Autism risk assessment method and device based on small sample data enhancement

Abstract

The present application relates to an autism risk assessment method and device based on small sample data enhancement. The method comprises: collecting a multi-channel electroencephalogram signal of a subject; calculating the multi-channel electroencephalogram signal using a Pearson correlation coefficient to obtain the correlation between channels, generating a brain functional connectivity matrix, and constructing a brain time sequence connected graph according to time sequence information of the multi-channel electroencephalogram signal; and performing data enhancement processing on the brain time sequence connected graph using a generative adversarial network, inputting the brain time sequence connected graph after data enhancement processing into a trained deep convolutional network and long short-term memory network joint model for classification, and outputting an autism risk assessment result of the subject by a classifier. According to the embodiments of the present application, the differences between patients with autism and healthy people are taken into account from the perspective of brain functional connectivity, and the brain functional connectivity and the time information contained in an electroencephalogram signal are both taken into account, so that autism risk assessment can be more accurately performed on a subject.

Description

Autism risk assessment method and device based on small sample data enhancement Technical Field

[0001] The present application belongs to the field of deep learning technology, and in particular relates to an autism risk assessment method and device based on small sample data enhancement. Background Art

[0002] Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by significant differences in brain functional connectivity compared to healthy individuals. Autism can lead to persistent deficits in social communication skills and behavior. As of 2020, the prevalence of autism among children aged 6-12 in China was approximately 0.70%, equivalent to one in every 143 children. People with ASD impose a significant economic burden on families and society. Traditional clinical diagnostic protocols for autism rely on observational scales and behavioral assessments, requiring a high level of expertise from physicians and resulting in low diagnostic efficiency.

[0003] Autism risk assessment can be achieved by building autism recognition models. Currently, many researchers are leveraging EEG signals with machine learning and deep learning methods to develop such models. For example, one domestic study extracted four entropy features from EEG (electroencephalogram) signals of children with ASD (Autistic Spectrum Disorders) and normal children, and analyzed group differences using independent sample t-tests. Finally, a support vector machine learning algorithm was used to develop classification models for each entropy measure across different brain regions, achieving a classification accuracy of up to 84.55%. Another international study used a discrete Fourier transform to convert EEG time-domain signals into frequency-domain signals. A reinforcement learning algorithm was then designed to rapidly update and reconstruct a convolutional neural network model. The resulting convolutional neural network model was then used for classification, achieving a recognition accuracy of up to 92.63%.

[0004] However, existing EEG-based autism detection methods generally utilize the raw EEG signal or spectral features. This is limited by the number of subjects, and the number of experimental subjects is generally small. Existing technologies often reuse data from different time segments of a subject as independent samples, using only a portion of the EEG data for analysis at a time. This can miss important EEG information and cause data contamination, leading to distorted analysis results. Alternatively, data from a single subject can be reused to create multiple samples, resulting in excessive similarity between samples. Furthermore, few existing studies have used EEG data to statistically analyze the functional connectivity characteristics of the subjects' brains and autism.

[0005] Summary of the Invention

[0006] The present application provides a method and device for autism risk assessment based on small sample data enhancement, aiming to solve at least one of the above-mentioned technical problems in the prior art to a certain extent.

[0007] In order to solve the above problems, this application provides the following technical solutions:

[0008] A method for autism risk assessment based on small sample data enhancement, comprising:

[0009] Collecting multi-channel EEG signals of the subjects;

[0010] Calculating the multi-channel EEG signals using the Pearson correlation coefficient to obtain the correlation between the channels, generating a brain functional connectivity matrix, and constructing a brain temporal connectivity map according to the temporal information of the multi-channel EEG signals;

[0011] The brain temporal connectivity map is subjected to data enhancement processing using a generative adversarial network, and the data-enhanced brain temporal connectivity map is input into a trained deep convolutional network and long short-term memory network joint model for classification, and the classifier outputs the subject's autism risk assessment result.

[0012] The technical solution adopted in the embodiment of the present application further includes: after collecting the multi-channel EEG signals of the subject, the following steps are also included:

[0013] Preprocessing the multi-channel EEG signals; the preprocessing process includes: filtering the multi-channel EEG signals using an anti-aliasing filter, digitizing and amplifying all EEG signals at a set sampling rate;

[0014] The amplified multi-channel EEG signal is corrected using a baseline correction algorithm, and the multi-channel EEG signal is artifact-removed using an artifact detection algorithm to obtain a preprocessed multi-channel EEG signal.

[0015] The technical solution adopted in the embodiment of the present application also includes: using the Pearson correlation coefficient to calculate the multi-channel EEG signals to obtain the correlation between each channel, generate a brain functional connectivity matrix, and construct a brain temporal connectivity map according to the timing information of the multi-channel EEG signals, specifically:

[0016] The correlation between each channel and other channels in a subject's EEG segment is calculated by the Pearson correlation coefficient to obtain a functional connectivity matrix of a set size;

[0017] Expanding the upper triangular matrix of the functional connectivity matrix into one dimension, and using the SelectKBest algorithm as a feature screening algorithm to screen the functional connectivity features, retaining a set number of main functional connectivity features;

[0018] The EEG signals of a set number of time segments collected from a subject are organized into a brain functional connectivity matrix of a set size in chronological order, and the brain functional connectivity matrix is ​​converted into a brain temporal connectivity map according to the gray value distribution.

[0019] The technical solution adopted in the embodiment of the present application further includes: after constructing the brain temporal connectivity map according to the temporal information of the multi-channel EEG signals, it also includes:

[0020] The brain temporal connectivity map is divided into a training set and a validation set according to a set ratio.

[0021] The technical solution adopted in the embodiment of the present application also includes: using a generative adversarial network to perform data enhancement processing on the brain temporal connectivity map, specifically:

[0022] The generative adversarial network includes a generator and a discriminator. The generator includes three deconvolution layers, which map the noise vector to the brain temporal connectivity map through the deconvolution layers to generate a new brain temporal connectivity map; the discriminator uses three convolution layers to determine the authenticity of the generated brain temporal connectivity map, and after the generator and the discriminator iterate and confront a set number of times, the brain temporal connectivity map generated by the generator is used as an expanded pseudo sample.

[0023] The technical solution adopted in the embodiment of the present application also includes: inputting the brain temporal connectivity map after data enhancement processing into a trained deep convolutional network and long short-term memory network joint model for classification, and the classifier outputs the subject's autism risk assessment result, specifically:

[0024] The deep convolutional network part of the joint model of deep convolutional network and long short-term memory network includes 5 convolutional layers and 4 pooling layers. The local features of the input data are learned through the convolutional layer, and the convolutional layer is activated by the ReLU function; the long short-term memory network includes a hidden layer, a recurrent network layer and a fully connected layer. After the input dimension is converted, the input data passes through the hidden layer shaping and two recurrent network layers in sequence, and then passes through two fully connected layers, and finally the classifier is connected to output a binary classification result. The binary classification result includes two categories: "autism" and "healthy".

[0025] Another technical solution adopted in the embodiment of the present application is: an autism risk assessment device based on small sample data enhancement, comprising:

[0026] Signal acquisition module: used to collect multi-channel EEG signals of subjects;

[0027] Connectivity map construction module: used to calculate the multi-channel EEG signals using the Pearson correlation coefficient to obtain the correlation between each channel, generate a brain functional connectivity matrix, and construct a brain temporal connectivity map according to the temporal information of the multi-channel EEG signals;

[0028] Data enhancement module: used for performing data enhancement processing on the brain temporal connectivity map using a generative adversarial network;

[0029] Risk assessment module: used to input the brain temporal connectivity map after data enhancement processing into the trained deep convolutional network and long short-term memory network joint model for classification, and the classifier outputs the subject's autism risk assessment result.

[0030] The technical solution adopted in the embodiment of the present application also includes: the connectivity map construction module calculates the multi-channel EEG signals using the Pearson correlation coefficient to obtain the correlation between each channel, generates a brain functional connectivity matrix, and constructs a brain temporal connectivity map according to the timing information of the multi-channel EEG signals, specifically:

[0031] The correlation between each channel and other channels in a subject's EEG segment is calculated by the Pearson correlation coefficient to obtain a functional connectivity matrix of a set size;

[0032] Expanding the upper triangular matrix of the functional connectivity matrix into one dimension, and using the SelectKBest algorithm as a feature screening algorithm to screen the functional connectivity features, retaining a set number of main functional connectivity features;

[0033] The EEG signals of a set number of time segments collected from a subject are organized into a brain functional connectivity matrix of a set size in chronological order, and the brain functional connectivity matrix is ​​converted into a brain temporal connectivity map according to the gray value distribution.

[0034] Another technical solution adopted by the embodiment of the present application is: a computer device, the computer device includes a processor and a memory coupled to the processor, wherein:

[0035] The memory stores program instructions for implementing the autism risk assessment method based on small sample data enhancement;

[0036] The processor is configured to execute the program instructions stored in the memory to control an autism risk assessment method based on small sample data enhancement.

[0037] Another technical solution adopted in the embodiment of the present application is: a storage medium storing program instructions executable by a processor, wherein the program instructions are used to execute the autism risk assessment method based on small sample data enhancement.

[0038] Compared with the prior art, the beneficial effect produced by the embodiment of the present application is that the autism risk assessment method and device based on small sample data enhancement of the embodiment of the present application calculates the brain functional connection matrix using EEG signals, and constructs a brain temporal connectivity map based on the time information of EEG signals. After the brain temporal connectivity map is expanded using a generative adversarial network, it is input into a deep convolutional network and a long short-term memory network joint model for classification, and the classifier outputs the autism risk assessment result. The embodiment of the present application considers the differences between autistic patients and healthy people from the perspective of brain functional connectivity, and simultaneously considers the time information of brain functional connectivity and EEG signals, and can more accurately assess the risk of autism for the subjects. Taking the EEG signals of each subject's overall time as a sample is also more in line with the actual diagnostic scenario. By designing a generative adversarial network model for sample expansion, the problem of generally small sample size is effectively solved. By constructing a deep convolutional network and a long short-term memory network joint model for autism risk assessment, it is possible to consider more comprehensive EEG signal features and further improve the accuracy of autism risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] FIG1 is a flow chart of an autism risk assessment method based on small sample data enhancement according to an embodiment of the present application;

[0040] FIG2 is a brain temporal connectivity diagram constructed according to an embodiment of the present application;

[0041] FIG3 is a schematic diagram of a model structure of a generative adversarial network according to an embodiment of the present application;

[0042] FIG4 is a schematic diagram of the structure of a joint model of a deep convolutional network and a long short-term memory network according to an embodiment of the present application;

[0043] FIG5 is a schematic diagram of the structure of an autism risk assessment device based on small sample data enhancement according to an embodiment of the present application;

[0044] FIG6 is a schematic diagram of the structure of a computer device according to an embodiment of the present application;

[0045] FIG7 is a schematic diagram of the structure of a storage medium according to an embodiment of the present application. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0047] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of such features. In the description of this application, "multiple" means at least two, for example, two, three, etc., unless otherwise specifically defined. All directional indications in the embodiments of this application (such as up, down, left, right, front, back...) are only used to explain the relative positional relationship, movement, etc. between the components under a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indications also change accordingly. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or computer device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products, or computer devices.

[0048] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0049] Specifically, please refer to Figure 1, which is a flow chart of the autism risk assessment method based on small sample data enhancement according to an embodiment of the present application. The autism risk assessment method based on small sample data enhancement according to an embodiment of the present application comprises the following steps:

[0050] S100: Collecting multi-channel EEG signals of the subjects;

[0051] In this step, the collected multi-channel EEG signal has 125 channels, and the specific number of channels can be set according to the actual application scenario.

[0052] S110: preprocessing multi-channel EEG signals;

[0053] In this step, since noise is inevitably introduced during the EEG signal acquisition process, it is necessary to eliminate noise such as eye contact signals in the preprocessing stage. In an embodiment of the present application, the preprocessing process of the multi-channel EEG signal includes: filtering the multi-channel EEG signal with an anti-aliasing filter, digitizing and amplifying all EEG signals at a set sampling rate. Finally, the amplified multi-channel EEG signal is corrected using a baseline correction algorithm, and the multi-channel EEG signal is artifact-removed using an artifact detection algorithm to obtain a preprocessed multi-channel EEG signal.

[0054] S120: Calculate the preprocessed multi-channel EEG signals using the Pearson product-moment correlation coefficient to obtain the correlation between each channel, generate a brain functional connectivity matrix, and construct a brain temporal connectivity map in time sequence;

[0055] In this step, the correlation between each channel and other channels in an EEG segment of the subject is calculated by the Pearson correlation coefficient, thereby obtaining a 125×125 functional connectivity matrix. The upper triangular matrix of the functional connectivity matrix is ​​expanded into one dimension, and then the SelectKBest algorithm is used as a feature screening algorithm to screen the functional connectivity features, retaining a set number of main functional connectivity features. Assuming that the main functional connectivity features retained are 400, the EEG signals of 40 time segments collected by a subject are organized into a 400×40 brain functional connectivity matrix in chronological order, and then the brain functional connectivity matrix is ​​converted into a brain temporal connectivity map according to the grayscale value distribution, and all brain temporal connectivity maps are divided into training sets and verification sets according to a set ratio. In the embodiment of the present application, the division ratio of the training set is 80%, and the division ratio of the verification set is 20%. The specific setting can be made according to the actual application scenario. Specifically, as shown in FIG2 , a brain temporal connectivity map constructed in an embodiment of the present application is shown, wherein (a) is a brain temporal connectivity map of autistic patients, and (b) is a brain temporal connectivity map of a healthy control group. As can be seen from FIG2 , there are clear differences between the brain temporal connectivity maps of autistic patients and the healthy control group. It is understood that in other embodiments of the present application, the phase locking value can also be used as a connection indicator for constructing a brain temporal connectivity map.

[0056] S130: Using a generative adversarial network (DCGAN) to perform data augmentation on the brain temporal connectivity graph in the training set to generate an expanded training set;

[0057] In this step, due to the general lack of autism subject sample size, the present application embodiment utilizes pytorch to design a generative adversarial network (DCGAN), and the generative adversarial network is used as a data enhancement method to expand the sample size of the brain temporal connectivity map. Specifically, the model structure of the generative adversarial network of the present application embodiment is shown in Figure 3, and the generative adversarial network includes two parts, a generator and a discriminator. The generator part is composed of 3 layers of deconvolution layers, and 100x1 noise vectors are mapped to the brain temporal connectivity map of 400x40x1 by deconvolution, generating a new brain temporal connectivity map; The discriminator part uses 3 layers of convolution layers to distinguish the authenticity of the generated brain temporal connectivity map. After generator and discriminator iterative confrontation 100 times, the brain temporal connectivity map generated by the generator is used as an expanded pseudo sample. In the present application embodiment, data enhancement processing is performed only for the training set, and the result of the verification set classification experiment will not be affected.

[0058] S140: Input the expanded training set into the constructed deep convolutional network and long short-term memory network joint model (CNN-LSTM) for classification, and the classifier outputs the subject's autism risk assessment result;

[0059] In this step, a deep convolutional network (DCNN) and long short-term memory (LSTM) joint model was constructed using PyTorch. The structure of the DCNN / LSTM joint model is shown in Figure 4. The DCNN consists of five convolutional layers and four pooling layers. Local features of the input data are learned through 3x3 and 2x2 convolutional layers, activated by Reluctant Unit (ReLU) functions. The hidden layer dimension of the LSTM network was set to 64, the number of recurrent layers was set to 2, and the number of fully connected layers was set to 2. Because the brain temporal connectivity map constructed is a grayscale image, the initial input data size is 400×40×1. After input dimensionality conversion, the image size is reduced to 400×1×1. Subsequently, the image size is reduced to 64×1×1 after hidden layer reshaping and two recurrent layers. Finally, the image is connected to a classifier to output the final binary classification result after two fully connected layers. The DCNN / LSTM joint model was experimentally validated using a 5-fold cross-validation technique, resulting in the optimal DCNN / LSTM joint model. The binary classification results include two categories: "autism" and "healthy". If the classification result is "autism", it means that the subject has a high risk of autism and needs to undergo a more professional diagnosis in the next step. It can be understood that the embodiment of the present application only takes the autism risk assessment based on small sample data enhancement as an example. The embodiment of the present application is also applicable to the assessment and classification of other medical images with smaller sample sizes and images with spatiotemporal characteristics.

[0060] In order to verify the feasibility and effectiveness of the embodiment of the present application, the recognition accuracy of the embodiment of the present application was evaluated through experiments. The present invention proved that the method is feasible through experiments. The recognition accuracy achieved by testing data from autistic patients and healthy controls is shown in Table 1 below:

[0061] Table 1 Recognition accuracy of autistic patients and healthy control group data

[0062] Experimental results show that the embodiments of the present application can quickly and accurately assess the risk of autism in subjects. In the resting-state EEG signal dataset, the recognition accuracy obtained by testing using the joint model of the deep convolutional network and long short-term memory network of the embodiments of the present application can reach up to 81.08%.

[0063] Based on the above, the autism risk assessment method based on small sample data enhancement of the embodiment of the present application utilizes EEG signals to calculate brain function connection matrix, and builds brain temporal connectivity map according to the time information of EEG signals, after using generative adversarial network (DCGAN) to carry out sample expansion to brain temporal connectivity map, it is input to deep convolutional network and long short-term memory network joint model (CNN-LSTM) for classification, and outputs autism risk assessment result by classifier. The embodiment of the present application considers the difference between autistic patients and healthy people from the perspective of brain function connectivity, while considering the time information possessed by brain function connectivity and EEG signals, and can more accurately carry out autism risk assessment to experimenter. The EEG signals of each experimenter's overall time are used as a sample and are more in line with actual diagnosis scene. Sample expansion is carried out by designing generative adversarial network model, effectively solving the problem that sample size is generally less. Autism risk assessment is carried out by constructing deep convolutional network and long short-term memory network joint model, more comprehensive EEG signal features can be considered, further improving the precision of autism risk assessment.

[0064] Please refer to FIG5 , which is a schematic diagram of the structure of an autism risk assessment device based on small sample data enhancement according to an embodiment of the present application. The autism risk assessment device 40 based on small sample data enhancement according to an embodiment of the present application comprises:

[0065] Signal acquisition module 41: used to collect multi-channel EEG signals of the subject;

[0066] Connectivity map construction module 42: used to calculate the multi-channel EEG signals using the Pearson correlation coefficient to obtain the correlation between each channel, generate a brain functional connectivity matrix, and construct a brain temporal connectivity map according to the temporal information of the multi-channel EEG signals;

[0067] Data enhancement module 43: configured to perform data enhancement processing on the brain temporal connectivity map using a generative adversarial network;

[0068] Risk assessment module 44: used to use a generative adversarial network to perform data enhancement processing on the brain temporal connectivity map, and input the brain temporal connectivity map after data enhancement processing into a trained deep convolutional network and long short-term memory network joint model for classification, and the classifier outputs the subject's autism risk assessment result.

[0069] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0070] The device provided in the embodiment of the present application can be applied in the aforementioned method embodiment. For details, please refer to the description of the aforementioned method embodiment, which will not be repeated here.

[0071] Please refer to FIG6 , which is a schematic diagram of the computer device structure of an embodiment of the present application. The computer device 50 includes:

[0072] A memory 51 storing executable program instructions;

[0073] a processor 52 connected to the memory 51;

[0074] The processor 52 is used to call the executable program instructions stored in the memory 51 and perform the following steps: collecting multi-channel EEG signals of the subject; using the Pearson correlation coefficient to calculate the multi-channel EEG signals to obtain the correlation between each channel, generating a brain functional connection matrix, and constructing a brain temporal connectivity map according to the timing information of the multi-channel EEG signals; using a generative adversarial network to perform data enhancement processing on the brain temporal connectivity map, and inputting the brain temporal connectivity map after data enhancement processing into a trained deep convolutional network and long short-term memory network joint model for classification, and the classifier outputs the subject's autism risk assessment result.

[0075] The processor 52 may also be referred to as a CPU (Central Processing Unit). The processor 52 may be an integrated circuit chip having signal processing capabilities. The processor 52 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The general-purpose processor may be a microprocessor or any conventional processor.

[0076] Refer to Figure 7, Figure 7 is a structural diagram of the storage medium of the embodiment of the present application. The storage medium of the embodiment of the present application stores a program instruction 61 that can implement the following steps: collect the multi-channel EEG signal of the subject; calculate the multi-channel EEG signal using the Pearson correlation coefficient, obtain the correlation between each channel, generate a brain functional connection matrix, and build a brain temporal connectivity map according to the timing information of the multi-channel EEG signal; use a generative adversarial network to perform data enhancement processing on the brain temporal connectivity map, and input the brain temporal connectivity map after the data enhancement processing into the deep convolutional network and long short-term memory network joint model trained for classification, and output the autism risk assessment result of the subject by the classifier. Wherein, the program instruction 61 can be stored in the above-mentioned storage medium in the form of a software product, including several instructions to make a computer device (which can be a personal computer, server, or network computer device, etc.) or processor (processor) perform all or part of the steps of each embodiment method of the present application. The aforementioned storage media include: various media that can store program instructions, such as USB flash drives, mobile hard drives, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks, or terminal computer devices such as computers, servers, mobile phones, and tablets. The server can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0077] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0078] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. The above is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of this application, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for autism risk assessment based on small sample data augmentation, characterized in that, Including: Collecting multi-channel electroencephalogram (EEG) signals of a subject; Calculating the correlations between the channels of the multi-channel EEG signals using the Pearson correlation coefficient, generating a brain functional connectivity matrix, and constructing a brain temporal connectivity graph according to the temporal information of the multi-channel EEG signals; Using a generative adversarial network to perform data augmentation on the brain temporal connectivity graph, and inputting the data-augmented brain temporal connectivity graph into a trained joint model of a deep convolutional network and a long short-term memory network for classification, and outputting an autism risk assessment result of the subject by a classifier.

2. The autism risk assessment method based on small sample data augmentation according to claim 1, wherein After collecting the multi-channel EEG signals of the subject, it further includes: Preprocessing the multi-channel EEG signals; the preprocessing process includes: filtering the multi-channel EEG signals using an anti-aliasing filter, digitizing and amplifying all EEG signals at a set sampling rate; Using a baseline correction algorithm to correct the amplified multi-channel EEG signals, and using an artifact detection algorithm to remove artifacts from the multi-channel EEG signals, obtaining preprocessed multi-channel EEG signals.

3. The autism risk assessment method based on small sample data augmentation according to claim 2, wherein The specific steps of calculating the correlations between the channels of the multi-channel EEG signals using the Pearson correlation coefficient, generating a brain functional connectivity matrix, and constructing a brain temporal connectivity graph according to the temporal information of the multi-channel EEG signals are as follows: Calculating the correlations between each channel and other channels in an EEG segment of a subject using the Pearson correlation coefficient, obtaining a functional connectivity matrix of a set size; Expanding the upper triangular matrix of the functional connectivity matrix into one dimension, and using the SelectKBest algorithm as a feature selection algorithm to screen the functional connectivity features, retaining a set number of main functional connectivity features; Composing the EEG signals of a set number of time segments collected from a subject into a brain functional connectivity matrix of a set size in chronological order, and converting the brain functional connectivity matrix into a brain temporal connectivity graph according to the gray value distribution.

4. The autism risk assessment method based on small sample data augmentation according to any one of claims 1 to 3, characterized in that, After constructing the brain temporal connectivity graph according to the temporal information of the multi-channel EEG signals, it further includes: Dividing the brain temporal connectivity graph into a training set and a validation set according to a set ratio.

5. The autism risk assessment method based on small sample data augmentation according to any one of claims 1 to 4, characterized in that, The specific steps of using a generative adversarial network to perform data augmentation on the brain temporal connectivity graph are as follows: The generative adversarial network includes a generator and a discriminator. The generator includes 3 deconvolutional layers, mapping a noise vector into a brain temporal connectivity graph through the deconvolutional layers to generate a new brain temporal connectivity graph; the discriminator uses 3 convolutional layers to discriminate the authenticity of the generated brain temporal connectivity graph, and after the generator and the discriminator iterate and confront for a set number of times, taking the brain temporal connectivity graph generated by the generator as an augmented pseudo-sample.

6. The autism risk assessment method based on small sample data augmentation according to claim 5, wherein The specific steps of inputting the data-augmented brain temporal connectivity graph into a trained joint model of a deep convolutional network and a long short-term memory network for classification, and outputting an autism risk assessment result of the subject by a classifier are as follows: The deep convolutional network part of the combined model of the deep convolutional network and the long short-term memory network includes 5 convolutional layers and 4 pooling layers. The local features of the input data are learned through the convolutional layers, and the convolutional layers are activated by the ReLU function; the long short-term memory network includes a hidden layer, a recurrent network layer, and a fully connected layer. After the input data undergoes input dimension conversion, it sequentially passes through the reshaping of the hidden layer and two recurrent network layers, then through two fully connected layers, and finally connects to a classifier to output a binary classification result. The binary classification result includes two categories: "autism" and "healthy".

7. An autism risk assessment device based on small sample data augmentation, characterized in that, Including: Signal acquisition module: used to acquire multi-channel electroencephalogram signals of the subject; Connectivity graph construction module: used to calculate the correlation between each channel of the multi-channel electroencephalogram signals using the Pearson correlation coefficient, obtain the correlation between each channel, generate a brain functional connectivity matrix, and construct a brain temporal connectivity graph according to the temporal information of the multi-channel electroencephalogram signals; Data augmentation module: used to perform data augmentation processing on the brain temporal connectivity graph using a generative adversarial network; Risk assessment module: used to input the brain temporal connectivity graph after the data augmentation processing into the trained combined model of the deep convolutional network and the long short-term memory network for classification, and the classifier outputs the autism risk assessment result of the subject.

8. The autism risk assessment device based on small sample data augmentation according to claim 7, wherein, The connectivity graph construction module calculates the correlation between each channel of the multi-channel electroencephalogram signals using the Pearson correlation coefficient, obtains the correlation between each channel, generates a brain functional connectivity matrix, and constructs a brain temporal connectivity graph according to the temporal information of the multi-channel electroencephalogram signals. Specifically: Calculate the correlation between each channel and other channels in an electroencephalogram segment of the subject using the Pearson correlation coefficient to obtain a functional connectivity matrix of a set size; Expand the upper triangular matrix of the functional connectivity matrix into one dimension, and use the SelectKBest algorithm as a feature screening algorithm to screen the functional connectivity features, and retain a set number of main functional connectivity features; The electroencephalogram signals of a set number of time segments collected from a subject are arranged in a functional connectivity matrix of a set size in chronological order, and the functional connectivity matrix of the brain is converted into a brain temporal connectivity graph according to the gray value distribution.

9. A computer device, characterized in that, The computer device includes a processor and a memory coupled to the processor. Among them, the memory stores program instructions for implementing the autism risk assessment method based on small sample data augmentation according to any one of claims 1-6; the processor is used to execute the program instructions stored in the memory to control the autism risk assessment method based on small sample data augmentation.

10. A storage medium, characterized in that, Stores program instructions that can be run by the processor, and the program instructions are used to execute the autism risk assessment method based on small sample data augmentation according to any one of claims 1 to 6.

Images