greater than a preset difference threshold, and the feature selection based on F-score is used to determine features whose importance is greater than a preset importance threshold;

[0028] The classification performance of the machine learning model is trained based on the target features to obtain the classification model.

[0029] In some embodiments, the method further comprises:

[0030] Obtaining consensus features during the training process, wherein the consensus features are used to characterize features whose contribution to the classification performance reaches a preset contribution requirement;

[0031] Mapping the consensus feature back to the original brain network space to locate the consensus feature in the original brain network space;

[0032] The brain network importance index analysis is performed on each brain network sub-region of the original brain network space mapped with the consensus features to obtain the brain network region that contributes most significantly to the quantitative classification performance.

[0033] In a second aspect, the present disclosure provides a structural magnetic resonance imaging analysis system, comprising:

[0034] An acquisition unit, configured to acquire sample data, wherein the sample data is a T1-weighted structural image obtained by structural magnetic resonance imaging;

[0035] A determination unit, used to determine a map of morphological indicators of the region of interest in the original brain network space according to the sample data, wherein the morphological indicators are used to characterize the anatomical and structural properties of the cerebral cortex, and the morphological indicators include cortical thickness, sulcus depth, cortical folds and fractal dimension;

[0036] A processing unit, used for resampling the map to a preset standard brain network space template, and performing Gaussian kernel smoothing to obtain standard brain network space template data;

[0037] A mapping unit, used for mapping the standard brain network space template data to a pre-constructed brain map template, wherein the brain map template divides the cerebral cortex surface into 148 regions of interest, and the 148 regions of interest are divided into 7 major brain networks, and the 7 major brain networks include: a visual network, a somatic motor network, a dorsal attention network, a ventral attention network, a limbic network, a frontoparietal network, and a default mode network;

[0038] The determining unit is further used to determine the feature input of the machine learning model according to the mapped brain atlas template;

[0039] A training unit is used to train the classification performance of the machine learning model based on the feature input to obtain a classification model, wherein the classification performance is the ability to distinguish the type corresponding to the T1 weighted structural image, and the classification model is used to classify the T1 weighted structural image to be classified to obtain a classification result.

[0040] In some embodiments, the sample data is preprocessed data, and the preprocessing includes at least one of image format conversion, image quality inspection, tissue segmentation, cortical reconstruction and smoothing, segmentation quality inspection and cortical reconstruction inspection.

[0041] In some embodiments, the determination unit is specifically used to extract the morphological indicators of each vertex in each of the 148 regions of interest, and determine the probability density of each region of interest; and determine the feature input according to each probability density.

[0042] In some embodiments, the determination unit is specifically used to determine the feature input based on each probability density similarity measure, wherein the similarity measure includes Kullback-Leibler divergence KLD and Jensen-Shannon divergence JSD, and the KLD and the JSD are used to characterize the morphological similarity between brain regions.

[0043] In some embodiments, the determination unit is specifically used to determine a similarity matrix based on the KLD and the JSD; and extract upper triangular data from the similarity matrix, and construct the extracted upper triangular data to obtain the feature input of the machine learning model.

[0044] In some embodiments, the machine learning model is an unsupervised Kmeans clustering model, or a hypergraph learning-based model;

[0045] In the case where the machine learning model is the model based on hypergraph learning, the hypergraph includes multiple nodes and multiple hyperedges, the nodes are used to represent elements or entities, and the hyperedges are used to represent the relationships between the elements or entities.

[0046] In some embodiments, the training unit is specifically used to select target features from the feature input based on a preset feature selection method, wherein the preset feature selection method includes statistical-based feature selection and / or F-score-based feature selection, the statistical-based feature selection is used to determine features whose differences are greater than a preset difference threshold, and the F-score-based feature selection is used to determine features whose importance is greater than a preset importance threshold; and the classification performance of the machine learning model is trained based on the target features to obtain the classification model.

[0047] In some embodiments, the acquisition unit is further used to obtain consensus features in the training process, wherein the consensus features are used to characterize features whose contribution to the classification performance reaches a preset contribution requirement;

[0048] The mapping unit is further used to map the consensus feature back to the original brain network space to locate the consensus feature in the original brain network space;

[0049] The determination unit is also used to perform brain network importance index analysis on each brain network sub-region of the original brain network space mapped with the consensus feature, so as to obtain the brain network region that contributes most significantly to the quantitative classification performance.

[0050] In a third aspect, the present disclosure provides a processor-readable storage medium, wherein the processor-readable storage medium stores a computer program, wherein the computer program is used to enable the processor to execute the method described in the first aspect above.

[0051] In a fourth aspect, the present disclosure provides a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0052] The present disclosure provides a structural magnetic resonance imaging analysis method, system and storage medium, comprising: obtaining sample data, wherein the sample data is a T1-weighted structural image obtained by structural magnetic resonance imaging, determining a map of morphological indicators of a region of interest in an original brain network space according to the sample data, wherein the morphological indicators are used to characterize the anatomical and structural properties of the cerebral cortex, and the morphological indicators include cortical thickness, sulcus depth, cortical folds and fractal dimension, resampling the map to a preset standard brain network space template, and performing Gaussian kernel smoothing to obtain standard brain network space template data, mapping the standard brain network space template data to a pre-constructed brain atlas template, wherein, The brain map template divides the surface of the cerebral cortex into 148 regions of interest, and the 148 regions of interest are divided into 7 major brain networks, including: visual network, somatic motor network, dorsal attention network, ventral attention network, limbic network, frontoparietal network, and default mode network. The feature input of the machine learning model is determined according to the mapped brain map template, and the classification performance of the machine learning model is trained based on the feature input to obtain a classification model, wherein the classification performance is the ability to distinguish the type corresponding to the T1 weighted structural image, and the classification model is used to classify the T1 weighted structural image to be classified to obtain a classification result. The effectiveness and reliability of the training can be improved, thereby improving the accuracy of classification based on the classification model obtained by training. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0054] Figure 1 A schematic diagram of a flow chart of a method for analyzing structural magnetic resonance imaging provided by an embodiment of the present disclosure;

[0055] Figure 2 A schematic diagram of the principle of the structural magnetic resonance imaging analysis method provided by the embodiment of the present disclosure;

[0056] Figure 3 A schematic diagram of the training principle of the classification model provided in the embodiment of the present disclosure;

[0057] Figure 4 A schematic diagram of a confusion matrix corresponding to the classification results provided in an embodiment of the present disclosure;

[0058] Figure 5 Brain network mapping diagrams for four morphological indicators;

[0059] Figure 6 A schematic diagram of the structure of the analysis system for structural magnetic resonance imaging provided by an embodiment of the present disclosure;

[0060] Figure 7A schematic block diagram of an electronic device provided in accordance with an embodiment of the present disclosure.

[0061] The above drawings show clear embodiments of the present disclosure, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present disclosure in any way, but to illustrate the concepts of the present disclosure to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0062] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0063] It should be understood that the terms "including" and "having" and any variations thereof in the embodiments of the present disclosure are intended to cover but not exclude inclusion. For example, a product or device comprising a series of components is not necessarily limited to those components explicitly listed, but may include other components not explicitly listed or inherent to these products or devices.

[0064] In the embodiments of the present disclosure, the term "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0065] The term "plurality" in the embodiments of the present disclosure refers to two or more than two, and other quantifiers are similar thereto.

[0066] The terms "first", "second", "third", etc. in the present disclosure are used to distinguish similar or similar objects or entities, and are not necessarily meant to limit a specific order or sequence, unless otherwise indicated. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, for example, they can be implemented in an order other than those given in the diagrams or descriptions of the embodiments of the present disclosure.

[0067] The term "unit / module" as used in this disclosure refers to any known or later developed hardware, software, firmware, artificial intelligence, fuzzy logic or a combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0068] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0069] According to one aspect of the present disclosure, the present disclosure provides an analysis method for structural magnetic resonance imaging (hereinafter referred to as the analysis method).

[0070] See also Figure 1 , Figure 1 Schematic diagram of the analysis method provided in the embodiment of the present disclosure. Figure 1 As shown, the analysis method includes:

[0071] S101: Obtain sample data, wherein the sample data is a T1-weighted structural image obtained by structural magnetic resonance imaging (sMRI).

[0072] The analysis method provided in this embodiment may be performed by an analysis system for structural magnetic resonance imaging (hereinafter referred to as the analysis system). The analysis system may be a server, an electronic device, a processor, etc. The analysis system is described elsewhere in this specification.

[0073] The data analysis of sMRI can be performed based on the surface-based morphometry (SBM) method. sMRI processing and analysis can be performed based on MATLAB's SPM12 software and CAT12 toolbox to obtain T1-weighted structural images, so that the T1-weighted structural images can be used as sample data to train a classification model.

[0074] In some embodiments, the sample data is preprocessed data, and the preprocessing includes at least one of image format conversion, image quality inspection, tissue segmentation, cortical reconstruction and smoothing, segmentation quality inspection and cortical reconstruction inspection.

[0075] Exemplarily, after obtaining the T1-weighted structural image, the analysis system may preprocess the T1-weighted structural image to use the preprocessed image as sample data.

[0076] The analysis system may use one or more of the above-mentioned preprocessing methods to preprocess the T1-weighted structural image. The reliability, accuracy and effectiveness of the sample data may be improved through preprocessing.

[0077] S102: Determine a map of morphological indicators of a region of interest (ROI) in the original brain network space based on the sample data, wherein the morphological indicators are used to characterize the anatomical and structural properties of the cerebral cortex, and the morphological indicators include cortical thickness (CT), sulcal depth (SD), gyrification index (GI) and fractal dimension (FD).

[0078] Among them, the original brain network space can be understood as the brain network space of the sample object.

[0079] For example, this step can be understood as the analysis system can calculate the morphological index of the region of interest in the original space of the sample object (ie, the original brain network space). It should be noted that the above preprocessing process is also performed in the original brain network space.

[0080] like Figure 2 As shown, the analysis system can obtain Figure 2 The data set shown in , the data set includes sample data, and the analysis system can determine the morphological indicators of the region of interest in the original brain network space based on S102, and the morphological indicators include (FD / GI / SD / CT).

[0081] S103: resampling the map to a preset standard brain network space template, and performing Gaussian kernel smoothing to obtain standard brain network space template data.

[0082] The standard brain network space template can be understood as the brain network space obtained after unified processing of each original brain network space.

[0083] This step can be understood as that after the analysis system obtains the map of morphological indicators of the region of interest in the original brain network space, it can resample the map into a unified standard brain network space template and perform Gaussian kernel smoothing to improve stability and reliability.

[0084] For example, the analysis system can select different Gaussian kernel sizes that match the distribution characteristics of the morphological index for smoothing. For example, the CT map uses a Gaussian kernel with a full width of 15 mm, while the SD, GI, and FD maps use a Gaussian kernel of 25 mm.

[0085] S104: Mapping the standard brain network space template data to a pre-built brain map template, wherein the brain map template divides the cerebral cortex surface into 148 regions of interest, and the 148 regions of interest are divided into 7 major brain networks, including: Visual Network (VN), Somatomotor Network (SMN), Dorsal Attention Network (DAN), Ventral Attention Network (VAN), Limbic Network (LN), Frontoparietal Network (FPN) and Default Mode Network (DMN).

[0086] Exemplarily, the analysis system can use the Destrieux brain atlas for data analysis. It provides detailed and comprehensive cerebral cortical partitioning and related anatomical information. The application of this atlas deepens our understanding of the complexity of brain structure and improves the accuracy of brain connection pattern analysis. This atlas template carefully divides the surface of the cerebral cortex into 148 regions of interest (ROIs). This partitioning method enables us to quantify the four brain morphological indicators of CT, SD, GI, and FD in each area.

[0087] Specifically, for each of the 148 ROIs, the analysis system can extract the average value of these indicators of all vertices in the ROI to represent the morphological characteristics of the region.

[0088] Based on the whole-brain cortical structural positioning of the Destrieux atlas, the analysis system can further divide the 148 ROIs into 7 major brain networks. This division helps to understand the relationship between brain function and structure at a more macro level.

[0089] For example, Figure 2 As shown, after the morphological indicators are determined, they can be mapped to the brain atlas (i.e., the brain atlas template).

[0090] S105: Determine feature input of the machine learning model according to the mapped brain map template.

[0091] In some embodiments, S105 may include:

[0092] S51: For each of the 148 regions of interest, extract the morphological index of each vertex in each region of interest, and determine the probability density of each region of interest.

[0093] Exemplarily, the analysis system can initially extract the values ​​of the four morphological indices (CT, SD, GI, FD) of all vertices from each of the 148 ROIs. Next, the analysis system can use the ksdensity function of MATLAB to estimate the probability density of the morphological indices for each of the 148 ROIs. This step involves using a normal kernel function to smooth the distribution of each indicator, thereby obtaining continuous probability density functions (PDFs).

[0094] S52: Determine feature input according to each probability density.

[0095] In some embodiments, S52 may include:

[0096] S521: Measuring similarity based on each probability density, wherein the similarity metrics include Kullback-Leibler divergence KLD and Jensen-Shannon divergence JSD, and KLD and JSD are used to characterize the morphological similarity between brain regions.

[0097] Exemplarily, in this embodiment, the analysis system can use two statistical measurement methods, KLD and JSD, to evaluate the morphological similarity between brain regions. Among them, KLD is an asymmetric measurement method used to calculate the difference between two probability distributions, revealing the degree of deviation of one distribution relative to the other. JSD is a symmetrical measurement method derived from KLD, which is used to estimate the similarity of two probability distributions and has a fixed upper bound, thereby providing an intuitive measure for comparing the similarity of two distributions.

[0098] Combined with the above analysis, after obtaining the PDFs, the analysis system can calculate KLDs and JSDs based on pairwise PDFs.

[0099] For example, KLDs can be calculated based on Formula 1:

[0100]

[0101] It can be calculated by formula 2:

[0102]

[0103] in, is a smoothed version of KLD, representing the probability distribution of two adjacent brain regions and similarities between; KLD is used to measure the probability distribution and The difference between for arrive KLD; for arrive KLD; is the number of vertices; for In the The probability of each vertex; for In the The probability of a vertex.

[0104] JSDs can be calculated by formula 3:

[0105]

[0106] It can be calculated by formula 4:

[0107]

[0108] in, is a smoothed version of JSD, used to represent and similarities between; Used to measure and The difference between JSD is used to measure and The difference between.

[0109] The value range of KLD and JSD is [0,1], where 0 and 1 respectively indicate that the two probability distribution functions are completely different and completely the same.

[0110] S522: Determine feature input according to KLD and JSD.

[0111] In some embodiments, S522 may include:

[0112] S5221: Determine a similarity matrix based on KLD and JSD.

[0113] Exemplary, combined Figure 2 It can be seen that after determining the similarity measure, the analysis system can determine a similarity matrix based on the similarity measure.

[0114] S5222: Extract the upper triangular data of the similarity matrix, and construct the extracted upper triangular data to obtain the feature input of the machine learning model.

[0115] Continue reading Figure 2,Based on the similarity matrix, the analysis system can extract the ,upper triangular data of the matrix (i.e., extract the upper triangular data of the ,similarity matrix) to use the upper triangular data as feature input.

[0116] S106: Training the classification performance of the machine learning model based on the feature input to obtain a classification model, wherein the classification performance is the ability to distinguish the type corresponding to the T1 weighted structural image, and the classification model is used to classify the T1 weighted structural image to be classified to obtain a classification result.

[0117] Continue reading Figure 2 , the machine learning module can perform classification based on feature input, and perform iterative learning based on the obtained classification results until a classification model with classification performance is obtained.

[0118] For different application scenarios, the classification results may correspond to different contents. For example, for the scenario of non-suicidal self-injury in adolescents with depression, the classification result of the T1-weighted structural image to be classified determined based on the classification model may be an image corresponding to the intention of self-injury behavior, or it may be an image corresponding to no intention of self-injury behavior. More specifically, the intention of self-injury behavior may be the intention of suicidal self-injury (NSSI) behavior, such as NSSI in adolescents with depression. This makes it possible to use neuroimaging methods to find objective neural markers to identify and predict the intention of self-injury behavior as early as possible.

[0119] In some embodiments, the machine learning model is an unsupervised Kmeans clustering model, or a model based on hypergraph learning (Hypergraph); when the machine learning model is a model based on hypergraph learning, the hypergraph includes multiple nodes and multiple hyperedges, the nodes are used to represent elements or entities, and the hyperedges are used to represent the relationship between elements or entities.

[0120] Exemplarily, during the model building process, the analysis system can obtain feature inputs for building a machine learning model based on upper triangular data extraction in a similarity measure matrix.

[0121] Here, the analysis system mainly focuses on two algorithms: unsupervised Kmeans clustering and hypergraph learning. For example, combined with the above example, the analysis system can classify based on unsupervised Kmeans clustering or based on hypergraph learning.

[0122] Furthermore, the analysis systems may adopt different validation methods to evaluate their performance. For example, the validation method may be a five-fold cross validation.

[0123] Among them, hypergraph learning is a more powerful data structure that expands on the traditional graph structure and is used to represent complex relational networks. In hypergraph learning, nodes represent elements or entities in the network, while hyperedges represent the relationships between these elements. Unlike traditional graphs, a hyperedge can connect any number of nodes, which enables hypergraph learning to better capture the complex associations between multiple elements. The advantage of hypergraph learning is that it can more accurately describe the structure and relationships of complex networks. Compared with traditional graph structures, it can better represent high-order associations and provide richer analysis and reasoning capabilities.

[0124] In some embodiments, S106 may include:

[0125] S61: Based on a preset feature selection method, select target features from feature inputs, wherein the preset feature selection method includes feature selection based on statistics and / or feature selection based on F-score, the feature selection based on statistics is used to determine features whose differences are greater than a preset difference threshold, and the feature selection based on F-score is used to determine features whose importance is greater than a preset importance threshold.

[0126] Exemplarily, the analysis system performs a two-sample T test on two types of sample data (positive sample data and negative sample data) based on statistical feature selection, checks the distribution of features in various brain regions, and selects features with significant differences.

[0127] The analysis system is based on F-score feature selection. In order to sort the feature importance according to the F-score, the features selected in each training are selected, their distribution in the brain areas is checked, and the feature distribution is compared with the two-sample T test to find the features with the best classification effect.

[0128] S62: Train the classification performance of the machine learning model based on the target features to obtain a classification model.

[0129] For example, in combination with the above-mentioned classification scenarios of the intention of self-harm behavior, Figure 2 as well as Figure 3 It can be seen that the feature input includes: NSSI positive group (NSSI+) and NSSI negative group (NSSI-), each group contains a different number of male and female samples. Through random selection (Random Select), a certain number of samples are randomly selected from the NSSI+ and NSSI- groups for training.

[0130] The data includes: inner training data, inner test data, outer training data, and outer test data. The inner test data is used for training and testing of the inner loop (inner loop). The outer test data is used for training and testing of the outer loop (outer loop).

[0131] In the inner training process, feature selection and classification performance training of the hypergraph model are included. Among them, the analysis system can filter out features that contribute significantly to the classification task from the high-dimensional feature space through the above-mentioned feature selection method (such as Figure 3 Consensus features in the dataset). For example, this step can use statistical methods (such as F-score) to evaluate the importance of each feature. Accordingly, the hypergraph model models the selected target features into the hypergraph model. The hypergraph model is an advanced graphical model that can capture complex relationships between features.

[0132] In the outer training process, the optimal parameters obtained through the internal training process are included, and the processing of the hypergraph model is performed to determine the significant contribution to the classification decision.

[0133] Among them, the optimal parameters can be achieved through parameter optimization. For example, grid search for parameter optimization: through the grid search method, the optimal parameters are found in multiple parameter combinations. Grid search is a systematic method that finds the best parameter settings by traversing all possible parameter combinations.

[0134] Combination Figure 3 It can be seen that the entire training process is divided into two loops, namely the inner loop and the outer loop. The inner loop is used for feature selection and model training, and the outer loop is used to evaluate model performance and parameter optimization.

[0135] Exemplarily, in the inner loop, the model is trained using internal training data, and the model is evaluated using internal test data. In the outer loop, the model is trained using external training data, and the final model is evaluated using external test data.

[0136] In some embodiments, the analysis method further comprises:

[0137] Step 1: Obtain consensus features during the training process, where the consensus features are used to characterize features whose contribution to the classification performance reaches a preset contribution requirement.

[0138] For example, in the modeling analysis of neuroimaging, the identification and location of brain features are crucial. In order to accurately find the brain network areas that have a significant impact on classification, the analysis system uses a method to calculate the importance of brain networks based on important features. This method is mainly divided into two key steps.

[0139] First, we use the ten-fold cross validation method to process a dataset with tens of thousands of features. In this process, feature traversal is a necessary part, the purpose is to screen out the features that are always selected in each cross validation fold, which we define as consensus features. In order to determine these consensus features, we use the F-score method to weight each feature, which determines its importance by analyzing the statistical correlation between the feature and the classification variable. The set including the consensus feature is considered to be a feature subset that contributes significantly to the classification decision.

[0140] Step 2: Map the consensus features back to the original brain network space to locate the consensus features in the original brain network space.

[0141] Step 3: Perform brain network importance index analysis on each brain network sub-region of the original brain network space mapped with consensus features to obtain the brain network region that contributes most significantly to the quantitative classification performance.

[0142] Combined with the above analysis, the analysis system then maps the consensus features back to the original brain network space. This step involves the conversion from the feature selection space to the brain anatomical structure template. Through this mapping, the analysis system can locate these features in the brain network and further analyze the cumulative weights of the selected features in each brain region. These weights represent the relative importance of each brain network sub-region in the classification task, thus forming a quantitative brain network importance indicator. Through the above two key steps, the analysis system can identify and quantify the brain network regions that contribute most significantly to the classification.

[0143] For the classification scenario of intention of self-harm behavior, in the analysis method provided by the present disclosure, a comparative experiment was conducted based on Kmeans and Hypergraph, SVM was used as the baseline, and the combination of the JSD method and the Destrieux brain map obtained a better classification effect, and the training results are shown in Table 1. Table 1 is used to represent the classification accuracy (%) of different classifiers (Kmeans as the baseline) constructed based on the sMRI data of NSSI patients.

[0144] Table 1

[0145]

[0146] As shown in Table 1, the classification accuracy based on the four morphological indicators varies under different selections of the number of features, and the highest accuracy (%) reaches 0.6750, 0.6649, 0.6643, and 0.6672, respectively.

[0147] According to the classification results, the corresponding confusion matrix is ​​obtained as follows Figure 4 As shown. Figure 4 Among them, four morphological indicators (browsed Figure 4 The perspective is used as the benchmark, from left to right are CT, FD, GI, SD) based on the confusion matrix of the classification results of the hypergraph learning model. This shows that the hypergraph learning model can be compared with the general machine learning method and can achieve a certain degree of performance improvement. It further verifies the feasibility of using the hypergraph learning model in magnetic resonance imaging compared with the machine learning model.

[0148] There were differences in CT, FD, GI, and SD morphological indicators in the whole brain between patients with NSSI and those without NSSI. Figure 5 As shown, there are differences in brain structural connection characteristics between the two, among which the main network with differences is the limbic network (LN) and its related networks. Figure 5 Four morphological indicators (browsed Figure 5 Based on the perspective, from left to right are brain network maps of CT, FD, GI, and SD).

[0149] In the technical solution disclosed herein, the collection, storage, use, processing, transmission, provision and disclosure of user personal information (such as T1 weighted structural images, etc.) involved are in compliance with the relevant laws and regulations and do not violate public order and good morals.

[0150] According to an embodiment of the present disclosure, the present disclosure also provides an analysis system for structural magnetic resonance imaging.

[0151] See also Figure 6 , Figure 6 The structural schematic diagram of the analysis system of structural magnetic resonance imaging provided by the embodiment of the present disclosure. Figure 6 As shown, the analysis system includes:

[0152] An acquisition unit 601 is used to obtain sample data, wherein the sample data is a T1-weighted structural image obtained by structural magnetic resonance imaging;

[0153] A determination unit 602 is used to determine a map of morphological indicators of the region of interest in the original brain network space according to the sample data, wherein the morphological indicators are used to characterize the anatomical and structural properties of the cerebral cortex, and the morphological indicators include cortical thickness, sulcus depth, cortical folds and fractal dimension;

[0154] The processing unit 603 is used to resample the map to a preset standard brain network space template and perform Gaussian kernel smoothing to obtain standard brain network space template data;

[0155] A mapping unit 604 is used to map the standard brain network space template data to a pre-constructed brain map template, wherein the brain map template divides the surface of the cerebral cortex into 148 regions of interest, and the 148 regions of interest are divided into 7 major brain networks, and the 7 major brain networks include: a visual network, a somatic motor network, a dorsal attention network, a ventral attention network, a limbic network, a frontoparietal network, and a default mode network;

[0156] The determining unit 602 is also used to determine the feature input of the machine learning model according to the mapped brain map template;

[0157] The training unit 605 is used to train the classification performance of the machine learning model based on the feature input to obtain a classification model, wherein the classification performance is the ability to distinguish the type corresponding to the T1 weighted structural image, and the classification model is used to classify the T1 weighted structural image to be classified to obtain a classification result.

[0158] In some embodiments, the sample data is preprocessed data, and the preprocessing includes at least one of image format conversion, image quality inspection, tissue segmentation, cortical reconstruction and smoothing, segmentation quality inspection and cortical reconstruction inspection.

[0159] In some embodiments, the determination unit 602 is specifically used to extract the morphological indicators of each vertex in each of the 148 regions of interest, and determine the probability density of each region of interest; and determine the feature input according to each probability density.

[0160] In some embodiments, the determination unit 602 is specifically used to determine the feature input based on each probability density similarity measure, wherein the similarity measure includes Kullback-Leibler divergence KLD and Jensen-Shannon divergence JSD, and the KLD and the JSD are used to characterize the morphological similarity between brain regions.

[0161] In some embodiments, the determination unit 602 is specifically used to determine a similarity matrix based on the KLD and the JSD; and to extract upper triangular data from the similarity matrix, and to construct the extracted upper triangular data to obtain the feature input of the machine learning model.

[0162] In some embodiments, the machine learning model is an unsupervised Kmeans clustering model, or a hypergraph learning-based model;

[0163] In the case where the machine learning model is the model based on hypergraph learning, the hypergraph includes multiple nodes and multiple hyperedges, the nodes are used to represent elements or entities, and the hyperedges are used to represent the relationships between the elements or entities.

[0164] In some embodiments, the training unit 605 is specifically used to select target features from the feature input based on a preset feature selection method, wherein the preset feature selection method includes statistical-based feature selection and / or F-score-based feature selection, the statistical-based feature selection is used to determine features whose differences are greater than a preset difference threshold, and the F-score-based feature selection is used to determine features whose importance is greater than a preset importance threshold; and the classification performance of the machine learning model is trained based on the target features to obtain the classification model.

[0165] In some embodiments, the acquisition unit 601 is further used to obtain consensus features in the training process, wherein the consensus features are used to characterize features whose contribution to the classification performance reaches a preset contribution requirement;

[0166] The mapping unit 604 is further used to map the consensus feature back to the original brain network space to locate the consensus feature in the original brain network space;

[0167] The determination unit 602 is also used to perform brain network importance index analysis on each brain network sub-region of the original brain network space mapped with the consensus feature, so as to obtain the brain network region that contributes most significantly to the quantitative classification performance.

[0168] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0169] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, which includes: a computer program, the computer program is stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the above embodiments.

[0170] Figure 7A schematic block diagram of an example electronic device 700 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0171] like Figure 7 As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0172] A number of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0173] The computing unit 701 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 701 performs the various methods and processes described above, such as the analysis method. For example, in some embodiments, the analysis method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the analysis method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the analysis method in any other appropriate manner (e.g., by means of firmware).

[0174] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0175] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0176] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0177] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0178] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0179] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0180] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program codes.

[0181] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer executable instructions. These computer executable instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0182] These processor executable instructions may also be stored in a processor readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the processor readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0183] These processor-executable instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0184] Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure fall within the scope of the claims of the present disclosure and their equivalents, the present disclosure is also intended to include these modifications and variations.

Claims

1. A method for analyzing structural magnetic resonance imaging, characterized in that: The method comprises: Obtaining sample data, wherein the sample data is a T1-weighted structural image obtained by structural magnetic resonance imaging; Determine a map of morphological indicators of the region of interest in the original brain network space according to the sample data, wherein the morphological indicators are used to characterize the anatomical and structural properties of the cerebral cortex, and the morphological indicators include cortical thickness, sulcus depth, cortical folds and fractal dimension; Resampling the map to a preset standard brain network space template, and performing Gaussian kernel smoothing to obtain standard brain network space template data, wherein when performing Gaussian kernel smoothing, the map of cortical thickness uses a Gaussian kernel with a full width of 15 mm, and the maps of sulcus depth, cortical folds, and fractal dimension use a Gaussian kernel of 25 mm; Mapping the standard brain network space template data to a pre-constructed brain atlas template, wherein the brain atlas template divides the cerebral cortex surface into 148 regions of interest, and the 148 regions of interest are divided into 7 major brain networks, the 7 major brain networks including: visual network, somatomotor network, dorsal attention network, ventral attention network, limbic network, frontoparietal network, and default mode network; Determining the feature input of the machine learning model according to the mapped brain map template, wherein determining the feature input of the machine learning model according to the mapped brain map template comprises: for each of the 148 regions of interest, extracting the morphological index of each vertex in the region of interest, and determining the probability density of each region of interest; performing similarity measurement according to each probability density, wherein the similarity measurement comprises Kullback-Leibler divergence KLD and Jensen-Shannon divergence JSD, and the KLD and the JSD are used to characterize the morphological similarity between brain regions; determining a similarity matrix according to the KLD and the JSD, extracting upper triangular data from the similarity matrix, and constructing the extracted upper triangular data to obtain the feature input of the machine learning model; The classification performance of the machine learning model is trained based on the feature input to obtain a classification model, wherein the classification performance is the ability to distinguish the type corresponding to the T1 weighted structural image, and the classification model is used to classify the T1 weighted structural image to be classified to obtain a classification result.

2. The method according to claim 1, characterized in that The sample data is preprocessed data, and the preprocessing includes at least one of image format conversion, image quality inspection, tissue segmentation, cortical reconstruction and smoothing, segmentation quality inspection and cortical reconstruction inspection.

3. The method according to claim 1, characterized in that: The machine learning model is an unsupervised Kmeans clustering model, or a hypergraph learning-based model; In the case where the machine learning model is the model based on hypergraph learning, the hypergraph includes multiple nodes and multiple hyperedges, the nodes are used to represent elements or entities, and the hyperedges are used to represent the relationships between the elements or entities.

4. The method according to claim 1, characterized in that: The training of the classification performance of the machine learning model based on the feature input to obtain a classification model includes: Based on a preset feature selection method, selecting a target feature from the feature input, wherein the preset feature selection method includes feature selection based on statistics and / or feature selection based on F-score, the feature selection based on statistics is used to determine features whose differences are greater than a preset difference threshold, and the feature selection based on F-score is used to determine features whose importance is greater than a preset importance threshold; The classification performance of the machine learning model is trained based on the target features to obtain the classification model.

5. The method according to claim 1, characterized in that The method further comprises: Obtaining consensus features during the training process, wherein the consensus features are used to characterize features whose contribution to the classification performance reaches a preset contribution requirement; Mapping the consensus feature back to the original brain network space to locate the consensus feature in the original brain network space; The brain network importance index analysis is performed on each brain network sub-region of the original brain network space mapped with the consensus features to obtain the brain network region that contributes most significantly to the quantitative classification performance.

6. A structural magnetic resonance imaging analysis system, characterized in that: include: An acquisition unit, configured to acquire sample data, wherein the sample data is a T1-weighted structural image obtained by structural magnetic resonance imaging; A determination unit, used to determine a map of morphological indicators of the region of interest in the original brain network space according to the sample data, wherein the morphological indicators are used to characterize the anatomical and structural properties of the cerebral cortex, and the morphological indicators include cortical thickness, sulcus depth, cortical folds and fractal dimension; a processing unit, for resampling the map to a preset standard brain network space template, and performing Gaussian kernel smoothing to obtain standard brain network space template data, wherein when performing Gaussian kernel smoothing, the map of cortical thickness uses a Gaussian kernel with a full width of 15 mm, and the maps of sulcus depth, cortical folds and fractal dimension use a Gaussian kernel of 25 mm; A mapping unit, used for mapping the standard brain network space template data to a pre-constructed brain map template, wherein the brain map template divides the cerebral cortex surface into 148 regions of interest, and the 148 regions of interest are divided into 7 major brain networks, and the 7 major brain networks include: a visual network, a somatic motor network, a dorsal attention network, a ventral attention network, a limbic network, a frontoparietal network, and a default mode network; The determination unit is also used to determine the feature input of the machine learning model according to the mapped brain map template, and the determination of the feature input of the machine learning model according to the mapped brain map template includes: for each of the 148 regions of interest, extracting the morphological index of each vertex in each region of interest, and determining the probability density of each region of interest; performing similarity measurement according to each probability density, wherein the similarity measurement includes Kullback-Leibler divergence KLD and Jensen-Shannon divergence JSD, and the KLD and the JSD are used to characterize the morphological similarity between brain regions; determining a similarity matrix according to the KLD and the JSD, extracting upper triangular data from the similarity matrix, and constructing the extracted upper triangular data to obtain the feature input of the machine learning model; A training unit is used to train the classification performance of the machine learning model based on the feature input to obtain a classification model, wherein the classification performance is the ability to distinguish the type corresponding to the T1 weighted structural image, and the classification model is used to classify the T1 weighted structural image to be classified to obtain a classification result.

7. A processor-readable storage medium, characterized in that: The processor-readable storage medium stores a computer program, and the computer program is used to cause the processor to execute the method according to any one of claims 1 to 5.

Images