A method of identifying a diagnostic biomarker for autism

Abstract

The application discloses a method for identifying autism diagnostic biomarkers, and belongs to the technical field of computer application. The method comprises the following steps: processing BOLD signals by using a PCMCI algorithm, regarding different regions of the brain as nodes, and regarding the strength of the causal relationship between each pair of brain regions as the weight of the edge, so as to generate a weighted directed causal network for each subject. Differences in brain causal networks between an autism group and a normal control group are evaluated by using a two-sample T test, and potential autism diagnostic biomarkers are determined. Further, a support vector machine is used to build an autism classification model, so as to verify the reliability of the potential autism diagnostic biomarkers. The application can better mine the causal relationship between various regions in the brain and identify autism diagnostic biomarkers, and provides a new perspective and method for autism research.

Description

Technical Field

[0001] This invention relates to a method for identifying diagnostic biomarkers for autism, belonging to the field of computer application technology. Background Technology

[0002] Autism spectrum disorder is a complex and highly heterogeneous neurodevelopmental disorder. Timely and accurate identification of autism facilitates targeted interventions for patients, thereby reducing their stress and burden. Current diagnostic methods for autism rely on relatively inefficient and somewhat subjective scale assessments. Therefore, research into objective diagnostic biomarkers is crucial. Identifying diagnostic biomarkers for autism based on effective brain connectivity is a commonly used approach.

[0003] Commonly used methods for calculating effective connectivity in the brains of autistic patients include Granger causal models, dynamic causal models, and entropy. Sato et al. ("Impaired social brain network for processing dynamic facial expressions in autism spectrum disorders." BMC neuroscience, 2012, 13, 1-17.) used dynamic causal models to construct brain networks. The study showed that, compared with normal controls, the autistic group had reduced activation in brain regions such as the middle temporal gyrus, fusiform gyrus, amygdala, medial prefrontal cortex, and inferior frontal gyrus in dynamic and static facial expression responses. However, dynamic causal models are highly dependent on prior knowledge, so the correctness of the causal relationships obtained is related to prior knowledge. Bernas et al. ("Brain resting state networks in adolescents with highfunctioning autism: Analysis of spatial connectivity and temporalneurodynamics." Brain and Behavior, 2018, 8(2), e00878.) used Granger causal analysis to construct brain networks and found significant differences in causal relationships originating from the temporal lobe and prefrontal cortex in autistic patients. However, a larger Granger causality value does not necessarily indicate a stronger true causal relationship. Zhao et al. ("Changes of entropy connectivity of the default mode and central executive networks in ASD." Research Square, 2021) used entropy connectivity to find that in the autism group, causal connectivity from the right dorsolateral prefrontal cortex to the left supramarginal gyrus was weakened, while causal connectivity from the right dorsolateral prefrontal cortex to the left piriform cortex was enhanced. However, calculating entropy requires long time series and is computationally complex.Cong et al. (“Altered default mode network causal connectivity patterns in autism spectrum disorder revealed by Liang information flow analysis”, Human Brain Mapping, 2023, 44(6), 2279-2293.) used Liang's information flow to observe increased differences in causal connectivity between regions involving the dorsal and ventromedial prefrontal cortex, hippocampus formation, and temporoparietal junction in autistic patients. However, Liang's information flow showed limited performance when dealing with larger sets of variables.

[0004] Therefore, current methods for identifying autism diagnostic biomarkers based on causal models still suffer from problems such as low accuracy and high computational cost, which limit the overall performance of autism identification. Summary of the Invention

[0005] To improve the accuracy of identifying autism diagnostic biomarkers, this invention provides a method for identifying autism diagnostic biomarkers, comprising:

[0006] Step 1: Obtain the BOLD signal of the subject, process the BOLD signal using the PCMCI algorithm, calculate the causal relationship between each pair of brain regions, and construct a causal brain network for the subject.

[0007] Step 2: Extract significantly differential causal connections from the causal brain network;

[0008] Step 3: Input the significant difference causal connection into the support vector machine classification model to obtain the identification results of autism diagnostic biomarkers;

[0009] The significant difference in causal connections was obtained by comparing the causal relationships between each pair of brain regions between the autism group and the normal control group using a two-sample t-test.

[0010] Optionally, step 1 includes:

[0011] Step 11: BOLD signal of the subject It can be viewed as a multivariable discrete-time stochastic process. Represents matrix transpose, brain region The BOLD signal is , For the number of brain regions, and Indicates the brain region number, Indicates time;

[0012] Time series plot is defined as The nodes in the diagram are formed by each Independent time-related brain region variables It is composed of Z, which is defined as a set of integers used to represent discrete time indices or time steps, and E, which represents the set of directed edges in the time series graph. Indicates the process at the current time. Representing a past process, initializing brain region variables. The set of causal parent nodes , Indicates a time lag. Indicates lag The maximum value that can be selected;

[0013] Step 12: In the first iteration, use partial correlation to perform an unconditional independence test. If Not independent ( Indicates lag, and can take values. ), then from Delete it, and then adjust it according to the test statistic. Midbrain region variable ranking, express Except The largest external statistical value A set consisting of variables from brain regions, where, The range is from 1 to the maximum value, and the theoretical maximum value is... Except The number of outside elements, at the same time It can also represent the number of iterations and sums. The number of elements in the middle.

[0014] In each subsequent iteration, the updated version was used. As a condition set, partial correlation is used to test conditional independence. Brain region variables that do not meet the independence condition are removed from Delete until no more conditions can be tested, then output. ;

[0015] Step 13: In step 12, we obtained... Estimate ,definition For set Translation time of all brain regions The resulting set is then used to perform a conditional independence test using partial correlation. Here, the p-value and test statistic for causal relationships between brain regions were estimated. If p Then the causal relationship is considered to be... Existence (i.e.) yes (The causal parent node), otherwise it does not exist;

[0016] The above steps use the PCMCI algorithm to estimate the causal parent node of each brain region, and the test statistic gives the strength of the causal relationship between each causal parent node and the brain region.

[0017] Step 14: Construct a weighted directed causal brain network with brain regions as nodes, causal relationships as edges, and the strength of causal relationships as edge weights.

[0018] Optionally, the screening process for significantly different causal connections includes:

[0019] Step 21: In the causal brain network, for each pair of causal connections between brain regions, calculate the mean for the autism group and the normal control group respectively, and then perform a two-sample t-test to test whether there is a significant difference between the means of the two groups on the causal connection. Sort all causal connections by their two-sample t-test p-values ​​from smallest to largest, and select the M causal connections with the smallest lag p-values ​​as the causal connections with significant differences.

[0020] Step 22: For causal connections with significant differences, subtract the mean of the causal connections of the corresponding brain region pairs in the normal control group from the mean in the autism group to obtain the changes in the strength of these causal connections with significant differences.

[0021] Optionally, in step 11 Set to 2.

[0022] Optionally, the training process of the support vector machine includes:

[0023] Step 31: Select the significantly different causal connections from the causal brain network of each subject as feature vectors for identifying autism;

[0024] Step 32: Using the feature vector as input, train the support vector machine classifier, and use the grid search method to evaluate the effectiveness of the classification model with accuracy as the evaluation index, and find the optimal support vector machine parameters;

[0025] Step 33: Set the parameters of the support vector machine to the optimal values ​​obtained in step 32, thereby obtaining the trained classification model.

[0026] Optionally, the support vector machine parameters include the kernel function, the penalty coefficient C, the kernel function coefficient gamma, and the degree of the polynomial kernel function.

[0027] A second objective of the present invention is to provide a system for identifying diagnostic biomarkers for autism, the system being used to implement the method described in any of the preceding claims, the system comprising:

[0028] The causal brain network construction module is configured to acquire the BOLD signal of the subject, process the BOLD signal using the PCMCI algorithm, calculate the causal relationship between each pair of brain regions, and construct a causal brain network for the subject.

[0029] A significantly differential causal connection extraction module is configured to extract significantly differential causal connections from the causal brain network;

[0030] The identification module is configured to input the significant difference causal connection into the support vector machine classification model to obtain the identification results of autism diagnostic biomarkers;

[0031] The significant difference in causal connections was obtained by comparing the causal relationships between each pair of brain regions between the autism group and the normal control group using a two-sample t-test.

[0032] A third object of the present invention is to provide an electronic device, including a memory and a processor;

[0033] The memory is used to store computer programs;

[0034] The processor is configured to, when executing the computer program, implement the method as described in any of the preceding descriptions.

[0035] A fourth object of the present invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in any of the preceding claims.

[0036] The beneficial effects of this invention are:

[0037] The autism diagnostic biomarker identification method of the present invention calculates causal relationships between brain regions based on the PCMCI algorithm and constructs an individual-level causal brain network. It extracts causal connections that are significantly different between autistic individuals and normal individuals from the causal brain network, and uses these causal connections as features to input a support vector machine classification model for identification and classification, providing a new method for identifying autism diagnostic biomarkers. Experimental results show that the autism classification model trained by the present invention achieves excellent performance in terms of accuracy, sensitivity, specificity, F1 score, and AUC curve, and achieves high accuracy and identification efficiency for autism diagnostic biomarker identification.

[0038] This invention uses a two-sample t-test to perform intergroup analysis between the autism group and the normal control group, thereby identifying causal connections with significant differences. These causal connections are then used as diagnostic biomarkers for autism, providing a new approach to the identification and diagnosis of autism. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a visualization of the significant differences in autism selected in Embodiment 2 of the present invention, connected in the brain. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0042] Example 1:

[0043] This embodiment provides a method for identifying diagnostic biomarkers for autism, including:

[0044] Step 1: Obtain the BOLD signal of the subject, process the BOLD signal using the PCMCI algorithm, calculate the causal relationship between each pair of brain regions, and construct a causal brain network for the subject.

[0045] Step 2: Extract significantly differential causal connections from causal brain networks;

[0046] Step 3: Input the significantly different causal connections into the support vector machine classification model to obtain the identification results of autism diagnostic biomarkers;

[0047] Significantly different causal connections were obtained by comparing the causal relationships between each pair of brain regions between the autism group and the normal control group using a two-sample t-test.

[0048] Example 2

[0049] This embodiment provides a method for identifying biomarkers for autism diagnosis. The subjects are data from 167 subjects selected from the ABIDE database NYU site, including 72 autistic subjects and 95 normal control subjects. The PCMCI algorithm is used to construct a causal brain network for each subject.

[0050] For the connections in the causal brain network, a two-sample t-test was used to conduct inter-group analysis between the autism group and the normal control group to select causal connections with significant differences, which were initially identified as potential diagnostic biomarkers for autism. Furthermore, features were selected from the causal brain network based on the significant differences in connections as inputs to train a support vector machine classifier, which was then used to verify the potential of these causal connections as diagnostic biomarkers for autism.

[0051] Specifically, it includes the following steps:

[0052] S1: Data from 167 subjects were obtained from the ABIDE database NYU site, including 72 ASD subjects and 95 normal control subjects. The data for each subject consisted of BOLD time-series signals corresponding to 116 brain regions divided by the AAL template. For the BOLD data of each subject, a causal brain network was constructed according to the following procedure.

[0053] (1) The BOLD signal of a subject It can be viewed as a multivariate discrete-time stochastic process, where (*)' denotes matrix transpose, brain region The BOLD signal is , For the number of brain regions, and Indicates the brain region number, Indicates time. Time series plot is defined as The nodes in the diagram are formed by each Independent time-related brain region variables It is composed of Z, which is defined as a set of integers used to represent discrete time indices or time steps, and E, which represents the set of directed edges in the time series graph. Indicates the process at the current time. Represents a process in the past. Initializes brain region variables. The set of causal parent nodes , Indicates a time lag. Indicates lag The maximum value that can be selected will Set to 2.

[0054] (2) In the first iteration ( Using partial correlation to perform unconditional independence tests ( , (indicating independence), if and Not independent ( Indicates lag, and can take values. ), then from Delete it, and then adjust it according to the test statistic (absolute value). Midbrain region variable ranking, express Except The largest external statistical value (absolute value) The set of brain region variables, in each subsequent iteration ( ), using the updated As a condition set, partial correlation is used to test conditional independence. ), removing brain region variables that do not meet the independence condition from Delete until no more conditions can be tested, then output. .

[0055] (3) In step (2), we obtained Estimate ,definition For set Translation time of all brain regions The resulting set is then used to perform a conditional independence test using partial correlation. Here, the p-value and test statistic for causal relationships between brain regions were estimated. If p Then the causal relationship is considered to be... Existence (i.e.) yes (The causal parent node), otherwise it does not exist.

[0056] The above steps use the PCMCI algorithm to estimate the causal parent node for each brain region, and the test statistic gives the strength of the causal relationship between each causal parent node and the brain region.

[0057] (4) Construct a weighted directed causal brain network with brain regions as nodes, causal relationships as edges, and the strength of causal relationships as edge weights.

[0058] S2: For connections in causal brain networks, a two-sample t-test was used to conduct between-group analysis between the autism group and the normal control group to select causal connections with significant differences.

[0059] (1) In the causal brain network, a two-sample t-test with a significance threshold of 0.05 was performed on all causal connections. The two-sample t-test p values ​​of all causal connections were sorted from smallest to largest. The 10 causal connections with the smallest p values ​​for each lag were selected as causal connections with significant differences (10 with lag of 0, 10 with lag of 1, and 10 with lag of 2).

[0060] (2) For causal connections with significant differences, the mean of the causal connections of the corresponding brain regions in the normal control group was subtracted from the mean in the autism group to obtain the intensity changes of these causal connections with significant differences, which are presented in Table 1 (causal connections with a lag of 0 have no direction, and the direction of causal connections with a lag of 1 or 2 is: brain region 1 → brain region 2).

[0061] Figure 1 In this study, BrainNet was used to visualize the selected significantly differential connections in the brain for easy observation.

[0062] Table 1. Significant differences in autism compared to normal controls.

[0063]

[0064] S3: Based on the significantly different connections selected in S2, select the corresponding connections from the causal brain network of each subject and arrange them to form features for identifying autism. Use these features as input to train a support vector machine classifier. During training, use a grid search method to find the optimal kernel function, penalty coefficient C, kernel function coefficient gamma, and degree of the polynomial kernel function. The kernel value is in ['linear', 'rbf', 'poly'], the C value is in [0.001, 0.01, 0.1, 1, 10, 100], the gamma value is in [0.001, 0.01, 0.1, 'scale', 'auto'], and the degree can be 2.

[0065] The classification model trained using the above process was used to identify autism. The performance of the model, evaluated using 10-fold cross-validation, is shown in Table 2.

[0066] Table 2 Classification Model

[0067]

[0068] Accuracy (ACC), Sensitivity (SEN), Specificity (SPE), F1 Score, and Area Under Curve (AUC) are metrics for measuring the classification performance of a model. Their definitions are as follows:

[0069] The calculation methods for each indicator are as follows ( The actual diagnosis was autism, which was correctly identified as autism by the classifier. The actual diagnosis was autism, but the classifier misclassified it as a normal control. The actual group consisted of normal controls, but the classifier misclassified them as autism. (This was actually a normal control group, which the classifier correctly identified as a normal control group.)

[0070] 1. Accuracy: Represents the proportion of correctly classified samples out of all samples, reflecting the overall classification ability of the model.

[0071]

[0072] 2. Sensitivity: This represents the proportion of autism correctly classified out of the total number of actual autism samples. Also known as recall, it reflects the model's ability to identify autism subjects.

[0073]

[0074] 3. Specificity: This represents the proportion of correctly classified normal controls to the total number of actual normal control samples, reflecting the model's ability to identify patients among normal control subjects.

[0075]

[0076] 4. F1 score: It is the harmonic mean of precision and recall, ranging from 0 to 1. The closer it is to 1, the better the model's output.

[0077]

[0078] 5. Area Under the Receiver Operating Characteristic Curve (AUC): This is the area under the Receiver Operating Characteristic Curve, with a value ranging from [0, 1]. AUC closer to 1 indicates better model performance; AUC closer to 0.5 indicates worse model performance.

[0079]

[0080]

[0081] in, For the number of autistic subjects, The number of subjects is the normal control group. To predict scores for autistic subjects, The predicted score is for the normal control subjects.

[0082] Example 3:

[0083] This embodiment provides a system for identifying diagnostic biomarkers for autism, used to implement the above-described method, the system comprising:

[0084] The causal brain network construction module is configured to acquire the BOLD signal of the subject, process the BOLD signal using the PCMCI algorithm, calculate the causal relationship between each pair of brain regions, and construct a causal brain network for the subject.

[0085] The module for extracting significantly different causal connections is configured to extract significantly different causal connections from causal brain networks.

[0086] The identification module is configured to input significantly different causal connections into a support vector machine classification model to obtain the identification results of autism diagnostic biomarkers.

[0087] Significantly different causal connections were obtained by comparing the causal relationships between each pair of brain regions between the autism group and the normal control group using a two-sample t-test.

[0088] Some steps in the embodiments of the present invention can be implemented using software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying diagnostic biomarkers for autism, characterized in that, The method includes: Step 1: Obtain the BOLD signal of the subject, process the BOLD signal using the PCMCI algorithm, calculate the causal relationship between each pair of brain regions, and construct a causal brain network for the subject. Step 2: Extract significantly differential causal connections from the causal brain network; Step 3: Input the significant difference causal connection into the support vector machine classification model to obtain the identification results of autism diagnostic biomarkers; The significant difference in causal connections was obtained by comparing the causal relationships between each pair of brain regions between the autism group and the normal control group using a two-sample t-test. Step 1 includes: Step 11: BOLD signal of the subject It can be viewed as a multivariable discrete-time stochastic process. Represents matrix transpose, brain region The BOLD signal is , For the number of brain regions, and Indicates the brain region number, Indicates time; Time series plot is defined as The nodes in the diagram are formed by each Independent time-related brain region variables It is composed of Z, which is defined as a set of integers used to represent discrete time indices or time steps, and E, which represents the set of directed edges in the time series graph. Indicates the process at the current time. Representing a past process, initializing brain region variables. The set of causal parent nodes , Indicates a time lag. Indicates lag The maximum value; Step 12: In the first iteration, use partial correlation to perform an unconditional independence test. If If not independent, then from Delete it, and then adjust it according to the test statistic. Midbrain region variable ranking, express Except The largest external statistical value A set consisting of variables from individual brain regions; In each subsequent iteration, the updated version was used. As a condition set, partial correlation is used to test conditional independence. Brain region variables that do not meet the independence condition are removed from Delete until no more conditions can be tested, then output. ; Step 13: In step 12, we obtained... Estimate ,definition For set Translation time of all brain regions The resulting set is then used to perform a conditional independence test using partial correlation. Here, the p-value and test statistic for causal relationships between brain regions were estimated. If p Then the causal relationship is considered to be... It exists, or it does not exist. The above steps use the PCMCI algorithm to estimate the causal parent node of each brain region, and the test statistic gives the strength of the causal relationship between each causal parent node and the brain region. Step 14: Construct a weighted directed causal brain network with brain regions as nodes, causal relationships as edges, and the strength of causal relationships as edge weights; The screening process for the significant difference causal connections includes: Step 21: In the causal brain network, for each pair of causal connections between brain regions, calculate the mean for the autism group and the normal control group respectively, and then perform a two-sample t-test to test whether there is a significant difference between the means of the two groups on the causal connection. Sort all causal connections by their two-sample t-test p-values ​​from smallest to largest, and select the M causal connections with the smallest lag p-values ​​as the causal connections with significant differences. Step 22: For causal connections with significant differences, subtract the corresponding mean from the normal control group using the mean in the autism group to obtain the changes in the strength of these causal connections with significant differences.

2. The method according to claim 1, characterized in that, In step 11 Set to 2.

3. The method according to claim 1, characterized in that, The training process of the support vector machine classification model in step 3 includes: Step 31: Select the significantly different causal connections from the causal brain network of each subject as feature vectors for identifying autism; Step 32: Using the feature vector as input, train the support vector machine classifier, and use the grid search method to evaluate the effectiveness of the classification model with accuracy as the evaluation index, and find the optimal support vector machine parameters; Step 33: Set the parameters of the support vector machine to the optimal values ​​obtained in step 32, thereby obtaining the trained classification model.

4. The method according to claim 3, characterized in that, The parameters of the support vector machine include the kernel function, the penalty coefficient C, the kernel function coefficient gamma, and the degree of the polynomial kernel function.

5. A system for identifying diagnostic biomarkers for autism, characterized in that, The system is used to implement the method according to any one of claims 1-4, the system comprising: The causal brain network construction module is configured to acquire the BOLD signal of the subject, process the BOLD signal using the PCMCI algorithm, calculate the causal relationship between each pair of brain regions, and construct a causal brain network for the subject. A significantly differential causal connection extraction module is configured to extract significantly differential causal connections from the causal brain network; The identification module is configured to input the significant difference causal connection into the support vector machine classification model to obtain the identification results of autism diagnostic biomarkers; The significant difference in causal connections was obtained by comparing the causal relationships between each pair of brain regions between the autism group and the normal control group using a two-sample t-test.

6. An electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to implement the method as described in any one of claims 1 to 4 when executing the computer program.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 4.

Images