A method for detecting attention deficit hyperactivity disorder based on BOLD signal

Abstract

The application provides a detection method for attention deficit hyperactivity disorder based on BOLD signals, the correlation features of the BOLD signals are extracted through a functional connection extraction network, then the time features of the BOLD signals are extracted through a time feature extraction network, the correlation features and the time features of the BOLD signals are fused, finally, a detection network is established based on a Softmax function, and the sample obtained through the feature fusion is input into the detection network to obtain a detection result; the application not only considers the fluctuation features of the BOLD signals in the time dimension, but also calculates the functional connections through the BOLD signals to further enhance the features and improve the classification accuracy, the multiple functional connections can be expressed at the same time through one output, the nonlinear features between the BOLD signals can be extracted, and the problems of too many features based on the traditional Pearson functional connection features and the linear correlation that can only be calculated are solved to a certain extent.

Description

Technical Field

[0001] This invention relates to the field of attention deficit hyperactivity disorder (ADHD) detection technology, and more specifically, to an ADHD detection method based on BOLD signals. Background Technology

[0002] Attention Deficit Hyperactivity Disorder (ADHD) is a common mental illness that predominantly affects adolescents. ADHD patients typically exhibit symptoms such as inattention, hyperactivity, and impulsivity, which can have many negative impacts on their lives. However, the pathogenesis of ADHD is not yet fully understood in modern medicine, and its clinical manifestations are not easily differentiated, thus accurate and efficient detection methods are currently lacking. Therefore, it is essential to explore an accurate and effective method for detecting ADHD using existing techniques.

[0003] Resting-state functional magnetic resonance imaging (rs-fMRI) focuses on reflecting brain activity in subjects over a period of time without any external stimulation. Each voxel or brain region in rs-fMRI contains a Bold Oxygen Level Dependent (BOLD) signal. Most ADHD detection methods that have emerged in recent years have focused on the time series of BOLD signals in fMRI. Generally speaking, existing detection methods use various techniques to extract features from BOLD signals and then use these features for classification. These methods can be divided into two types based on their emergence time: (1) methods based on traditional machine learning; and (2) methods based on deep learning.

[0004] However, most of the above methods only consider a single feature and have simple model structures. For example, some simply use recurrent neural networks (RNNs) to extract the fluctuation features of BOLD signals, while others simply use Pearson-related functional connectivity calculated from BOLD signals as features for classification. Current methods generally treat the temporal features and functional connectivity of BOLD signals as independent features, either analyzing only the temporal features of BOLD or only analyzing functional connectivity or correlations between brain regions. However, functional connectivity is calculated pairwise from BOLD signals from different brain regions, and it is a homologous feature of BOLD signals. Traditional Pearson-based functional connectivity features are too numerous and can only calculate linearly correlated features.

[0005] Based on this, this application proposes a method for detecting attention deficit and hyperactivity disorder based on BOLD signals to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide a method for detecting attention deficit and hyperactivity disorder based on BOLD signals, which can calculate functional connectivity through BOLD signals, enhance temporal characteristics, and improve detection accuracy.

[0007] The technical solution of this invention is as follows:

[0008] In a first aspect, this application provides a method for detecting attention deficit and hyperactivity disorder based on BOLD signals, which includes the following steps:

[0009] S1. Extract the correlation features of the BOLD signal by establishing a functional connectivity extraction network;

[0010] S2. Extract the temporal features of the BOLD signal using a temporal feature extraction network;

[0011] S3. Perform feature fusion on the correlation and temporal features of the BOLD signal;

[0012] S4. Establish a detection network based on the Softmax function, and input the samples obtained by feature fusion into the detection network to obtain the detection results.

[0013] Further, in step S1, the aforementioned functional connection extraction network includes a feature extraction network and a correlation calculation network; the aforementioned feature extraction network employs the Leaky ReLU activation function to extract features of the BOLD signal; the aforementioned correlation calculation network employs a self-attention mechanism to perform correlation calculation on the features of the BOLD signal extracted by the aforementioned feature extraction network to obtain correlation features.

[0014] Furthermore, the calculation process for the above correlation calculation includes:

[0015] a i =Wx i q i =W q a i k i =W k a i v i =W v a i ,

[0016]

[0017] score i,j =Softmax(β) i,j ),

[0018]

[0019] Where, x i Represents the input features, x j Let W and W' represent the input features of part j. q W k and W v Both are parameter matrices, a i x represents i The embedding result, q i Indicates the content to be queried, k i Indicates with q i Matching index, v i Indicates the query result, β i,j x represents i With x j The correlation between them q i transpose, k j Indicates with q j Matching index, score i,j Indicates β i,j The mapping, v j x represents j The linear transformation characteristics of f i Represents each part x in the input features j The weighted sum.

[0020] Further, in step S2, the aforementioned time feature extraction network comprises three one-dimensional convolutional kernels and a GRU. The BOLD signal passes through the three one-dimensional convolutional kernels, and then the GRU extracts the time features of the BOLD signal. The calculation process for the GRU to extract the time features of the BOLD signal includes:

[0021] r t =σ(W r B t +U r h t-1 +b r ),

[0022]

[0023] z t =σ(W z B t +W z h t-1 +b z ),

[0024]

[0025] Where, rt This represents the reset gate, σ represents the sigmoid function, and W... r Indicates r t The corresponding parameter matrix, B represents the time series of the BOLD signal, B t U represents the BOLD signal value at the current time point. r Let h represent the hidden state of the GRU. t-1 This represents the previous hidden state of the GRU, b r Indicates r t The corresponding paranoia. U represents the candidate state. h h t The corresponding parameter matrix, ⊙ represents the Hadamard product, b h h t The corresponding paranoia quantity, z t Indicates an update gate, W z Indicate z t The corresponding parameter matrix, b z Indicate z t The corresponding paranoia, o t This indicates the final time characteristic.

[0026] Furthermore, in step S4, the expression for establishing the detection network based on the Softmax function is as follows:

[0027]

[0028] Among them, z i Let z represent the output of the i-th node, C represent the number of output nodes of the neuron, and z k This represents the output of the k-th node.

[0029] Secondly, this application provides an electronic device, characterized in that it comprises:

[0030] Memory, used to store one or more programs;

[0031] processor;

[0032] When one or more of the above programs are executed by the above processor, a method for detecting attention deficit and hyperactivity disorder based on BOLD signals, as described in any of the first aspects above, is implemented.

[0033] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for detecting attention deficit and hyperactivity disorder based on BOLD signals as described in any of the first aspects above.

[0034] Compared with the prior art, the present invention has at least the following advantages or beneficial effects:

[0035] (1) The present invention provides a method for detecting attention deficit and hyperactivity disorder based on BOLD signal, which not only considers the fluctuation characteristics of BOLD signal in the time dimension, but also calculates functional connectivity through BOLD signal to further enhance features and improve classification accuracy.

[0036] (2) By calculating the correlation between BOLD signals in brain regions, this invention can not only express multiple functional connections through one output, but also extract nonlinear features between BOLD signals, which to some extent solves the problems of excessive Pearson functional connection features and the inability to calculate linear correlations in traditional Pearson-based systems. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart illustrating the steps of a method for detecting attention deficit and hyperactivity disorder based on BOLD signals according to the present invention.

[0039] Figure 2 A schematic diagram of the functional connectivity extraction network;

[0040] Figure 3 This is a diagram illustrating the principle of correlation calculation.

[0041] Figure 4 This is a schematic diagram of the structure of a time feature extraction network;

[0042] Figure 5 This is a schematic structural block diagram of an electronic device according to an embodiment of the present invention.

[0043] Icons: 101, memory; 102, processor; 103, communication interface. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0045] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0046] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0047] It should be noted that, in this document, the term "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0048] In the description of this application, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0049] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the various embodiments and features described below can be combined with each other.

[0050] Example 1

[0051] Please see Figure 1 , Figure 1 The figure shown is a flowchart of a method for detecting attention deficit and hyperactivity disorder based on BOLD signals provided in Embodiment 1 of this application.

[0052] In a first aspect, this application provides a method for detecting attention deficit and hyperactivity disorder based on BOLD signals, which includes the following steps:

[0053] S1. Extract the correlation features of the BOLD signal by establishing a functional connectivity extraction network;

[0054] S2. Extract the temporal features of the BOLD signal using a temporal feature extraction network;

[0055] S3. Perform feature fusion on the correlation and temporal features of the BOLD signal;

[0056] S4. Establish a detection network based on the Softmax function, and input the samples obtained by feature fusion into the detection network to obtain the detection results.

[0057] Functional connectivity essentially represents the linear correlation between two BOLD signals, quantified by normalizing the correlation to [-1, 1]. Deep learning excels at linear representation and can fit arbitrary nonlinear functions by introducing nonlinear activation functions. Therefore, deep learning can be considered for calculating the correlation between two BOLD signals. Through deep learning, the correlation between BOLD signals is no longer limited to linearity but can also learn nonlinear features. Based on the above analysis, this invention proposes a deep learning-based functional connectivity extraction network, the structure of which can be found in [link to network diagram]. Figure 2 , Figure 2 The diagram shows the structure of the functional connectivity extraction network.

[0058] In a preferred embodiment, in step S1, the functional connectivity extraction network includes a feature extraction network and a correlation calculation network; the feature extraction network uses the Leaky ReLU activation function to extract features of the BOLD signal; the correlation calculation network uses a self-attention mechanism to perform correlation calculation on the features of the BOLD signal extracted by the feature extraction network to obtain correlation features.

[0059] The functional connectivity extraction network mainly consists of a feature extraction network and a correlation calculation network. Each BOLD signal first passes through the feature extraction network to further extract fluctuation features. All features obtained from the feature extraction network are then fed into the correlation calculation network to calculate correlations. The feature extraction network consists of three layers, each containing 1D convolutions, batch normalization, and activation functions. The difference is that the first layer includes a max pooling operation with a pooling window size of 2 at the end. The 1D convolution kernel size is 3, with 8, 16, and 8 kernels used in each layer, respectively. The purpose of batch normalization is similar to dividing by the standard deviation when calculating functional connectivity—to normalize the data and eliminate the influence of dimensions on correlations. Leaky ReLU is used for all activation functions. The correlation calculation network mainly relies on the self-attention mechanism. The attention mechanism is a mechanism that aligns internal experience with external sensations to increase the precision of observation in certain areas. The self-attention mechanism is an improvement on the attention mechanism; it reduces the influence of external data and focuses more on the correlations within features. The feature extraction network abstracted the fluctuation features of BOLD signals from 90 brain regions, and the self-attention mechanism can be used to explore the correlations of these fluctuations. Self-attention mechanisms can consider the correlations between different parts of the input. The relationship between a certain part of the input and other parts is represented by the weighted sum of the remaining parts. Strong correlation is achieved by assigning larger weights to parts with strong correlations, and vice versa by assigning lower weights to parts with weak correlations.

[0060] Taking the calculation of functional connectivity among BOLD signals in 90 brain regions as an example, please refer to the working principle of the correlation calculation network. Figure 3 , Figure 3 The diagram shows the principle of correlation calculation. Only the calculation process of f1 is shown in the diagram. This represents the Softmax function. This represents the dot product operation. This represents vector addition. Analyze the working principle of the correlation calculation network, with input x. i This can be understood as the feature extraction network extracting features from the BOLD signal, and outputting f. i This can be understood as representing the characteristics of functional connectivity. Input x i First, it will undergo an embedding operation to obtain a. i Immediately afterwards, a i By using the parameter matrix W q W k and W v Multiply to get q i k i and v i .

[0061] As a preferred implementation, the correlation calculation process includes:

[0062] a i =Wx i q i =W q a i k i =W k a i v i =W v a i ,

[0063] Get a i The corresponding q i k i and v i Then you need to use q i Each with each k j Matching (j = 1, 2, 3, ..., 90) to calculate the remaining parts of the input for x i The impact on the score. First, q i It will be with all k respectively j Perform matching and dot product operation to obtain β. i,j In fact, it represents x i With x j The degree of correlation between them:

[0064]

[0065] Where, q i In relation to k j When performing matching, you can choose to calculate or not calculate the correlation with yourself; that is, j can be equal to i or not equal to i. When calculating functional connections, the correlation with itself is not calculated. Figure 3 In the calculation of f1, q1 and k1 matching is not used to calculate β. 1,1 Immediately afterwards, β i,j The value is mapped to the range (0,1) using the Softmax function, and is denoted as score. i,j , representing the j-th part x in the input j For x i The impact on scores, and all scores i,j satisfy The above process can be expressed by the following formula:

[0066] score i,j =Softmax(β) i,j ),

[0067] At this point, the score can be used to... i,j To calculate f i f iThis is represented as the weighted sum of the input components, where the weights are the previously calculated scores. i,j The specific calculation formula is as follows:

[0068]

[0069] Where, x i Represents the input features, x j Let W and W' represent the input features of part j. q W k and W v Both are parameter matrices, a i x represents i The embedding result, q i Indicates the content to be queried, k i Indicates with q i Matching index, v i Indicates the query result, β i,j x represents i With x j The correlation between them q i transpose, k j Indicates with q j Matching index, score i,j Indicates β i,j The mapping, v j x represents j The linear transformation characteristics of f i Represents each part x in the input features j The weighted sum.

[0070] It should be noted that the output f i In fact, it calculates x in the original input. i Other parts x j The correlation. Specifically, f i It is derived from the weighted sum of the input parts, where the parts with higher weights have a certain relationship with f. i The impact is significant, corresponding to the original input x i Strong correlation; conversely, the low-weighted portion is more relevant to f. i The influence of the earth is relatively small, corresponding to the original input x i The correlation is weak.

[0071] In a preferred embodiment, in step S2, the time feature extraction network consists of three one-dimensional convolutional kernels and a GRU. After the BOLD signal passes through the three one-dimensional convolutional kernels, the GRU extracts the time features of the BOLD signal.

[0072] The temporal feature extraction network is mainly used to analyze the characteristics of the BOLD signal fluctuation over time in 90 brain regions in the Automated Anatomical Labeling (AAL) brain map. It is mainly composed of 1D-CNN and GRU. Figure 4 The specific structure of the temporal feature extraction network is given.

[0073] like Figure 4 The diagram shows the structure of the temporal feature extraction network. The three one-dimensional convolutional kernels have sizes of 2, 4, and 8, respectively. The time interval between each value in the BOLD signal is the time required to scan the entire brain, generally called the time of repetition (TR). Therefore, the three convolutional kernels correspond to 2TR, 4TR, and 8TR, respectively. Furthermore, because the convolutional feature maps will be stitched together later, to facilitate stitching, the multi-scale convolutional layers choose to keep the sequence length after 4TR and 8TR convolutions consistent with the sequence length after 2TR convolutions. Therefore, padding is used during 4TR and 8TR convolutions. Padding calculation is relatively simple. Assuming the convolutional kernel size is f, the padding size that ensures the sequence length before and after convolution is the original sequence length minus 1 (i.e., the sequence length after 2TR convolution) is:

[0074] In the current context, the dynamic features extracted by GRU can be understood as long-term dependency features of the BOLD signal time series fluctuations, such as periodicity and trend features. Specifically, the BOLD signal in certain brain regions of ADHD patients may fluctuate with a trend of first rising and then falling over a period of time, and this pattern of fluctuation will repeat within a certain period; or the BOLD signal of ADHD patients may generally show an upward or downward trend compared to normal individuals. GRU can effectively control the transmission and retention of information through its update and reset gates, thereby capturing long-term dependencies in the time series. Assuming B represents the BOLD signal time series with a length of n, when the BOLD signal value B at each time point... t When passing through the GRU, the gate r is reset. t It will be based on the BOLD signal value B at the current time point. t And the previously hidden state h t-1 Calculate a value between (0, 1) that controls the amount of past information forgotten:

[0075] r t =σ(W r B t +U r h t-1 +b r ),

[0076] According to the reset gate r t The BOLD signal value B at the current time point t And the previously hidden state h t-1 A candidate state can be calculated.

[0077]

[0078] Update Gate Z t Similar to the reset gate, it combines the current BOLD signal value with the previous hidden state to generate a value between (0, 1), which determines how much past information to forget and how much current information to retain:

[0079] z t =σ(W z B t +W z h t-1 +b z ),

[0080] Therefore, the output o at the current time can be calculated. t :

[0081]

[0082] Where, r t This represents the reset gate, σ represents the sigmoid function, and W... r Indicates r t The corresponding parameter matrix, B represents the time series of the BOLD signal, B t U represents the BOLD signal value at the current time point. r Let h represent the hidden state of the GRU. t-1 This represents the previous hidden state of the GRU, b r Indicates r t The corresponding paranoia. U represents the candidate state. h h t The corresponding parameter matrix, ⊙ represents the Hadamard product, b h h t The corresponding paranoia quantity, z t Indicates an update gate, W z Indicate z t The corresponding parameter matrix, b z Indicate z t The corresponding paranoia, o t This indicates the final time characteristic.

[0083] Therefore, through the above-mentioned gating mechanism, the discarding and transmission of information can be dynamically controlled, thereby extracting long-term dependent features such as periodicity and trend from sequence data.

[0084] In a preferred implementation, in step S4, the expression for establishing the detection network based on the Softmax function is as follows:

[0085]

[0086] Among them, z i Let z represent the output of the i-th node, C represent the number of output nodes of the neuron, and z k This represents the output of the k-th node.

[0087] It should be noted that the detection network mainly consists of three fully connected layers and a softmax function. The softmax function is commonly used in multi-class classification tasks. It maps multiple values ​​to the interval (0,1), and the sum of these multiple values ​​is 1. In the context of multi-class classification tasks, these multiple values ​​represent the probability value of each class.

[0088] Example 2

[0089] Please see Figure 5 , Figure 5 The diagram shown is a schematic structural block diagram of an electronic device provided in Embodiment 2 of this application.

[0090] An electronic device includes a memory 101, a processor 102, and a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules. The processor 102 executes the software programs and modules stored in the memory 101 to perform various functional applications and data processing. The communication interface 103 can be used for signaling or data communication with other node devices.

[0091] The memory 101 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0092] The processor 102 can be an integrated circuit chip with signal processing capabilities. The processor 102 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0093] It is understood that the structure shown in the figure is for illustrative purposes only. An attention deficit and hyperactivity disorder detection method based on BOLD signals may include more or fewer components than shown in the figure, or have a different configuration. The components shown in the figure can be implemented in hardware, software, or a combination thereof.

[0094] In the embodiments provided in this application, it should be understood that the disclosed methods can also be implemented in other ways. The embodiments described above are merely illustrative. For example, the flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, each block in the flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0095] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0096] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0097] In summary, the attention deficit and hyperactivity disorder (ADHD) detection method based on BOLD signals provided in this application not only considers the fluctuation characteristics of BOLD signals in the time dimension, but also calculates functional connectivity through BOLD signals to further enhance features and improve classification accuracy. Unlike traditional Pearson-based functional connectivity, this invention does not employ a novel deep learning-based functional connectivity extraction network to calculate the correlation between BOLD signals in brain regions. This can be understood as a "functional connectivity-like" network, called deep functional connectivity. Deep functional connectivity can not only express multiple functional connections through a single output, but also extract nonlinear features between BOLD signals, thus addressing to some extent the limitations of traditional Pearson-based functional connectivity methods, which suffer from excessive features and can only calculate linear correlations.

[0098] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0099] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for detecting attention deficit hyperactivity disorder based on BOLD signal, characterized in that, The method comprises the following steps: S1, extracting the correlation feature of the BOLD signal through a functional connection extraction network; In step S1, the functional connection extraction network comprises a feature extraction network and a correlation calculation network; the feature extraction network comprises a Leaky ReLU activation function, and is used to extract the feature of the BOLD signal; the correlation calculation network adopts a self-attention mechanism, and is used to calculate the correlation of the feature of the BOLD signal extracted by the feature extraction network to obtain a correlation feature; The calculation process of the correlation calculation comprises: ， wherein, represents an input feature, represents an input feature of the j-th part, and are parameter matrices, represents an embedding result of represents a content to be queried, represents an index matched with represents a query result, represents a relevance between and represents a transpose of represents an index matched with represents a mapping of represents a linear transformation feature of represents a weighted sum of each part of an input feature;​​​​​​​​​ S2, extracting the time feature of the BOLD signal through a time feature extraction network; In step S2, the time feature extraction network comprises three one-dimensional convolution kernels and a GRU, and the BOLD signal passes through the three one-dimensional convolution kernels and then the GRU extracts the time feature of the BOLD signal; wherein the calculation process of the GRU for extracting the time feature of the BOLD signal comprises: wherein, denotes a reset gate, denotes a sigmoid function, denotes the corresponding parameter matrix of denotes a BOLD signal time series, denotes the BOLD signal value at the current time point, denotes, denotes the hidden state of the GRU, denotes the previous hidden state of the GRU, denotes the corresponding bias of denotes a candidate state, denotes the corresponding parameter matrix of denotes the Hadamard product, denotes the corresponding bias of denotes an update gate, denotes the corresponding parameter matrix of denotes the corresponding bias of denotes the final temporal feature; S3, fusing the correlation feature and the time feature of the BOLD signal; S4, establishing a detection network based on a Softmax function, and inputting the sample fused by the features into the detection network to obtain a detection result.

2. The method for detecting attention deficit hyperactivity disorder based on BOLD signal according to claim 1, wherein, In step S4, the expression of the detection network established based on the Softmax function is: wherein, represents the output of the i-th node, represents the number of output nodes of the neuron, represents the output of the k-th node.

3. An electronic device, comprising: comprise: a memory for storing one or more programs; a processor; when the one or more programs are executed by the processor, a BOLD signal-based attention deficit hyperactivity disorder detection method according to any one of claims 1-2 is implemented.

4. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement a BOLD signal-based attention deficit hyperactivity disorder detection method according to any one of claims 1-2.

Images