Method for generating multi-task fMRI based on common-specific functional brain network guidance and related equipment
By using diffusion models and functional brain network analysis, noise is gradually added and removed, and common and characteristic components are extracted. This solves the problem of insufficient quality in fMRI data generation, improves model training efficiency and accuracy, and is suitable for multi-task fMRI data generation.
Patent Information
- Application Number
- CN202610775274.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-01
- Publication Date
- 2026-08-25
AI Technical Summary
Existing generative models suffer from insufficient data scale and diversity, resulting in limited fMRI data generation quality and generalization ability. Direct application of multi-task generation methods leads to poor performance, affecting the accuracy of brain disease diagnostic models.
By acquiring fMRI data for different diagnostic tasks, noise is gradually added and removed using a diffusion model. Combined with Pearson correlation coefficient and graph-constrained robust principal component analysis, common and characteristic components are extracted to guide the generation of multi-task fMRI data.
It improved the performance of fMRI data generation, enhanced the efficiency and accuracy of model training, and the generated data performed excellently in downstream brain disease classification tasks.
Smart Images
Figure CN122636799A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and related equipment for generating multi-task fMRI based on common-specific functional brain networks. Background Technology
[0002] Functional magnetic resonance imaging (fMRI) data has been widely used to train various deep learning-based diagnostic models for brain diseases. However, the high cost of acquisition results in a limited sample size of fMRI data, while these brain disease diagnostic models typically require large-scale training data, limiting their performance. To alleviate the data scarcity problem, recent research has focused on developing generative models such as variational autoencoders (VAEs), generative adversarial networks (GANs), and their variants for fMRI signal generation. Among them, diffusion models (DMs) have shown significant advantages over VAE and GAN-based methods due to their higher fidelity of generated samples and model stability. However, these generative models, mostly developed based on single-task datasets, often face the challenge of insufficient training data. Limited by the scale and diversity of data, the generation quality and generalization ability of existing generative models are affected.
[0003] Multi-task generation methods proposed in other application areas can, to some extent, address the limitations imposed by the small scale and insufficient diversity of current data. However, when these methods are directly applied to fMRI generation, their performance is poor due to a failure to fully consider the essential characteristics of fMRI data, further reducing the accuracy of the data model. Therefore, there are still technical problems that need to be solved in these related technologies. Summary of the Invention
[0004] The purpose of this application is to at least partially solve one of the technical problems existing in the prior art.
[0005] Therefore, one objective of this application is to provide a multi-task fMRI generation method and related equipment based on common-specific functional brain networks, which can improve the performance of multi-task fMRI used for model training and improve the efficiency and quality of model training.
[0006] To achieve the above-mentioned technical objectives, the technical solution adopted in the embodiments of this application includes: a multi-task fMRI generation method based on common-specific functional brain networks, comprising the following steps:
[0007] Acquire first fMRI data for different diagnostic tasks; each task corresponds to several fMRI sub-data.
[0008] Noise is gradually added to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, so as to train the diffusion model to learn the transition distribution from the first fMRI data to the pure noise data;
[0009] The noise in the pure noise data input to the diffusion model is gradually removed in chronological order until the pure noise data is transformed into noise-free data as the second fMRI data, so as to train the diffusion model to learn the probability distribution of the pure noise data to the second fMRI data;
[0010] The first fMRI data is decomposed to obtain the common and characteristic components of the fMRI classification task;
[0011] Based on the common components, the characteristic components, and the diffusion model, the second fMRI data is guided to obtain a multi-task fMRI dataset.
[0012] Furthermore, in this embodiment of the application, the method further includes: resampling and cropping the first fMRI data to obtain the first fMRI data with the same time scale.
[0013] Furthermore, in this embodiment of the application, the first fMRI data is decomposed to obtain the common components and characteristic components of the fMRI classification task, including:
[0014] The functional brain network of each sample in the first fMRI data is calculated using the Pearson correlation coefficient, and the category functional brain network is obtained by calculating the average value of the functional brain networks of all categories of samples in the first fMRI data.
[0015] The functional brain network of the aforementioned categories was decomposed using graph-constrained robust principal component analysis to obtain common components shared by all tasks and characteristic components of the fMRI classification task.
[0016] Furthermore, in this embodiment of the application, the decomposition of the categorical functional brain network using graph-constrained robust principal component analysis to obtain common components shared by all tasks and characteristic components of the fMRI classification task includes:
[0017] For the j-th class of data in the i-th classification task, the category functional brain network is input into the calculation formula to obtain the common components shared by all tasks and the characteristic components of the fMRI classification task, wherein the calculation formula is:
[0018]
[0019] in, For category-functional brain networks, C is the characteristic component, and C is the common component.
[0020] Furthermore, in this embodiment of the application, the second fMRI data is guided based on the common components, the characteristic components, and the diffusion model to obtain a multi-task fMRI dataset, specifically including:
[0021] The second fMRI data, the common components, and the characteristic components are input into the diffusion model to obtain a multi-task fMRI dataset; the diffusion model is:
[0022]
[0023] F t This is the second time series data, where C is the common component and X is... t This is the second fMRI data.
[0024] Further, in this embodiment of the application, the step of progressively adding noise to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, so as to train the diffusion model to learn the transition distribution from the first fMRI data to the pure noise data, includes:
[0025] During the forward pass of the training phase, for the input-aligned time series X0, Gaussian noise will be superimposed sequentially over T time steps to obtain X1, X2, ..., X... T ; where X t Let X represent the noisy data at step t, and the final X... T It is a pure noise time series; from X t−1 To X t The mathematical process follows the following transition distribution:
[0026]
[0027] Where, β t I represents the noise variance that controls the amount of noise introduced at each step, and I is the identity matrix.
[0028] Further, in this embodiment of the application, the step of sequentially superimposing Gaussian noise in T time steps includes the following steps: adding noise to the time series generated in the previous t-1 time steps in the t-th time step to generate the time series in the t-th time step; the expression for adding noise to generate the time series is as follows:
[0029]
[0030] in, From the standard normal distribution The sample drawn from the middle, and This represents the cumulative noise variance.
[0031] On the other hand, embodiments of this application also provide a multi-task fMRI generation system guided by common-specific functional brain networks, comprising:
[0032] The first processing unit is used to acquire first fMRI data for different diagnostic tasks; where each task corresponds to several fMRI sub-data.
[0033] The second processing unit is used to add noise to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, so as to train the diffusion model to learn the transition distribution from the first fMRI data to the pure noise data.
[0034] The third processing unit is used to gradually remove noise from the pure noise data input to the diffusion model in chronological order until the pure noise data is transformed into noise-free data as the second fMRI data, so as to train the diffusion model to learn the probability distribution of the pure noise data to the second fMRI data.
[0035] The fourth processing unit is used to decompose the first fMRI data to obtain the common and characteristic components of the fMRI classification task.
[0036] The fifth processing unit is used to guide the second fMRI data based on the common components, the characteristic components, and the diffusion model to obtain a multi-task fMRI dataset.
[0037] On the other hand, this application also provides a multi-task fMRI generation device guided by common-specific functional brain networks, comprising:
[0038] At least one processor;
[0039] At least one memory for storing at least one program;
[0040] When the at least one program is executed by the at least one processor, the at least one processor implements a multi-task fMRI generation method based on common-specific functional brain networks as described in any one of the inventions.
[0041] Furthermore, this application also provides a computer-readable storage medium storing processor-executable instructions, which, when executed by a processor, are used to perform a multi-task fMRI generation method guided by a common-specific functional brain network as described in any of the preceding claims.
[0042] The advantages and beneficial effects of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application:
[0043] This application can acquire first fMRI data for different diagnostic tasks; each task corresponds to several fMRI sub-data; noise is progressively added to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, so as to train the diffusion model to learn the transition distribution from the first fMRI data to the pure noise data; noise is progressively removed from the pure noise data input to the diffusion model in chronological order until the pure noise data is transformed into noise-free data as second fMRI data, so as to train the diffusion model to learn the probability distribution from the pure noise data to the second fMRI data; the first fMRI data is decomposed to obtain common components and characteristic components of the fMRI classification task; based on the common components, the characteristic components, and the diffusion model, the second fMRI data is guided to obtain a multi-task fMRI dataset. This application can improve the performance of the generated fMRI data by combining the diffusion model to generate fMRI data, thereby improving the efficiency and accuracy of model training. Attached Figure Description
[0044] Figure 1 This is a schematic diagram illustrating the steps of a multi-task fMRI generation method guided by common-specific functional brain networks in a specific embodiment of the present invention;
[0045] Figure 2 This is a flowchart illustrating a multi-task fMRI generation method based on common-specific functional brain networks in a specific embodiment of the present invention.
[0046] Figure 3 This is a schematic diagram of the structure of a multi-task fMRI generation system guided by a common-specific functional brain network in another specific embodiment of the present invention;
[0047] Figure 4 This is a schematic diagram of the structure of a multi-task fMRI generation device guided by a common-specific functional brain network in a specific embodiment of the present invention;
[0048] Figure 5 This is a schematic diagram illustrating the effects of different data generation methods on the ADNI dataset in a specific embodiment of the present invention;
[0049] Figure 6 This is a schematic diagram illustrating the effects of different data generation methods on the Huashan dataset in a specific embodiment of the present invention;
[0050] Figure 7 This is a schematic diagram illustrating the effects of different data generation methods on the Zhongshan dataset in a specific embodiment of the present invention. Detailed Implementation
[0051] The following description, in conjunction with the accompanying drawings, details the principles and processes of the multi-task fMRI generation method and related equipment based on common-specific functional brain networks guided by the present invention.
[0052] Reference Figure 1 This application provides a multi-task fMRI generation method based on common-specific functional brain networks, including steps S101-S105.
[0053] S101. Acquire the first fMRI data for different diagnostic tasks; where each task corresponds to several fMRI sub-data.
[0054] S102. Add noise to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, so as to train the diffusion model to learn the transition distribution from the first fMRI data to the pure noise data.
[0055] S103. The noise in the pure noise data input to the diffusion model is gradually removed in chronological order until the pure noise data is transformed into noise-free data as the second fMRI data, so as to train the diffusion model to learn the probability distribution of pure noise data to the second fMRI data.
[0056] S104. Perform data decomposition on the first fMRI data to obtain the common and characteristic components of the fMRI classification task.
[0057] S105. Based on common components, characteristic components, and diffusion models, the second fMRI data is guided to obtain a multi-task fMRI dataset.
[0058] It is understandable that the first fMRI data can be a dataset that includes multiple fMRI sub-data; in this context, each task can correspond to three or more fMRI sub-data, and the first fMRI data can correspond to multiple classification diagnostic tasks.
[0059] Furthermore, in this embodiment of the application, the multi-task fMRI generation method guided by common-specific functional brain networks further includes: resampling and cropping the first fMRI data to obtain first fMRI data with the same time scale. It is understood that this step can be performed before step S102; that is, in step S102, noise is gradually added to the first fMRI data input to the diffusion model in chronological order. At this time, the first fMRI data input to the diffusion model is resampled and cropped, cropping the time of all first fMRI data to a uniform length.
[0060] Furthermore, in this embodiment of the application, the first fMRI data is decomposed to obtain the common components and characteristic components of the fMRI classification task, including:
[0061] The functional brain network of each sample in the first fMRI data was calculated using the Pearson correlation coefficient, and the category functional brain network was obtained by calculating the average of the functional brain networks of all categories in the first fMRI data.
[0062] The categorical functional brain network was decomposed using graph-constrained robust principal component analysis to obtain common components shared by all tasks and characteristic components of the fMRI classification task.
[0063] Furthermore, in this embodiment, the categorical functional brain network is decomposed using graph-constrained robust principal component analysis to obtain common components shared by all tasks and characteristic components of the fMRI classification task, including:
[0064] For the j-th class of data in the i-th classification task, the category functional brain network is input into the calculation formula to obtain the common components shared by all tasks and the characteristic components of the fMRI classification task, where the calculation formula is:
[0065]
[0066] in, For category-functional brain networks, C is the characteristic component, and C is the common component.
[0067] Furthermore, in this embodiment, based on common components, characteristic components, and a diffusion model, the second fMRI data is guided to obtain a multi-task fMRI dataset, specifically including:
[0068] The second fMRI data, common components, and characteristic components are input into the diffusion model to obtain the multi-task fMRI dataset; the diffusion model is as follows:
[0069]
[0070] F tThis is the second time series data, where C is the common component and X is... t This is the second fMRI data.
[0071] Furthermore, in this embodiment, noise is progressively added to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, so as to train the diffusion model to learn the transition distribution from the first fMRI data to the pure noise data, including:
[0072] During the forward pass of the training phase, for the input-aligned time series X0, Gaussian noise will be superimposed sequentially over T time steps to obtain X1, X2, ..., X... T ; where X t Let X represent the noisy data at step t, and the final X... T It is a pure noise time series; from X t−1 To X t The mathematical process follows the following transition distribution:
[0073]
[0074] Where, β t I represents the noise variance that controls the amount of noise introduced at each step, and I is the identity matrix.
[0075] Furthermore, in this embodiment, Gaussian noise is sequentially superimposed over T time steps, including the following steps: In the t-th time step, noise is added to the time series generated in the preceding t-1 time steps to generate the time series at the t-th time step; the expression for adding noise to generate the time series is as follows:
[0076]
[0077] in, From the standard normal distribution The sample drawn from the middle, and This represents the cumulative noise variance.
[0078] Specifically, the specific process of this application is as follows: Figure 2 To mitigate temporal heterogeneity in multi-task functional magnetic resonance imaging (fMRI) data, this embodiment employs a temporal alignment method that includes resampling and temporal clipping operations. Figure 2 In this process, stage (b) is the training stage of the diffusion model. Its purpose is to learn the denoising model parameters to recover the original data from noisy data, rather than to generate the final data. The X constructed during this process... t and X T These represent noisy data and pure noise data at different time steps, respectively, both of which are intermediate variables in the training process.
[0079] Figure 2 (b) From X0 to X T The arrows indicate the forward diffusion process, which involves gradually adding Gaussian noise to the original sample until it becomes pure noise; from X... T The arrow to X0 indicates the reverse denoising process, which is used to train the model to learn to predict noise and gradually remove it from pure noise to restore the original sample. Figure 2 (d) is the generation process of the diffusion model guided by common-specific FBNs. Arrow X T →X t This represents the inverse denoising process in (b). In this denoising process, the Pearson correlation coefficient is also used to denoise X. t The calculated FBN, decomposition strategy generates common and specific component pairs in the noisy fMRI data X. t Guided to generate the final multi-task fMRI. Figure 2 (d) The X0 obtained in the process is the multi-task fMRI data that this application expects to generate.
[0080] First, for the different repetition times (TRs) in the multi-task fMRI dataset, the original fMRI time series need to be temporally resampled to unify the target TR. For the fMRI time series of task i... N represents the number of regions of interest, L i (representing the total number of original time points), resampling length L i Defined as
[0081]
[0082] in, and These represent the original and target TR values, respectively. This is the length of the resampled fMRI time series. Subsequently, to ensure consistency in the input dimension among the multi-task fMRI datasets, the resampled time series are truncated to a uniform length L, ultimately obtaining the aligned datasets for all tasks. .
[0083] Specifically, the aligned dataset The input diffusion model is used for training. First, during the forward pass of the training phase, for the aligned input time series X0, Gaussian noise is sequentially superimposed over T time steps, resulting in X1, X2, ..., X... T Where Xt represents the noisy data at step t, X tThis refers to intermediate state data obtained at time step t by progressively adding Gaussian noise to the original fMRI time series. This data is not generated by model inference, but rather serves as noise perturbation samples during model training. The final X... T It is a purely noisy time series. From X t-1 To X t The mathematical process follows the following transition distribution:
[0084]
[0085] Where βt represents the noise variance that controls the amount of noise introduced at each step, and I is the identity matrix. This indicates that X t Follow the mean The variance is β t I follows a normal distribution. To generate X at a specific step t... t The noise addition process can be reparameterized:
[0086]
[0087] in, From the standard normal distribution The sample drawn from the middle, and This represents the cumulative noise variance. Subsequently, during the reverse process of training, the final noisy data X is processed. T Perform noise reduction processing, and gradually reduce noise from X. T To recover the original data X0, i.e., to learn X T The joint probability distribution to X0 can be expressed as:
[0088]
[0089] Among them, from X t To X t-1 The inverse conditional probability distribution can be expressed as:
[0090]
[0091] Among them, the mean Covariance Noise prediction can be performed using a noise prediction network, where θ represents the network parameters. To make the predicted noise... More closely approximating the actual noise during the forward diffusion process The loss function L can be defined as:
[0092]
[0093] The above training process can capture the underlying distribution of fMRI data and gradually recover the original data from the noise.
[0094] Specifically, given that fMRI data from different tasks inherently possess both common functional connectivity patterns and disease-specific heterogeneity, this study employs a decomposition strategy to generate common and specific components, providing guidance for the data generation process to improve the reliability and specificity of the synthesized data. First, the functional brain network of each sample in the fMRI data is calculated using the Pearson correlation coefficient. Then, the categorical functional brain network is obtained by calculating the average value of the functional brain networks at the sample level for each category within each dataset. Functional brain network decomposition is achieved through graph-constrained robust principal component analysis (PCA): the low-rank regularization technique can capture shared connectivity structures and effectively handle the sparsity model differences between multi-task data. For the j-th class of data in the i-th classification task, its categorical functional brain network... It can be decomposed into low-rank common components shared by all tasks. and sparse specific components The decomposed patterns are then incorporated as structural guiding factors into the generation process to improve the effectiveness and reliability of the synthesized data.
[0095] Specifically, the data Xt generated in a particular step is guided by the following mechanism to obtain functional connectivity information similar to that in the prior FBN:
[0096]
[0097] Among them, F t This indicates that by using the Pearson correlation coefficient from The calculated FBN is also the second time series data. It is a weighting factor that adjusts the intensity of guidance from common components and specific components. The Frobenius norm is used to quantify differences. These are the update coefficients; the above FBN guides each This is performed once per step. In the early denoising stage of the generation process, The small value indicates that the model initially focused on common components to preserve shared functional connectivity patterns. As denoising progresses, the number of common components increases... The focus shifts to generated data with disease-specific characteristics.
[0098] Specifically, to verify the performance of a general multi-task fMRI generation method guided by common-specific functional brain networks in downstream brain disease classification tasks, this embodiment conducted experiments on three real-world datasets. The datasets included the public Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, the Huashan dataset for distinguishing between mild cognitive impairment (MCI) and normal controls (NC), and the Zhongshan dataset for distinguishing between vascular cognitive impairment (VCI) and normal controls (NC). The ADNI dataset contained 407 subjects (183 MCI patients and 224 NC controls), the Huashan dataset contained 513 subjects (204 MCI patients and 309 NC controls), and the Zhongshan dataset contained 297 subjects (151 VCI patients and 146 NC controls). All data were preprocessed using the DPARSFA toolkit according to standard procedures, and mean time series of gray matter from 90 brain regions were extracted using an automated anatomical labeling atlas.
[0099] Specifically, for the first fMRI data, the number of brain regions N was set to 90, and the time series length L was set to 130, 488, and 220, respectively, corresponding to the ADNI, Huashan, and Zhongshan datasets. The aligned length was set to 2000 ms, and the uniform length L was 195 time points. During training, the batch size was set to 24, the number of training epochs was set to 1000, the learning rate was set to 0.0001, and the Adam optimizer was used. The noise addition step T was set to 200, with linear sampling from [0.0015, 0.0195]. During generation, the update coefficient was set to 1, and the FBN guidance interval ∆t was set to 5. These parameters were selected by cross-validation of the training dataset. In this embodiment, 100 samples (50 MCI: 50 NC) were generated for the ADNI and Huashan datasets, and 100 samples (50 VCI: 50 NC) were generated for the Zhongshan dataset. Five-fold cross-validation was performed, and the same data partitioning method was used in all competing methods. Finally, the amplified data were classified. The method proposed in this application is named UM-fMRI-GM. The classification performance of the model after data augmentation using it is as follows: Figure 5 , Figure 6 as well as Figure 7 As shown, Figure 5 To illustrate the performance of different data generation methods on the ADNI dataset, Figure 6 To illustrate the performance of different data generation methods on the Huashan dataset, Figure 7The performance of different data generation methods on the Zhongshan dataset.
[0100] The method proposed in this embodiment, after data augmentation, achieves higher classification accuracy on different datasets than other data generation methods. This indicates that joint training with multi-task fMRI data, and using both shared functional connectivity and disease-specific functional connectivity modes to guide the generation process, is beneficial for improving the quality of generated fMRI samples and model stability, resulting in biologically plausible and diagnostically valuable samples.
[0101] In addition, refer to Figure 3 ,and Figure 1 Corresponding to the method described above, embodiments of this application also provide a multi-task fMRI generation system guided by a common-specific functional brain network. This system may include a first processing unit 1001, a second processing unit 1002, a third processing unit 1003, a fourth processing unit 1004, and a fifth processing unit 1005. The first processing unit 1001 is used to acquire first fMRI data for different diagnostic tasks; each task corresponds to several fMRI sub-data. The second processing unit 1002 is used to progressively add noise to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, thereby training the diffusion model to learn the transition distribution from the first fMRI data to pure noise data. The third processing unit 1003 is used to progressively remove noise from the pure noise data input to the diffusion model in chronological order until the pure noise data is transformed into noise-free data as second fMRI data, thereby training the diffusion model to learn the probability distribution from the pure noise data to the second fMRI data. The fourth processing unit 1004 is used to perform data decomposition on the first fMRI data to obtain the common components and characteristic components of the fMRI classification task. The fifth processing unit 1005 is used to guide the second fMRI data based on common components, characteristic components and diffusion models to obtain a multi-task fMRI dataset.
[0102] It should be noted that the first processing unit can be any integrated circuit unit or microprocessor unit obtained by integrating a chip with processing functions and its peripheral circuits using existing integration technology. The first processing unit and the second processing unit can also be any integrated circuit module or microprocessor module obtained by integrating a chip with processing functions and its peripheral circuits using existing integration technology. Furthermore, the first processing unit and the second processing unit may include one or more memories.
[0103] It should be noted that the content of the above-described embodiments of the multi-task fMRI generation method based on common-specific functional brain networks is applicable to this embodiment of the multi-task fMRI generation system based on common-specific functional brain networks. The specific functions implemented by this embodiment of the multi-task fMRI generation system based on common-specific functional brain networks are the same as those of the above-described embodiments of the multi-task fMRI generation method based on common-specific functional brain networks, and the beneficial effects achieved are also the same as those achieved by the above-described embodiments of the multi-task fMRI generation method based on common-specific functional brain networks.
[0104] and Figure 1 Corresponding to the method described herein, embodiments of this application also provide a multi-task fMRI generation device guided by common-specific functional brain networks, the specific structure of which can be referred to Figure 4 ,include:
[0105] At least one processor 1011;
[0106] At least one memory 1012 is used to store at least one program;
[0107] When the at least one program is executed by the at least one processor, the at least one processor implements the multi-task fMRI generation method guided by common-specific functional brain networks.
[0108] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0109] and Figure 1 Corresponding to the method described above, embodiments of this application also provide a computer-readable storage medium storing processor-executable instructions, which, when executed by a processor, are used to perform the multi-task fMRI generation method guided by common-specific functional brain networks.
[0110] The contents of the above embodiments of the multi-task fMRI generation method based on common-specific functional brain networks are all applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above embodiments of the multi-task fMRI generation method based on common-specific functional brain networks, and the beneficial effects achieved are also the same as those achieved in the above embodiments of the multi-task fMRI generation method based on common-specific functional brain networks.
[0111] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.
[0112] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional technology for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.
[0113] If the aforementioned functions are implemented as software functional units 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 programs 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.
[0114] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable programs for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, a program execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can retrieve and execute a program from or in conjunction with such a program execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit a program for use by or in conjunction with a program execution system, apparatus, or device.
[0115] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0116] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable program execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0117] In the foregoing description of this specification, the references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0118] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.
[0119] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.
Claims
1. A multi-task fMRI generation method based on common-specific functional brain networks, characterized in that, Includes the following steps: Acquire first fMRI data for different diagnostic tasks; each task corresponds to several fMRI sub-data. Noise is gradually added to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, so as to train the diffusion model to learn the transition distribution from the first fMRI data to the pure noise data; The noise in the pure noise data input to the diffusion model is gradually removed in chronological order until the pure noise data is transformed into noise-free data as the second fMRI data, so as to train the diffusion model to learn the probability distribution of the pure noise data to the second fMRI data; The first fMRI data is decomposed to obtain the common and characteristic components of the fMRI classification task; Based on the common components, the characteristic components, and the diffusion model, the second fMRI data is guided to obtain a multi-task fMRI dataset.
2. The method for generating multi-task fMRI based on common-specific functional brain networks according to claim 1, characterized in that, The method further includes: resampling and cropping the first fMRI data to obtain the first fMRI data with the same time scale.
3. The method for generating multi-task fMRI based on common-specific functional brain networks according to claim 1, characterized in that, The first fMRI data is decomposed to obtain the common and characteristic components of the fMRI classification task, including: The functional brain network of each sample in the first fMRI data is calculated using the Pearson correlation coefficient, and the category functional brain network is obtained by calculating the average value of the functional brain networks of all categories of samples in the first fMRI data. The functional brain network of the aforementioned categories was decomposed using graph-constrained robust principal component analysis to obtain common components shared by all tasks and characteristic components of the fMRI classification task.
4. The method for generating multi-task fMRI based on common-specific functional brain networks according to claim 3, characterized in that, The decomposition of the functional brain network using graph-constrained robust principal component analysis yields common components shared by all tasks and characteristic components specific to the fMRI classification task, including: For the j-th class of data in the i-th classification task, the category functional brain network is input into the calculation formula to obtain the common components shared by all tasks and the characteristic components of the fMRI classification task, wherein the calculation formula is: in, For category-functional brain networks, C is the characteristic component, and C is the common component.
5. The method for generating multi-task fMRI based on common-specific functional brain networks according to claim 4, characterized in that, Based on the common components, the characteristic components, and the diffusion model, the second fMRI data is guided to obtain a multi-task fMRI dataset, specifically including: The second fMRI data, the common components, and the characteristic components are input into the diffusion model to obtain a multi-task fMRI dataset; the diffusion model is: F t This is the second time series data, where C is the common component and X is... t This is the second fMRI data.
6. The method for generating multi-task fMRI based on common-specific functional brain networks according to claim 1, characterized in that, The step of progressively adding noise to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, in order to train the diffusion model to learn the transition distribution from the first fMRI data to the pure noise data, includes: During the forward pass of the training phase, for the input-aligned time series X0, Gaussian noise will be superimposed sequentially over T time steps to obtain X1, X2, ..., X... T ; where X t Let X represent the noisy data at step t, and the final X... T It is a pure noise time series; from X t−1 To X t The mathematical process follows the following transition distribution: Where, β t I represents the noise variance that controls the amount of noise introduced at each step, and I is the identity matrix.
7. The method for generating multi-task fMRI based on common-specific functional brain networks according to claim 1, characterized in that, The method of sequentially superimposing Gaussian noise over T time steps includes the following steps: In the t-th time step, noise is added to the time series generated in the preceding t-1 time steps to generate the time series at the t-th time step; the expression for adding noise to generate the time series is as follows: in, From the standard normal distribution The sample drawn from the middle, and This represents the cumulative noise variance.
8. A multi-task fMRI generation system guided by common-specific functional brain networks, characterized in that, include: The first processing unit is used to acquire first fMRI data for different diagnostic tasks; Each task corresponds to several fMRI sub-data sets; The second processing unit is used to add noise to the first fMRI data input to the diffusion model in chronological order until the first fMRI data is transformed into pure noise data, so as to train the diffusion model to learn the transfer distribution from the first fMRI data to the pure noise data. The third processing unit is used to gradually remove noise from the pure noise data input to the diffusion model in chronological order until the pure noise data is transformed into noise-free data as the second fMRI data, so as to train the diffusion model to learn the probability distribution of the pure noise data to the second fMRI data. The fourth processing unit is used to decompose the first fMRI data to obtain the common and characteristic components of the fMRI classification task. The fifth processing unit is used to guide the second fMRI data based on the common components, the characteristic components, and the diffusion model to obtain a multi-task fMRI dataset.
9. A multi-task fMRI generation device guided by common-specific functional brain networks, characterized in that... include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the multi-task fMRI generation method based on common-specific functional brain networks as described in any one of claims 1-7.
10. A computer-readable storage medium storing processor-executable instructions, characterized in that, The processor-executable instructions, when executed by the processor, are used to perform a multi-task fMRI generation method based on a common-specific functional brain network as described in any one of claims 1-7.