Autism classification feature extraction method based on dynamic brain function network attributes

Abstract

The invention discloses an autism classification feature extraction method based on dynamic brain function network attributes, and the method comprises the steps: carrying out the data preprocessing of the rs-fMRI data of a participant, and extracting a plurality of BOLD signals of the participant; the method comprises the following steps: segmenting a brain into 90 regions by using an anatomical automatic marking map, and constructing a first brain function network based on Pearson correlation coefficients of BOLD signals extracted among the regions; calculating three network topology characteristic indexes of network strength, global efficiency and an average clustering coefficient in the first brain function network of each participant; a sliding time window technology is utilized to dynamically and continuously extract BOLD signals, and three network topology characteristic indexes of network strength, global efficiency and average clustering coefficient of a rich club network, a feeder network and a local network extracted under each sliding window are calculated; and representing the key brain region in the whole time interval by adopting a dynamic Rich-club evaluation index. Verification shows that after the features participate in training, the trained model can accurately and quickly classify autism patients and health samples.

Description

Technical Field

[0001] The invention relates to an autism classification feature extraction method based on dynamic brain function network attributes, belonging to the technical field of autism classification. Background Art

[0002] Autism, also known as Autism Spectrum Disorder (ASD), is a neurodevelopmental disorder that primarily affects social interaction, communication, and behavior. The incidence of autism has gradually received attention in recent years. It usually starts in childhood and eventually causes the patient to be unable to function normally in society. In severe cases, the patient becomes disabled and requires lifelong care.

[0003] Autism is relatively hidden in children, and early symptoms are not obvious. Parents and children with autism themselves lack awareness of the social difficulties manifested by autism, which further increases the difficulty of identification. Therefore, screening and professional evaluation of children with autism are very important in order to accurately identify them in a timely manner and provide medical intervention for children in need.

[0004] At present, common diagnostic schemes are mainly based on doctors' experience and judgment. However, due to the subjective judgment of doctors, there is uncertainty in the diagnosis method and a high misdiagnosis rate. With the development of machine learning, a new solution is provided for autism classification. The key to machine learning is the construction of input feature sets, and the existing feature sets are not sufficient to support the model to achieve the expected training effect. Summary of the invention

[0005] The present invention provides an autism classification feature extraction method based on dynamic brain function network attributes, which is used to realize autism classification feature extraction based on dynamic brain function network, and then construct a feature set according to the extracted autism classification features; and further perform category classification according to the constructed feature set.

[0006] The technical solution of the present invention is:

[0007] According to a first aspect of the present invention, a method for extracting autism classification features based on dynamic brain functional network attributes is provided, comprising the following steps:

[0008] Step S1: performing data preprocessing on the resting-state functional magnetic resonance imaging data of the participant to extract the first blood oxygen level dependent signal of the participant;

[0009] Step S2: using an anatomical automatic labeling atlas to segment the brain into 90 regions; using the 90 brain regions as network nodes, calculating the Pearson correlation coefficient of the first blood oxygen level dependent signal between every two regions in the 90 regions to construct a first brain functional network;

[0010] Step S3: Calculate the three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network of each participant;

[0011] Step S4: using a sliding time window technique, dynamically and continuously extracting the first blood oxygen level dependent signal to construct a second brain functional network under each sliding window;

[0012] Step S5: Calculate the node degree centrality value of each node in the second brain functional network under each sliding window, and determine the rich club area and the non-rich club area; the connection between the nodes in the rich club area constitutes the rich club network, the connection between the nodes in the rich club area and the non-rich club area constitutes the feeder network, and the connection between the nodes in the non-rich club area constitutes the local area; calculate the network strength, global efficiency and average clustering coefficient of the three network topology characteristic indicators of the rich club network, feeder network and local network extracted under each sliding window respectively;

[0013] Step S6: using dynamic Rich-club evaluation indicators to characterize the key brain regions in the entire time interval and quantitatively describe the dynamic Rich-club characteristics of each participant;

[0014] Step S7: The three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network calculated in the above steps and the three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient of the rich club network, feeder network and local network extracted under each sliding window and the dynamic Rich-club evaluation indicator are combined to construct a feature set.

[0015] Furthermore, the network strength is specifically expressed as follows:

[0016]

[0017] Where G is the network strength; n is the number of nodes in the network; d i is the node degree centrality value of the i-th network node.

[0018] Furthermore, the global efficiency is specifically formulated as follows:

[0019]

[0020] Among them, E global is the global efficiency, n is the number of nodes in the network; N = 1, 2, ... n; d ij It is the shortest path length between the ith network node and the jth network node. If there is no path connecting the two nodes, it is defined as infinity.

[0021] Furthermore, the specific formula of the average clustering coefficient is:

[0022]

[0023] Among them, C is the average clustering coefficient, n is the number of nodes in the network, T i is the number of nodes directly connected to the ith network node, d i is the node degree centrality value of the i-th network node.

[0024] Furthermore, the method of dynamically and continuously extracting the first blood oxygen level dependent signal using the sliding time window technology to construct a second brain functional network under each sliding window specifically includes:

[0025] The first blood oxygen level dependence signal is continuously extracted according to the preset sliding window size and time step. The brain is divided into 90 regions using the anatomical automatic labeling atlas. For the second blood oxygen level dependence signal under each sliding window, the 90 brain regions are used as network nodes. The Pearson correlation coefficient of the second blood oxygen level dependence signal between every two regions in the 90 brain regions is calculated to construct the second brain functional network under each sliding window.

[0026] Furthermore, the dynamic Rich-club evaluation indicators specifically include:

[0027] Step S61: Calculate the importance of each network node in all second brain functional networks. The specific formula is as follows:

[0028]

[0029] Among them, D i represents the importance of the i-th network node in all sliding windows, F represents the total number of sliding windows, and d ij represents the node degree centrality value of the i-th network node as the core node of the brain region in the j-th sliding window; f i represents the number of times the i-th network node is a node in the rich club region in all sliding windows;

[0030] Step S62: Calculate the importance percentage of the brain regions in the rich club area. The specific formula is as follows:

[0031]

[0032] Where P is the percentage of importance of the brain region in the rich club area, D k is the importance value of the kth network node, D global is the sum of the importance of all network nodes.

[0033] According to a second aspect of the present invention, there is provided an apparatus for extracting features for autism classification based on dynamic brain functional network attributes, comprising a module for executing a method for extracting features for autism classification based on dynamic brain functional network attributes as described above.

[0034] According to a third aspect of the present invention, there is provided a terminal device comprising a memory, a processor, and a program stored in the memory and executable by the processor, wherein when the processor executes the program, the method for extracting autism classification features based on dynamic brain functional network properties as described in any one of the above is implemented.

[0035] The beneficial effects of the present invention are:

[0036] The present invention can perform data preprocessing on the rs-fMRI data of the participants, extract multiple BOLD signals of the participants; use the anatomical automatic labeling atlas to divide the brain into 90 regions, and construct the first brain functional network based on the Pearson correlation coefficient of the BOLD signals between the extracted regions; calculate the three network topological characteristic indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network of each participant; further, use the sliding time window technology to dynamically and continuously extract the BOLD signal; calculate the degree centrality value of each node in the dynamic second brain functional network, and extract the top 10 nodes with the largest degree centrality values ​​to form a rich person network. Club (Rich-club) area; dynamic Rich-club evaluation indicators are used to characterize the key brain areas within the entire time interval, and the dynamic Rich-club characteristics of each participant are quantitatively described to obtain the autism classification feature characteristics; further, a feature set is constructed according to the autism classification features, and a support vector machine classification method is used to classify autistic patients and healthy participants to verify the training effect of the feature set of the present invention. The verification shows that after the feature set of the present invention participates in the training, the trained model can accurately and quickly classify autistic patients and healthy samples. The process is objective and accurate, and has reliability, ease of use and extensiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other arbitrarily without conflict.

[0039] Example 1: Figure 1 As shown, according to a first aspect of an embodiment of the present invention, a method for extracting autism classification features based on dynamic brain functional network attributes is provided, comprising the following steps:

[0040] Step S1: Preprocess the resting-state functional magnetic resonance imaging data of the participant to extract the first blood oxygen level-dependent BOLD signal of the participant; the preprocessing includes removing unstable time points, correcting time layers, correcting head motion, spatial standardization, spatial smoothing, regressing oblique variables and filtering.

[0041] Step S2: using an anatomical automatic labeling atlas to segment the brain into 90 regions; using the 90 brain regions as network nodes, calculating the Pearson correlation coefficient of the first blood oxygen level dependent signal between every two regions in the 90 regions to construct a first brain functional network;

[0042] According to the above steps, the first brain functional network constructed is a 90×90 matrix;

[0043] Step S3: Calculate the three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network of each participant;

[0044] The calculation of the three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network of each participant specifically includes:

[0045] Step S31: Calculate the network strength of the first brain functional network of each participant. The specific formula is:

[0046]

[0047] Wherein, G is the network strength, which is the sum of the node degrees of all nodes in the first brain functional network; n is the number of nodes in the network (90 in the embodiment of the present invention); d i is the node degree centrality value of the i-th network node.

[0048] Step S32: Calculate the global efficiency of the first brain functional network of each participant. The specific formula is:

[0049]

[0050] Among them, E global is the global efficiency, n is the number of nodes in the network; N = 1, 2, ... n; d ij It is the shortest path length between the ith network node and the jth network node. If there is no path connecting the two nodes, it is defined as infinity.

[0051] Step S33: Calculate the average clustering coefficient in the first brain functional network of each participant. The specific formula is:

[0052]

[0053] Among them, C is the average clustering coefficient, n is the number of nodes in the network, T i is the number of nodes directly connected to the ith network node, d i is the node degree centrality value of the i-th network node.

[0054] Step S4: using a sliding time window technique, dynamically and continuously extracting the first blood oxygen level dependent signal to construct a second brain functional network under each sliding window;

[0055] The method uses a sliding time window technology to dynamically and continuously extract the first blood oxygen level dependence signal to construct a second brain functional network under each sliding window, specifically including: using a sliding window size of 30 and a time step of 1 to continuously extract the first blood oxygen level dependence signal; using an anatomical automatic labeling atlas to divide the brain into 90 regions, and for the second blood oxygen level dependence signal under each sliding window, using 90 brain regions as network nodes, calculating the Pearson correlation coefficient of the second blood oxygen level dependence signal between every two regions in the 90 brain regions to construct the second brain functional network under each sliding window.

[0056] Step S5: Calculate the node degree centrality of each node in the second brain functional network under each sliding window, and extract the nodes corresponding to the top 10 largest node degree centrality values ​​to form the rich club area (Rich club), and the remaining nodes form the non-rich club area; the connections between the nodes in the rich club area form the rich club network, the connections between the nodes in the rich club area and the non-rich club area form the feeder network (Freeder), and the connections between the nodes in the non-rich club area form the local area; calculate the network strength, global efficiency and average clustering coefficient of the three network topology characteristic indicators of the rich club network, feeder network and local network extracted under each sliding window respectively, and the calculation method of each indicator is the same as step S3.

[0057] Step S6: using dynamic Rich-club evaluation indicators to characterize the key brain regions in the entire time interval and quantitatively describe the dynamic Rich-club characteristics of each participant;

[0058] The dynamic Rich-club evaluation indicators specifically include:

[0059] Step S61: Calculate the importance of each network node in all second brain functional networks. The specific formula is as follows:

[0060]

[0061] Among them, D i represents the importance of the i-th network node in all sliding windows, F represents the total number of sliding windows, and d ij represents the node degree centrality value of the i-th network node as the core node of the brain region in the j-th sliding window. At this time, if the network node is not a node in the rich club area in the j-th sliding window, then d ij is 0; if the network node is a node in the rich club area in the jth sliding window, then d ij is the node degree centrality value; f i represents the number of times the i-th network node is a node in the rich club region in all sliding windows.

[0062] Step S62: Calculate the importance percentage of the brain regions in the rich club region, that is, the top 10 largest D i The ratio of the sum of the values ​​to the sum of the importance of all network nodes. The specific formula is as follows:

[0063]

[0064] Where P is the percentage of importance of the brain region in the rich club area, D k is the importance value of the kth network node, that is, the top 10 largest importance values ​​are selected; D global is the sum of the importance of all network nodes, that is, D1+D2+....+D 90 .

[0065] Step S7: The three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network calculated in the above steps and the three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient of the rich club network, feeder network and local network extracted under each sliding window and the dynamic Rich-club evaluation indicator are combined to construct a feature set; the feature set is trained with a support vector machine and then a 10-fold cross validation is used to evaluate the generalization performance of the model.

[0066] Specifically include:

[0067] S71: Randomly divide the entire feature set into 10 subsets of equal size.

[0068] S72: In each iteration, 9 subsets are selected as training sets and the remaining 1 subset is used as validation set. The training process uses the support vector machine algorithm, and then the performance of the model is tested on the validation set to calculate the model's accuracy, recall, and F1 score.

[0069] In order to illustrate the effectiveness of the present invention, the following experiments are carried out:

[0070] The experiment used a public dataset published by the Kennedy Krieger Institute (https: / / fcon_1000.projects.nitrc.org / ind i / abide / abide_II.html), with a total of 211 data samples of children aged 8-13 years old. After data quality control and standardized diagnosis of autism, a total of 206 samples were included in the study, including 152 children with autism and 54 healthy controls.

[0071] The experiment constructed three feature sets, and the feature sets were trained using support vector machines, and the generalization performance of the model was evaluated using 10-fold cross validation. The accuracy, recall and F1 scores of the models were calculated to evaluate the performance of the models trained with different feature sets on the validation set. Feature set 1 includes the three network topology feature indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network of each participant. Feature set 2 includes the three network topology feature indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network, and the three network topology feature indicators of network strength, global efficiency and average clustering coefficient of the rich club network, feeder network and local network extracted under each sliding window. Feature set 3 includes the three network topology feature indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network, and the three network topology feature indicators of network strength, global efficiency and average clustering coefficient of the rich club network, feeder network and local network extracted under each sliding window, as well as the dynamic Rich-club evaluation index. The results are shown in Table 1. The results show that the accuracy, recall and F1 score of the model trained with feature set 3 in classifying autism and healthy samples are higher than those of feature set 1 and feature set 2, indicating that the classification accuracy and generalization of the model trained with feature set 3 are improved.

[0072] Table 1 Classification results of three feature sets

[0073]

[0074] According to a second aspect of an embodiment of the present invention, there is provided an autism classification feature extraction device based on dynamic brain functional network attributes, specifically comprising: a first module for executing step S1: performing data preprocessing on the resting-state functional magnetic resonance imaging data of the participant to extract the participant's first blood oxygen level dependent signal; a second module for executing step S2: using an anatomical automatic labeling atlas to divide the brain into 90 regions; using the 90 brain regions as network nodes, calculating the Pearson correlation coefficient of the first blood oxygen level dependent signal between each two regions in the 90 regions to construct a first brain functional network; a third module for executing step S3: calculating three network topology feature indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network of each participant; a fourth module for executing step S4: using a sliding time window technology to dynamically and continuously extract the first blood oxygen level dependent signal to construct a second brain functional network under each sliding window; a fifth module for executing step S5: calculating the node degree center of each node in the second brain functional network under each sliding window value, and determine the rich club area and the non-rich club area; the connection between the nodes of the rich club area constitutes the rich club network, the connection between the nodes of the rich club area and the non-rich club area constitutes the feeder network, and the connection between the nodes of the non-rich club area constitutes the local area; the network strength, global efficiency and average clustering coefficient of the rich club network, feeder network and local network extracted under each sliding window are calculated respectively; the sixth module is used to execute step S6: the dynamic Rich-club evaluation index is used to characterize the key brain areas in the entire time interval, and the dynamic Rich-club characteristics of each participant are quantitatively described; the seventh module is used to execute step S7: the network strength, global efficiency and average clustering coefficient of the first brain functional network calculated in the above steps, the three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient of the rich club network, feeder network and local network extracted under each sliding window, and the dynamic Rich-club evaluation index are combined to construct a feature set. As used above, the term "module" can be a combination of software and / or hardware to implement a predetermined function. Although the system described in the above embodiment is preferably implemented by software, implementation by hardware, or a combination of software and hardware is also possible and conceivable. For parts that are not described in detail in each module, please refer to the relevant description of this embodiment.

[0075] According to a third aspect of an embodiment of the present invention, there is provided a terminal device, comprising a memory, a processor, and a program stored in the memory and executable by the processor, wherein when the processor executes the program, the method for extracting autism classification features based on dynamic brain functional network properties as described in any one of the above is implemented.

[0076] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A method for extracting autism classification features based on dynamic brain functional network attributes, characterized in that: The following steps are involved: Step S1: performing data preprocessing on the resting-state functional magnetic resonance imaging data of the participant to extract the first blood oxygen level dependent signal of the participant; Step S2: using an anatomical automatic labeling atlas to segment the brain into 90 regions; using the 90 brain regions as network nodes, calculating the Pearson correlation coefficient of the first blood oxygen level dependent signal between every two regions in the 90 regions to construct a first brain functional network; Step S3: Calculate the three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network of each participant; Step S4: using a sliding time window technique, dynamically and continuously extracting the first blood oxygen level dependent signal to construct a second brain functional network under each sliding window; Step S5: Calculate the node degree centrality value of each node in the second brain functional network under each sliding window, and determine the rich club area and the non-rich club area; the connection between the nodes in the rich club area constitutes the rich club network, the connection between the nodes in the rich club area and the non-rich club area constitutes the feeder network, and the connection between the nodes in the non-rich club area constitutes the local area; calculate the network strength, global efficiency and average clustering coefficient of the three network topology characteristic indicators of the rich club network, feeder network and local network extracted under each sliding window respectively; Step S6: using dynamic Rich-club evaluation indicators to characterize the key brain regions in the entire time interval and quantitatively describe the dynamic Rich-club characteristics of each participant; Step S7: The three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient in the first brain functional network calculated in the above steps and the three network topology characteristic indicators of network strength, global efficiency and average clustering coefficient of the rich club network, feeder network and local network extracted under each sliding window and the dynamic Rich-club evaluation indicator are combined to construct a feature set.

2. The method for extracting autism classification features based on dynamic brain functional network attributes according to claim 1, characterized in that: The network strength is specifically formulated as follows: Where G is the network strength; n is the number of nodes in the network; d i is the node degree centrality value of the i-th network node.

3. The autism classification feature extraction method based on dynamic brain function network attributes according to claim 1, characterized in that: The global efficiency is specifically formulated as follows: Among them, E global is the global efficiency, n is the number of nodes in the network; N = 1, 2, ... n; d ij It is the shortest path length between the ith network node and the jth network node. If there is no path connecting the two nodes, it is defined as infinity.

4. The autism classification feature extraction method based on dynamic brain function network attributes according to claim 1, characterized in that: The specific formula of the average clustering coefficient is: Among them, C is the average clustering coefficient, n is the number of nodes in the network, T i is the number of nodes directly connected to the ith network node, d i is the node degree centrality value of the i-th network node.

5. The autism classification feature extraction method based on dynamic brain function network attributes according to claim 1, characterized in that: The method of dynamically and continuously extracting the first blood oxygen level dependent signal by using the sliding time window technology to construct the second brain functional network under each sliding window specifically includes: The first blood oxygen level dependence signal is continuously extracted according to the preset sliding window size and time step. The brain is divided into 90 regions using the anatomical automatic labeling atlas. For the second blood oxygen level dependence signal under each sliding window, the 90 brain regions are used as network nodes. The Pearson correlation coefficient of the second blood oxygen level dependence signal between every two regions in the 90 brain regions is calculated to construct the second brain functional network under each sliding window.

6. The autism classification feature extraction method based on dynamic brain function network attributes according to claim 1, characterized in that: The dynamic Rich-club evaluation indicators specifically include: Step S61: Calculate the importance of each network node in all second brain functional networks. The specific formula is as follows: Among them, D i represents the importance of the i-th network node in all sliding windows, F represents the total number of sliding windows, and d ij represents the node degree centrality value of the i-th network node as the core node of the brain region in the j-th sliding window; f i represents the number of times the i-th network node is a node in the rich club region in all sliding windows; Step S62: Calculate the importance percentage of the brain regions in the rich club area. The specific formula is as follows: Where P is the percentage of importance of the brain region in the rich club area, D k is the importance value of the kth network node, D global is the sum of the importance of all network nodes.

7. A device for extracting autism classification features based on dynamic brain functional network attributes, characterized in that: It comprises a module for executing the autism classification feature extraction method based on dynamic brain functional network attributes as described in any one of claims 1-6.

8. A terminal device, characterized in that: It includes a memory, a processor, and a program stored in the memory and executable by the processor. When the processor executes the program, it implements the autism classification feature extraction method based on dynamic brain functional network attributes as described in any one of claims 1 to 6.

Images