Denoising method, system and equipment for functional magnetic resonance image data and medium

Abstract

The invention belongs to the technical field of medical image processing, and particularly relates to a functional magnetic resonance image data denoising method, system and device and a medium, and the method comprises the steps: obtaining child brain functional magnetic resonance original time sequence image data, child brain development prior map data, historical same-age denoising standard data and physiological movement synchronous collection data; brain tissue region segmentation and region-of-interest positioning are completed, positioning results of different brain regions are obtained, and time sequence signal features and space structure features of the different brain regions are extracted; constructing a children brain functional magnetic resonance multi-dimensional adaptive denoising model, generating a brain region noise type thermodynamic diagram, and generating a brain region hierarchical differential denoising strategy through a signal-noise separation algorithm to obtain denoised functional magnetic resonance image data; and generating a de-noising quality evaluation report and updating the parameter weight of the multi-dimensional adaptive de-noising model. Therefore, the problems of low denoising accuracy and the like in the prior art are solved.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, specifically relating to methods, systems, devices, and media for denoising functional magnetic resonance imaging data. Background Technology

[0002] Functional magnetic resonance imaging (fMRI), a non-invasive neuroimaging technique, has become a core tool for studying typical and atypical brain development in children and diagnosing neurodevelopmental disorders. It effectively captures oxygen-dependent signals in the brain, reflecting the spontaneous activity and functional connectivity characteristics of neurons. Denoising techniques for pediatric brain fMRI images are crucial for ensuring the accuracy of subsequent image analysis. Existing denoising methods have formed a multi-dimensional technical system, mainly including traditional filtering, head motion parameter regression, independent component analysis, and volume censoring. These methods can specifically remove various types of noise generated during the scanning process—suppressing equipment-related noise such as scanner magnetic field inhomogeneity and electromagnetic interference, mitigating head motion artifacts caused by children's hyperactivity and difficulty concentrating, and reducing signal interference caused by physiological activities such as breathing and heartbeat. The application of these existing technologies effectively improves the signal-to-noise ratio of pediatric brain fMRI images, reduces noise interference in brain functional network analysis and lesion feature identification, provides fundamental data support for the study of pediatric brain development trajectory, early screening and mechanism exploration of neurodevelopmental disorders such as epilepsy and attention deficit hyperactivity disorder, and promotes the development and clinical application of pediatric neuroimaging.

[0003] Current denoising techniques for pediatric brain functional magnetic resonance imaging (fMRI) images suffer from several unresolved problems, including: Firstly, existing technologies struggle to fully acquire and integrate prior information about children's brain development, age-appropriate gold standard data for denoising, and simultaneously acquired physiological motion data. They rely heavily on adult data, failing to adapt to the unique physiological characteristics of children, such as high respiratory rates and low-amplitude, high-frequency head movements. This makes it difficult to accurately distinguish noise from real neural signals, hindering precise denoising. Secondly, they struggle to accurately segment brain tissue regions and locate regions of interest, and cannot effectively extract temporal signal features and spatial structural features from each brain region. Furthermore, they struggle to accurately analyze noise types in different brain regions, making it impossible to develop targeted, differentiated denoising strategies that address significant differences in noise across brain regions. Thirdly, they struggle to construct multi-dimensional adaptive denoising models suitable for children. Faced with complex scenarios involving overlapping equipment noise, motion artifacts, and physiological noise, they cannot achieve efficient and accurate signal-noise separation, resulting in insufficient denoising accuracy. Fourthly, they struggle to simultaneously evaluate denoising quality, cannot calculate image signal-to-noise ratio and brain region neural signal fidelity in real time, and cannot update the parameter weights of the denoising model based on relevant data, making it difficult to guarantee the stability and reliability of the denoising effect. Summary of the Invention

[0004] This application provides a method, system, device, and medium for denoising functional magnetic resonance imaging data to address the problem of insufficient denoising accuracy in the prior art.

[0005] The first aspect of this application provides a method for denoising functional magnetic resonance imaging (fMRI) data, comprising the following steps: acquiring raw temporal fMRI image data of a child's brain, prior brain development atlas data, historical age-matched denoised gold standard data, and synchronously acquired physiological motion data; based on the raw temporal fMRI image data and prior brain development atlas data, performing brain tissue region segmentation and region of interest localization to obtain localization results for different brain regions, and extracting temporal signal features and spatial structural features of different brain regions; based on the historical age-matched denoised gold standard data, synchronously acquired physiological motion data, and the temporal signal features and spatial structural features of the brain regions... Based on the structural features of the brain, a multi-dimensional adaptive denoising model of functional magnetic resonance imaging (fMRI) of children's brain is constructed. Combining the localization results of different brain regions, a heatmap of brain region noise types is generated. Through a signal-noise separation algorithm, the distinction between neural signals and noise components in each brain region is obtained, generating a brain region differentiation-based denoising strategy. Adaptive denoising processing is then performed on the original time-series fMRI image data to obtain denoised fMRI image data. Based on the denoised fMRI image data, image signal-to-noise ratio evaluation data and brain region neural signal fidelity data are calculated simultaneously to generate a denoising quality evaluation report and update the parameter weights of the multi-dimensional adaptive denoising model.

[0006] Preferably, the brain tissue region segmentation and region of interest localization are completed to obtain localization results for different brain regions, and temporal signal features and spatial structural features of different brain regions are extracted, including: constructing a pediatric brain tissue segmentation and brain region localization model; based on the pediatric brain tissue segmentation and brain region localization model, combined with age-matched brain region anatomical structure and functional zoning features in the pediatric brain development prior atlas data, the brain gray and white matter, cerebrospinal fluid, and non-brain tissues of the skull are segmented using an improved 3DU-Net network, and the spatial association features of each functional brain region are captured using a graph attention network to complete the region of interest localization and obtain localization results for different brain regions; based on the localization results of different brain regions, the temporal sequence fluctuation features and frequency domain features of the functional magnetic resonance signals of each brain region are extracted as temporal signal features, and the anatomical boundary features, gray-level distribution features, and spatial texture features of the brain regions are extracted as spatial structural features.

[0007] Preferably, generating a brain region noise type heatmap includes: constructing a regional spatial grid partitioning model adapted to the characteristics of children's brain development; based on the regional spatial grid partitioning model adapted to the characteristics of children's brain development, combined with the brain region partitioning boundaries and key developmental region annotations in the prior brain development atlas data of children, as well as the noise fluctuation amplitude in the temporal signal characteristics and the artifact distribution characteristics in the spatial structure characteristics, calculating the noise density value and noise type confidence of each grid unit through a kernel density estimation algorithm; mapping the noise density value and noise type confidence to a preset color gradient interval, superimposing it onto the three-dimensional brain region template of the prior brain development atlas of children, and generating a brain region noise type heatmap annotating the noise type, density level, confidence, and update timestamp of each grid unit.

[0008] Preferably, the signal-noise separation algorithm is used to obtain the distinction results between neural signals and noise components in each brain region, including: constructing a brain region-specific signal-noise separation model; based on the brain region-specific signal-noise separation model, using the sliding window method to extract the non-stationary fluctuation features of the temporal signals of each brain region, and using independent component analysis combined with gradient boosting tree algorithm to complete the component separation of neurophysiological signals from motion noise, physiological noise, and instrument noise, and outputting the distinction results between neural signals and noise components in each brain region; based on the distinction results between neural signals and noise components in each brain region, obtaining the noise component proportion of the corresponding brain region; when the proportion of the noise component exceeds the preset noise threshold of the corresponding brain region, triggering the brain region adaptive weighted denoising mechanism to generate targeted denoising parameter correction instructions.

[0009] Preferably, the simultaneous calculation of image signal-to-noise ratio (SNR) assessment data and brain region neural signal fidelity data includes: constructing a two-dimensional quantitative assessment model for the denoising effect of functional magnetic resonance imaging (fMRI) in children's brains; based on the two-dimensional quantitative assessment model for the denoising effect of fMRI in children's brains, combined with denoised fMRI image data, original temporal fMRI image data of children's brains, historical gold standard data for denoising at the same age, and localization results of different brain regions, calculating the SNR, peak SNR, and structural similarity index of the whole brain and each brain region using a medical image SNR statistical algorithm to obtain image SNR assessment data; simultaneously calculating the temporal correlation, low-frequency oscillation amplitude fidelity, and functional connectivity strength fidelity of neural signals in each brain region using a neurophysiological feature matching algorithm to obtain brain region neural signal fidelity data; performing brain region matching and association between the image SNR assessment data and the brain region neural signal fidelity data, and outputting a two-dimensional quantitative assessment dataset with brain region labels.

[0010] Preferably, the signal-noise separation algorithm formula is as follows: ; ; ; ; ; ; ; ; in, A matrix of raw time-series images from functional magnetic resonance imaging (fMRI) of the brain in children; A design matrix that includes both neural networks and noise; This is the regression coefficient matrix; The neural signal is in residual form; For the first Brain region spatial mask matrix; For the first Brain region neural signal regression matrix; These are the regression coefficients of the neural signal; This is the noise reference regression matrix; The noise regression coefficient; For the first residual neural signals in brain regions; It is a signal mixing matrix; It is a set of independent components; This is the total noise matrix; for Time-major filtering weight vector; for Time-weight vector; This is the step size parameter; To prevent zero constant; for Timing error signal; for Time-of-flight noise reference vector; The estimated neural signal matrix; To find the optimal solution operator; For regularization weights; For the development of the prior space regularization matrix; These are real neural signals; These are the wavelet coefficients after thresholding. The sign function for wavelet coefficients; These are the original wavelet coefficients; For the first Noise-reducing neural signals in brain regions; For the first Adaptive brain region separation operator; This is the physiological motion noise reference matrix; A priori atlas of child development; These are the model weight parameters; This is the denoised functional magnetic resonance imaging data.

[0011] The second aspect of this application provides a denoising system for functional magnetic resonance imaging (fMRI) data, comprising: a data acquisition module for acquiring raw time-series fMRI images of a child's brain, prior brain development atlas data, historical age-matched denoising gold standard data, and synchronously acquired physiological motion data; a brain region feature extraction module for performing brain tissue region segmentation and region of interest localization based on the raw time-series fMRI images of the child's brain and the prior brain development atlas data, obtaining localization results for different brain regions, and extracting temporal signal features and spatial structural features of different brain regions; and an adaptive denoising module for performing denoising based on the historical age-matched denoising gold standard data, synchronously acquired physiological motion data, and the raw time-series fMRI images of the child's brain and the prior brain development atlas data, obtaining the localization results for different brain regions, and extracting temporal signal features and spatial structural features of different brain regions; and an adaptive denoising module for performing denoising based on the raw time-series fMRI images of the child's brain, the historical age-matched denoising gold standard data, the synchronously acquired physiological motion data, and the synchronously acquired physiological motion data. Based on the sequence signal features and spatial structure features, a multi-dimensional adaptive denoising model for functional magnetic resonance imaging (fMRI) of the child's brain is constructed. Combined with the localization results of different brain regions, a heatmap of brain region noise types is generated. Through a signal-noise separation algorithm, the distinction between neural signals and noise components in each brain region is obtained, generating a differentiated denoising strategy for brain region differentiation. Adaptive denoising processing is then performed on the original time-series fMRI image data to obtain denoised fMRI image data. An evaluation and optimization module simultaneously calculates image signal-to-noise ratio evaluation data and brain region neural signal fidelity data based on the denoised fMRI image data, generating a denoising quality evaluation report and updating the parameter weights of the multi-dimensional adaptive denoising model.

[0012] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the denoising method for functional magnetic resonance image data as described in the above embodiments.

[0013] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a method for denoising functional magnetic resonance image data as described in the above embodiments.

[0014] Therefore, this application includes the following beneficial effects: by integrating prior brain development atlases of children, the gold standard for denoising within the same age group, and synchronously acquired physiological and motor data, it achieves precise segmentation and region of interest localization of brain tissue regions, efficiently extracting temporal and spatial structural features of brain regions; furthermore, it constructs a multi-dimensional adaptive denoising model, generates a heatmap of brain region noise types, and uses a signal-noise separation algorithm to form a graded differentiated denoising strategy to perform adaptive denoising processing on the original data; simultaneously, it evaluates the image signal-to-noise ratio and the fidelity of neural signals in brain regions, generates a quality assessment report, and dynamically updates the model parameter weights, which can significantly improve the accuracy of denoising and the fidelity of neural signals in children's functional magnetic resonance imaging (fMRI) data, better adapt to the physiological characteristics of children's brain development, effectively suppress various noise interferences, provide high-quality data support for the analysis of children's brain functional imaging data, and help improve the accuracy and reliability of research related to children's brain development. Thus, it solves the problem of insufficient denoising accuracy in existing technologies.

[0015] Additional aspects and advantages 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. Attached Figure Description

[0016] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a method for denoising functional magnetic resonance image data according to an embodiment of this application; Figure 2 This is a schematic diagram of a method for denoising functional magnetic resonance image data according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a denoising system for functional magnetic resonance imaging data according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0017] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0018] The following describes a method, system, device, and medium for denoising functional magnetic resonance imaging (fMRI) data according to embodiments of this application, with reference to the accompanying drawings. Addressing the issue of insufficient denoising accuracy mentioned in the background art, this application provides a method for denoising fMRI image data. This method integrates prior brain development atlases for children, the gold standard for denoising within the same age group, and synchronously acquired physiological motion data to perform precise segmentation and region of interest localization of brain tissue regions, efficiently extracting temporal and spatial structural features of brain regions. Furthermore, a multi-dimensional adaptive denoising model is constructed, generating a heatmap of brain region noise types and employing a signal-noise separation algorithm to form a graded differentiated denoising strategy, performing adaptive denoising processing on the original data. Simultaneously, the image signal-to-noise ratio and the fidelity of neural signals in brain regions are evaluated, generating a quality assessment report and dynamically updating the model parameter weights. This significantly improves the accuracy of denoising and the fidelity of neural signals in children's fMRI data, better adapting to the physiological characteristics of children's brain development, effectively suppressing various noise interferences, providing high-quality data support for the analysis of children's brain functional imaging data, and helping to improve the accuracy and reliability of research related to children's brain development. Thus, it solves the problem of insufficient denoising accuracy in the prior art.

[0019] Specifically, Figure 1 This application provides a method for denoising functional magnetic resonance imaging data.

[0020] like Figure 1 As shown, the denoising method for functional magnetic resonance imaging data includes the following steps: In step S101, the raw time-series image data of the child's functional magnetic resonance imaging (fMRI), the child's brain development prior atlas data, the historical gold standard data for the same age group after noise reduction, and the synchronously acquired physiological and motor data are obtained.

[0021] It is understood that the acquisition of raw time-series images of children's brain functional magnetic resonance imaging (fMRI), prior atlas data of children's brain development, historical gold standard data for denoising at the same age, and synchronously acquired physiological motion data in the embodiments of this application are the basic prerequisites for the entire process of denoising images of children's brain functional magnetic resonance imaging. This provides comprehensive data support that is adapted to the characteristics of children's brain development for subsequent brain tissue region segmentation, region of interest localization, construction of multi-dimensional adaptive denoising models, signal-noise separation, and evaluation of denoising effects. The developmental prior atlas ensures the accuracy of brain region segmentation and localization, the denoising model is calibrated based on historical gold standard data, and the physiological motion data is used to distinguish neural signals from various noise components. The raw time-series data is used as the object of denoising processing and the benchmark for effect comparison. At the same time, it provides a reliable basis for updating the parameters of the denoising model and provides high-quality image data support for children's brain-related medical research and clinical diagnosis.

[0022] In step S102, based on the original temporal image data of the child's functional magnetic resonance imaging and the prior atlas data of the child's brain development, the brain tissue region segmentation and region of interest localization are completed, the localization results of different brain regions are obtained, and the temporal signal features and spatial structural features of different brain regions are extracted.

[0023] Brain tissue region segmentation refers to the technical process of dividing brain medical images into different anatomical structures and tissue regions through image analysis and algorithm processing.

[0024] It is understood that the brain tissue region segmentation in this application can accurately delineate the boundaries of non-brain tissues such as gray and white matter, cerebrospinal fluid, and skull in children's brains, clarify the spatial location and extent of each functional brain region, and provide accurate spatial reference for subsequent region of interest localization. At the same time, it can separate the temporal signals and spatial structural features of different brain regions, making subsequent signal-noise separation more targeted, avoiding artifacts of non-brain tissues and mutual interference of noise from different brain regions, ensuring the integrity and accuracy of neural signals in each brain region, providing precise support for the construction of multi-dimensional adaptive denoising models, analysis of brain region noise types, and formulation of graded differentiated denoising strategies, ensuring that the denoising process effectively removes noise without damaging brain region neural signals, and providing a reliable brain region structure and functional basis for the accurate interpretation of functional magnetic resonance imaging of children's brains and related medical research and clinical diagnosis.

[0025] For example, using raw time-series images of functional magnetic resonance imaging (fMRI) of a 3-year-old male child's brain, combined with a priori atlas of brain development in 3-year-old children, a brain tissue segmentation and brain region localization model was constructed. An improved 3DU-Net network was used to complete brain tissue region segmentation, with a segmentation volume of approximately 420 cm³ for gray and white matter, approximately 180 cm³ for cerebrospinal fluid, and approximately 250 cm³ for skull and non-brain tissue. The segmentation accuracy reached 96.8%. The improved 3DU-Net network, by introducing an attention gating mechanism, effectively suppressed interference from non-brain tissue regions, improving the clarity of the segmentation boundaries between gray and white matter and cerebrospinal fluid. Simultaneously, a graph attention network was used to capture the spatial correlation features of each functional brain region, completing the localization of 12 core regions of interest, including the prefrontal cortex, hippocampus, and cerebellum. The localization error was controlled within 0.3 mm, and the overlap of each brain region's localization reached 95.3%, ultimately obtaining clear localization results for each brain region. Simultaneously, features such as the temporal series fluctuation amplitude of the temporal signals of each brain region and the standard deviation of the gray-scale distribution of the spatial structure were extracted.

[0026] In this embodiment, brain tissue region segmentation and region of interest localization are completed to obtain localization results for different brain regions. Temporal signal features and spatial structural features of different brain regions are extracted, including: constructing a pediatric brain tissue segmentation and brain region localization model; based on the pediatric brain tissue segmentation and brain region localization model, combined with age-matched brain region anatomical structure and functional zoning features in the pediatric brain development prior atlas data, using an improved 3DU-Net network to complete the region segmentation of gray and white matter, cerebrospinal fluid, and non-brain tissues of the skull, and combining graph attention network to capture the spatial association features of each functional brain region to complete the region of interest localization and obtain localization results for different brain regions; based on the localization results of different brain regions, the temporal sequence fluctuation features and frequency domain features of functional magnetic resonance signals of each brain region are extracted as temporal signal features, and the anatomical structure boundary features, gray-level distribution features, and spatial texture features of brain regions are extracted as spatial structural features.

[0027] Among them, the pediatric brain tissue segmentation and brain region localization model is a computational model used to automatically segment brain tissue in pediatric brain medical images and accurately identify and locate the spatial location of each brain region.

[0028] The formula for the segmentation and localization model of pediatric brain tissue is as follows: ; ; ; in, The convolution outputs a feature map; This is a 3D convolution operation; Input feature map; These are bias parameters; For node attention weights; This is the normalization function; For activation functions; for The transpose of ; It is a linear transformation matrix; For nodes eigenvectors; For nodes eigenvectors; For feature splicing operations; This is the total loss function; Cross-entropy loss; This is the loss balance coefficient; This is a loss for Dice.

[0029] It is understood that the pediatric brain tissue segmentation and brain region localization model in this application can fully combine the age-matched anatomical structure and functional zoning features of brain regions in the prior atlas data of pediatric brain development. It accurately completes the regional segmentation of gray and white matter, cerebrospinal fluid, and non-brain tissues of the skull through the improved 3DU-Net network. At the same time, it captures the spatial correlation features of each functional brain region with the help of graph attention network, realizes the accurate localization of regions of interest, and clarifies the spatial location and range of different brain regions. This provides reliable support for the accurate extraction of temporal signal features and spatial structural features of each brain region, effectively avoids non-brain tissue artifacts and mutual interference of signals from different brain regions, and ensures the pertinence and accuracy of subsequent signal-noise separation, brain region noise type analysis, and graded differentiated denoising strategy formulation. It adapts to the specificity of pediatric brain development and provides core technical support for the efficient advancement and improvement of the denoising effect of the entire pediatric brain functional magnetic resonance imaging denoising process. It provides accurate brain region structure and functional localization basis for subsequent medical image interpretation, brain development research, and clinical diagnosis.

[0030] It should be noted that by combining graph attention networks to capture the spatial association features of each functional brain region, the region of interest (ROI) is located, and the localization results of different brain regions are obtained. After the brain tissue regions are segmented based on the improved 3DU-Net network, each functional brain region is constructed as a graph structure node. The spatial topological relationship and functional connectivity features of each node are modeled using graph attention networks. The spatial association weights and interaction features between different brain regions are learned through the attention mechanism, accurately capturing the spatial dependencies of key functional brain regions, thereby completing the ROI localization and finally obtaining the localization results of different brain regions.

[0031] The temporal fluctuation features and frequency domain features of functional magnetic resonance imaging (fMRI) signals from each brain region are extracted as temporal signal features. The anatomical boundary features, gray-level distribution features, and spatial texture features of brain regions are extracted as spatial structural features. Based on the localization results of each brain region, the temporal fluctuation features of the fMRI signals from each brain region are extracted through sliding window temporal domain analysis, and the frequency domain features are extracted by combining Fourier transform and wavelet analysis, thus integrating them to form complete temporal signal features. At the same time, edge detection, gray-level statistics, and texture modeling are performed on the spatial images of brain regions to accurately extract anatomical boundary features, gray-level distribution, and spatial texture features, thus comprehensively completing the extraction of spatial structural features.

[0032] For example, a model for segmenting and locating brain regions in children was constructed. Based on an improved 3DU-Net network, it features a 5-layer encoding structure and a 5-layer decoding structure. Feature extraction is performed using 3×3×3 convolutional kernels, coupled with a graph attention network with four attention heads. Inputting raw functional magnetic resonance imaging (fMRI) images of 3-year-old children and corresponding developmental prior maps, the model completes brain tissue segmentation and region of interest (ROI) localization. The model is trained using a combination of cross-entropy and Dice loss functions. 200 cases of annotated MRI data from 3-5-year-old children were selected as the training set, and 50 cases as the validation set. Data augmentation techniques such as random flipping and Gaussian blur were used to improve the model's generalization ability. The average segmentation accuracy on the validation set reached 96.7%. The Dice coefficient for gray and white matter segmentation reached 0.972, the Dice coefficient for cerebrospinal fluid segmentation reached 0.958, and the Dice coefficient for skull and non-brain tissue segmentation reached 0.981. It successfully located 15 core functional brain regions, including the prefrontal cortex, hippocampus, thalamus, and cerebellum. The overlap of each brain region was higher than 95.5%, and the localization error was stable within 0.28 mm. The model inference speed was 12.3 seconds per image, which effectively adapted to the characteristics of children's immature brain development and relatively blurred brain region boundaries. It accurately completed the tissue segmentation and brain region localization tasks, providing reliable basic data support for subsequent multi-dimensional adaptive denoising.

[0033] In step S103, based on historical age-matched gold standard data for denoising, synchronous physiological motion data, and time-series signal and spatial structural features, a multi-dimensional adaptive denoising model for functional magnetic resonance imaging (fMRI) of children's brain is constructed. Combined with the localization results of different brain regions, a heatmap of brain region noise types is generated. Through a signal-noise separation algorithm, the distinction between neural signals and noise components in each brain region is obtained, a brain region differentiation-based denoising strategy is generated, and adaptive denoising processing is performed on the original time-series fMRI image data to obtain denoised fMRI image data.

[0034] Among them, the multi-dimensional adaptive denoising model for functional magnetic resonance imaging of children's brain is a denoising analysis model that integrates multi-dimensional noise features and has adaptive optimization capabilities for functional magnetic resonance imaging data of children's brain. It aims to improve the signal-to-noise ratio of data and the accuracy and reliability of brain functional activity analysis.

[0035] The formula for the multidimensional adaptive denoising model of functional magnetic resonance imaging (fMRI) of the brain in children is as follows: ; ; ; ; ; ; ; ; in, Total loss due to brain region segmentation; Cross-entropy loss; This is the loss balance coefficient; For Dice's loss; This represents the grid noise density value of the brain region. This represents the number of noise feature samples. The kernel function bandwidth; The Gaussian kernel function; This is the current noise feature vector; For the first One noise feature sample; This is the vector of signal-noise separation results; The separation matrix; This is the original observed signal vector; The percentage of noise components in a single brain region; This is a noise signal vector; It is a neural signal vector; This is a denoised functional magnetic resonance imaging (fMRI) image; Adaptive denoising weights for brain regions; Original functional magnetic resonance imaging (fMRI) image; This serves as the gold standard reference image for the same age group. Signal-to-noise ratio of brain region images; This represents the average neural signal value in the brain region. Standard deviation of brain region noise; It is an image structure similarity index; The average pixel value of the original image; The average pixel value of the denoised image; This is a constant used to maintain numerical stability in brightness contrast calculations; The covariance between the original and denoised images; The standard deviation of the original image; Standard deviation of the denoised image; This is a constant used to maintain numerical stability in contrast calculations. The total loss of the multi-dimensional adaptive denoising model; Image denoising and reconstruction loss; This is the signal fidelity weighting coefficient; For loss of fidelity of neural signals; These are the regularization weight coefficients; This is the regularization loss for the model parameters.

[0036] It is understood that the multi-dimensional adaptive denoising model for pediatric brain functional magnetic resonance imaging (fMRI) in this application can integrate historical gold standard data for denoising at the same age, synchronously acquired physiological motion data, and temporal signal characteristics and spatial structural characteristics of each brain region. It adapts to the specificity of children's brain development, generates accurate heatmaps of brain region noise types by combining the localization results of different brain regions, and clearly distinguishes neural signals and noise components in each brain region through a signal-noise separation algorithm. Then, it formulates a graded and differentiated denoising strategy for brain regions, achieving targeted adaptive denoising of the original temporal image data. This can effectively avoid the problems of excessive denoising damaging neural signals or incomplete denoising leaving artifacts in traditional denoising methods. At the same time, it provides reliable model support for subsequent denoising quality assessment. The denoising effect is continuously optimized through parameter weight updates, ensuring that the denoised image has both a high signal-to-noise ratio and completely preserves the physiological characteristics of neural signals in each brain region. This provides high-quality technical support for the accurate analysis of pediatric brain fMRI images, brain development research, and clinical diagnosis.

[0037] It should be noted that the construction of the multidimensional adaptive denoising model of pediatric brain functional magnetic resonance imaging (fMRI) was based on historical gold standard data for denoising of the same age group, synchronous physiological motion data, and extracted temporal signal features and spatial structural features of each brain region. After the model was built, the specific location and extent of each brain region were determined by combining the localization results of different brain regions. Then, based on this localization information, the noise density value and noise type confidence of each brain region grid unit were calculated by the kernel density estimation algorithm, generating a heatmap of brain region noise type labeled with noise-related information, which provides support for subsequent differentiated denoising.

[0038] A graded and differentiated denoising strategy for brain regions was generated, and adaptive denoising was performed on the original time-series functional magnetic resonance imaging (fMRI) images to obtain denoised fMRI images. When generating the graded and differentiated denoising strategy for brain regions, based on the heatmap of noise types in brain regions and the differentiation results of neural signals and noise components in each brain region, combined with the proportion of noise components in each brain region and the preset noise threshold for the corresponding brain region, differentiated denoising parameters and denoising intensities were formulated for different noise types and density levels in different brain regions. Subsequently, based on this graded and differentiated denoising strategy and combined with the constructed multi-dimensional adaptive denoising model of pediatric brain fMRI, adaptive denoising was performed on each brain region in the original time-series fMRI images to accurately separate neural signals and noise components in each brain region, preserving the integrity and fidelity of neural signals, and finally obtaining denoised fMRI images.

[0039] For example, the constructed multi-dimensional adaptive denoising model for pediatric functional magnetic resonance imaging (fMRI) was trained based on the gold standard data of denoising from 200 historical cases of 3-5 year old children's fMRI. It integrates brain region developmental features from prior brain development atlases, respiratory and heart rate interference data collected synchronously with physiological movements, and temporal signal features and spatial structural features of each brain region extracted previously. The model's main body is constructed using a convolutional neural network combined with an attention mechanism, and includes three feature fusion modules and two adaptive adjustment modules. The model takes raw temporal fMRI images of 3-year-old children as input, combines them with the localization results of 15 core brain regions such as the prefrontal cortex and hippocampus, and generates a heatmap of brain region noise types. A signal-noise separation algorithm is used to accurately separate neural signals from various types of noise, achieving an accuracy rate of 97.3% for motion noise separation, 96.8% for physiological noise separation, and 98.1% for instrument noise separation. Before denoising, the average signal-to-noise ratio of the whole brain in the original image was 18.2 dB, which was improved to 32.5 dB after denoising. The fidelity of neural signals in each brain region was higher than 95.7%. The model was trained iteratively for 200 rounds with a learning rate of 0.001, and the generalization error was controlled within 3.2%. The denoising intensity of each brain region was automatically adjusted to take into account the differences in brain regions at different developmental stages of children's brains. Low-intensity denoising was used for key developmental brain regions such as the hippocampus to preserve subtle neural signals, while high-intensity denoising was used for noise-concentrated areas such as the periphery of the skull. The final output was clear denoised and neural signal-fidelity functional magnetic resonance imaging data of children's brains. At the same time, a denoising parameter report for each brain region was generated for dynamic updating of model parameter weights.

[0040] In this embodiment, generating a brain region noise type heatmap includes: constructing a regional spatial grid partitioning model adapted to the characteristics of children's brain development; based on the regional spatial grid partitioning model adapted to the characteristics of children's brain development, combined with the brain region partitioning boundaries and key developmental region annotations in the prior brain development atlas data of children, as well as the noise fluctuation amplitude in the temporal signal characteristics and the artifact distribution characteristics in the spatial structure characteristics, calculating the noise density value and noise type confidence of each grid unit through a kernel density estimation algorithm; mapping the noise density value and noise type confidence to a preset color gradient interval, superimposing it onto the three-dimensional brain region template of the prior brain development atlas of children, and generating a brain region noise type heatmap annotating the noise type, density level, confidence, and update timestamp of each grid unit.

[0041] Among them, the regional spatial grid division model adapted to the characteristics of children's brain development is a special model that performs refined spatial grid division and modeling of brain regions based on the structural and functional characteristics of children's brains at different developmental stages.

[0042] The formula for a regional spatial grid partitioning model adapted to the characteristics of children's brain development is as follows: ; ; ; in, This represents the grid noise density value of the brain region. This represents the number of noise feature samples. The kernel function bandwidth; The Gaussian kernel function; This is the current noise feature vector; For the first One noise feature sample; Confidence level for noise type; This is the transpose of the weight parameter vector; The input vector is the grid feature. It is a normalized exponential function; Heatmap of noise types in brain regions; This is the color gradient mapping function; This is an estimate of the noise density; This represents the minimum noise density. This represents the maximum noise density.

[0043] It is understood that the embodiments of this application are adapted to the regional spatial grid division model of children's brain development characteristics. Combining the brain region boundaries and developmental key area annotations in the prior map of children's brain development, and integrating the noise fluctuation amplitude of time-series signals and the artifact distribution characteristics of spatial structures, the noise density and type confidence of grid cells are quantified through kernel density estimation algorithm, and three-dimensional brain region noise heat map is generated by mapping colors. This can finely characterize the spatial distribution, type and level of brain region noise, providing accurate spatial positioning basis for subsequent signal-noise separation and graded differential denoising. It is adapted to the specificity of children's brain development, improves the fineness of noise identification and characterization, and ensures the pertinence and effectiveness of denoising processing.

[0044] It should be noted that, by combining the brain region boundaries and key developmental region annotations in the prior brain development atlas data of children, as well as the noise fluctuation amplitude in the temporal signal features and the artifact distribution characteristics in the spatial structure features, the brain region boundaries and key developmental region annotation information are accurately extracted from the prior brain development atlas data of children. At the same time, by combining the noise fluctuation amplitude in the temporal signal features and the artifact distribution characteristics in the spatial structure features, the noise feature quantification analysis of each grid cell is carried out through the kernel density estimation algorithm, and the corresponding noise density value and noise type confidence are calculated. This provides accurate data support and spatial positioning basis for the subsequent generation of brain region noise type heatmaps and the formulation of differentiated denoising strategies.

[0045] The noise density value and noise type confidence level are mapped to a preset color gradient range and superimposed onto a 3D brain region template of the child's brain development prior map. This generates a brain region noise type heatmap labeled with the noise type, density level, confidence level, and update timestamp of each grid unit. The calculated noise density value and noise type confidence level of each grid unit are mapped one by one to a preset color gradient range, and different color systems such as blue, green, yellow, and red are divided according to the noise intensity from low to high, accurately corresponding to different density levels and confidence levels. Then, a high-precision spatial registration algorithm is used to accurately superimpose the color-coded grid information onto the 3D brain region template of the child's brain development prior map, ensuring that the spatial position is aligned without deviation. Finally, a brain region noise type heatmap with clear labels of noise type, density level, confidence level, and real-time update timestamp of each grid unit is generated.

[0046] For example, a regional spatial grid partitioning model adapted to the characteristics of children's brain development was constructed. Based on the three-dimensional anatomical space of a 3-year-old child's brain, and combined with the annotation locations of key developmental brain regions such as brain region boundaries, hippocampus, and visual cortex in the prior brain development atlas, the entire brain space was divided into cubic grid units with a side length of 2 mm, totaling approximately 128,000 grid units, completely covering all core brain regions such as the cerebral cortex, subcortical nuclei, and cerebellum. Based on this grid model, the noise fluctuation amplitude in temporal signal features and the artifact distribution density in spatial structural features were integrated. The noise density value and noise type confidence of each grid unit were calculated using a kernel density estimation algorithm. Among them, the noise density of motion artifact grid units was concentrated between 0.15 and 0.25 with a confidence of 89.2%, the noise density of physiological noise grid units was between 0.08 and 0.15 with a confidence of 85.7%, and the noise density of instrument noise grid units was below 0.08 with a confidence of 91.3%. The noise density value and confidence level are mapped to a blue-yellow-red gradient color range and superimposed on a standardized three-dimensional brain region template to generate a brain region noise type heatmap labeled with the noise type, density level, confidence level and update timestamp of each grid unit. The overall consistency between grid division and noise feature mapping reaches 94.6%, realizing a refined and visual representation of the spatial noise characteristics of the whole brain of children.

[0047] In this embodiment, a signal-noise separation algorithm is used to obtain the distinction between neural signals and noise components in each brain region. This includes: constructing a brain region-specific signal-noise separation model; based on the brain region-specific signal-noise separation model, using the sliding window method to extract the non-stationary fluctuation features of the temporal signals in each brain region, and using independent component analysis combined with the gradient boosting tree algorithm to complete the separation of neurophysiological signals from motion noise, physiological noise, and instrument noise, and outputting the distinction between neural signals and noise components in each brain region; based on the distinction between neural signals and noise components in each brain region, obtaining the noise component ratio of the corresponding brain region; when the noise component ratio exceeds the preset noise threshold of the corresponding brain region, triggering an adaptive weighted denoising mechanism for the brain region to generate targeted denoising parameter correction instructions.

[0048] Among them, the brain region-specific signal-noise separation model is a neural data analysis model that focuses on a specific brain region and accurately separates the effective neural signals from irrelevant noise.

[0049] It is understood that the brain region-specific signal-noise separation model in this application can extract the non-stationary fluctuation characteristics of temporal signals in each brain region. By combining independent component analysis with gradient boosting tree algorithm, it can accurately separate the components of neurophysiological signals from motion noise, physiological noise, and instrument noise, clarify the distinction between neural signals and noise in each brain region, quantify the proportion of noise components and trigger an adaptive weighted denoising mechanism, and specifically correct denoising parameters to effectively improve the accuracy of noise separation. While removing various types of noise, it can preserve the physiological characteristics of neural signals in children's brain regions to the greatest extent, providing a reliable basis for the formulation of graded differentiated denoising strategies.

[0050] It should be noted that the noise component proportion of the corresponding brain region is obtained. When the proportion of the noise component exceeds the preset noise threshold of the corresponding brain region, the brain region adaptive weighted denoising mechanism is triggered to generate targeted denoising parameter correction instructions. Based on the differentiation results of neural signals and noise components in each brain region, the noise component proportion of the corresponding brain region is calculated by quantifying the signal amplitude proportion of noise components in each brain region. This proportion is compared one by one with the preset noise thresholds for each brain region based on the characteristics of children's brain development and historical gold standard data. When the noise component proportion exceeds the corresponding threshold, the brain region adaptive weighted denoising mechanism is automatically triggered to dynamically adjust the denoising weight and filtering intensity of the brain region, and generate targeted denoising parameter correction instructions based on the noise intensity and type.

[0051] Gradient boosting tree algorithm formula: ; ; in, For the first The predicted output for each sample; This represents the total number of decision trees; For the first Decision tree model; For the first The input feature vector of each sample; The sequence number of the decision tree; This is the total loss function of the model; The total number of training samples; For the first The true label of each sample; The regularization coefficient is used. For the first The complexity of a decision tree is determined by a regularization term. This is the sample number.

[0052] For example, a brain region-specific signal-noise separation model was constructed. This model uses a sliding window method to extract the non-stationary fluctuation characteristics of temporal signals from each brain region. Combined with independent component analysis (ICA) and gradient boosting tree algorithm, it was used to separate neural signals from various types of noise in functional magnetic resonance imaging (fMRI) data from 3-year-old children. The model was trained and optimized using 200 cases of data from the same age group. The sliding window length was set to 30 time points, and the step size to 5 time points. ICA extracted 62 independent components, of which 21 were real neurophysiological activity signals, and 41 were motion noise, physiological noise, and instrument noise. The gradient boosting tree algorithm achieved a 97.8% accuracy rate in classifying noise components and a 96.5% accuracy rate in distinguishing between signal and noise components. The model outputs the separation results of signals and noise in each brain region. The statistics show that the noise component accounts for 18.3% in the prefrontal cortex, 12.7% in the hippocampus, and 15.6% in the thalamus. When the noise content exceeds the preset threshold of 20%, an adaptive weighted denoising mechanism is automatically triggered to generate targeted denoising parameter correction instructions, thereby achieving graded and accurate signal-noise separation in different brain regions.

[0053] In this embodiment of the application, the signal-noise separation algorithm formula is as follows: ; ; ; ; ; ; ; ; in, A matrix of raw time-series images from functional magnetic resonance imaging (fMRI) of the brain in children; A design matrix that includes both neural networks and noise; This is the regression coefficient matrix; The neural signal is in residual form; For the first Brain region spatial mask matrix; For the first Brain region neural signal regression matrix; These are the regression coefficients of the neural signal; This is the noise reference regression matrix; The noise regression coefficient; For the first residual neural signals in brain regions; It is a signal mixing matrix; It is a set of independent components; This is the total noise matrix; for Time-major filtering weight vector; for Time-weight vector; This is the step size parameter; To prevent zero constant; for Timing error signal; for Time-of-flight noise reference vector; The estimated neural signal matrix; To find the optimal solution operator; For regularization weights; For the development of the prior space regularization matrix; These are real neural signals; These are the wavelet coefficients after thresholding. The sign function for wavelet coefficients; These are the original wavelet coefficients; For the first Noise-reducing neural signals in brain regions; For the first Adaptive brain region separation operator; This is the physiological motion noise reference matrix; A priori atlas of child development; These are the model weight parameters; This is the denoised functional magnetic resonance imaging data.

[0054] Understandably, the real-time example in this application extracts the non-stationary fluctuation features of temporal signals from each brain region, and combines independent component analysis and gradient boosting tree algorithm to achieve accurate separation of neural signals from various types of noise such as motion, physiology, and instrumentation. It clarifies the distinction between neural signals and noise components in each brain region, quantifies the proportion of noise components, and triggers an adaptive weighted denoising mechanism. This effectively removes noise interference while preserving the physiological characteristics of neural signals in brain regions to the greatest extent. It provides a core basis for the formulation and implementation of graded differentiated denoising strategies and significantly improves the pertinence and accuracy of the denoising process.

[0055] For example, based on a brain region-specific signal-noise separation model, non-stationary fluctuation features of temporal signals in each brain region are extracted. A sliding window length of 30 time points and a step size of 5 time points are set. A signal-noise separation algorithm combined with independent component analysis is used to extract 62 independent components. Then, a gradient boosting tree algorithm is used to classify the components, accurately distinguishing neurophysiological signals from various types of noise. During processing, the algorithm achieves an accuracy rate of 97.2% for motion noise separation, 96.5% for physiological noise separation, and 98.3% for instrument noise separation. The overall signal-to-noise component differentiation accuracy reaches 96.8%, effectively identifying and removing interference signals from different sources. Based on this, the statistical results of the noise component proportion in each brain region are obtained. Cerebellar noise accounts for 21.2%, exceeding the preset threshold. The algorithm automatically triggers an adaptive weighted denoising mechanism for the brain region and generates correction instructions, avoiding the loss of neural signals while improving noise removal efficiency, achieving accurate separation of neural signals and noise.

[0056] In step S104, based on the denoised functional magnetic resonance imaging data, image signal-to-noise ratio evaluation data and brain region neural signal fidelity data are calculated simultaneously to generate a denoising quality evaluation report and update the parameter weights of the multi-dimensional adaptive denoising model.

[0057] Among them, the denoising quality assessment report refers to the technical document or process that systematically analyzes the denoising algorithm to preserve the original signal details and edge effects while eliminating noise through quantitative indicators and qualitative observation.

[0058] Understandably, the real-time example in this application quantifies and accurately evaluates the denoising effect by simultaneously calculating image signal-to-noise ratio assessment data and brain region neural signal fidelity data. This directly reflects the degree to which denoising processing preserves image quality and neurophysiological characteristics, providing a scientific basis for subsequent parameter weight updates of multi-dimensional adaptive denoising models. It continuously optimizes denoising strategies and effects, ensuring that denoising processing effectively suppresses noise and improves image quality while fully protecting the authenticity and integrity of neural signals in children's brain regions. This provides reliable quality assurance and data support for subsequent analysis of functional magnetic resonance imaging of children's brains, brain development research, and clinical diagnosis.

[0059] It should be noted that the generation of the denoising quality assessment report and the updating of the parameter weights of the multi-dimensional adaptive denoising model are based on the constructed dual-dimensional quantitative assessment model of the denoising effect of functional magnetic resonance imaging in children's brains. Combining the image data before and after denoising, the gold standard data of the same age group in history and the brain region localization results, the signal-to-noise ratio of the whole brain and each brain region, structural similarity and other image signal-to-noise ratio assessment data, as well as the brain region neural signal fidelity data such as neural signal temporal correlation and functional connectivity strength fidelity, are calculated. The two types of assessment data are matched and associated by brain region to generate the denoising quality assessment report. Then, the parameter weights of the multi-dimensional adaptive denoising model are adaptively adjusted and updated according to the assessment results.

[0060] For example, based on a two-dimensional quantitative evaluation model of denoising effects in pediatric brain functional magnetic resonance imaging (fMRI), a comprehensive calculation and analysis of image data and neural signal characteristics before and after denoising was performed. The results showed that the average signal-to-noise ratio (SNR) of the original whole-brain images was 18.2 dB, which increased to 32.5 dB after denoising, with a 79.3% increase in the peak SNR and a structural similarity index consistently above 0.92. The temporal correlation of each core brain region was higher than 0.94, the fidelity of low-frequency oscillation amplitude remained above 95.7%, and the fidelity of functional connectivity strength reached 94.2%. The evaluation report clearly marked the quantitative indicators for the whole brain and 15 brain regions, including the prefrontal cortex, hippocampus, and thalamus, demonstrating a denoising quality compliance rate of 96.3%. It also detailed key indicators such as noise residue ratio and signal distortion rate, and completed cross-comparison with historical gold standard data for the same age group. Finally, based on the evaluation results, the parameter weights of the multi-dimensional adaptive denoising model were automatically updated, dynamically adjusting the denoising coefficients of different brain regions to achieve continuous optimization and iterative improvement of model performance.

[0061] In this embodiment, the simultaneous calculation of image signal-to-noise ratio (SNR) assessment data and brain region neural signal fidelity data includes: constructing a two-dimensional quantitative assessment model for the denoising effect of functional magnetic resonance imaging (fMRI) in children's brains; based on the two-dimensional quantitative assessment model for the denoising effect of fMRI in children's brains, combined with denoised fMRI image data, original temporal fMRI image data of children's brains, historical gold standard data for denoising at the same age, and localization results of different brain regions, calculating the SNR, peak SNR, and structural similarity index of the whole brain and each brain region using a medical image SNR statistical algorithm to obtain image SNR assessment data; simultaneously calculating the temporal correlation, low-frequency oscillation amplitude fidelity, and functional connectivity strength fidelity of neural signals in each brain region using a neurophysiological feature matching algorithm to obtain brain region neural signal fidelity data; performing brain region matching and association between the image SNR assessment data and the brain region neural signal fidelity data, and outputting a two-dimensional quantitative assessment dataset with brain region labels.

[0062] Among them, the dual-dimensional quantitative evaluation model of denoising effect of functional magnetic resonance imaging of children's brain is an analytical model that conducts quantitative evaluation of the denoising effect of functional magnetic resonance imaging data of children's brain from two different dimensions.

[0063] The formula for a two-dimensional quantitative evaluation model of the noise reduction effect of functional magnetic resonance imaging (fMRI) in children's brains is as follows: ; ; ; ; ; ; ; ; ; ; in, brain region Signal-to-noise ratio; brain region Mean value of internal fMRI signal; brain region Standard deviation of the corresponding background noise region; The mean square error of the images before and after denoising; The length of the image in pixels; The width of the image in pixels; These are the pixel values ​​of the original fMRI image; These are the pixel values ​​of the denoised fMRI image; The horizontal pixel coordinates of the image; The vertical pixel coordinates of the image; Peak signal-to-noise ratio; The maximum grayscale value of a pixel in an fMRI image; Logarithmic operations with base 10; The structural similarity index of the images before and after denoising; The average pixel value of the original image; The average pixel value of the denoised image; This is a constant used to maintain numerical stability in brightness contrast calculations; The covariance between the original and denoised images; The standard deviation of the original image; Standard deviation of the denoised image; This is a constant used to maintain numerical stability in contrast calculations. The correlation coefficient of temporal signals in brain regions; Covariance is a covariance operator used to measure the degree of linear correlation between two variables. The standard deviation of the original signal; The standard deviation of the denoised signal; The original temporal neural signals from functional magnetic resonance imaging of the brain in children; The denoised functional magnetic resonance imaging (fMRI) sequence of neural signals in children's brains; For the signal at frequency Power spectral density at; For signal Fast Fourier Transform; To preserve the fidelity of low-frequency oscillation amplitude; The low-frequency power spectral density of the denoised signal; The low-frequency power spectral density of the original signal; The low-frequency frequency of the fMRI signal; This represents the strength of functional connectivity between brain regions; Pearson correlation coefficient between the temporal signals of brain region A and brain region B; This represents the time-series fMRI signal of brain region A. This represents the time-series fMRI signal of brain region B. For functional connection strength fidelity; Functional connectivity strength of the denoised data; The functional connection strength of the original data; brain region The denoising effect of the dataset was evaluated from two dimensions. brain region The correlation coefficient of the time-series signal; brain region Low-frequency oscillation amplitude fidelity; brain region Functional connection strength fidelity.

[0064] Understandably, this application's real-time example quantifies multiple signal-to-noise ratio indicators for the whole brain and individual brain regions using a medical image signal-to-noise ratio statistical algorithm. Simultaneously, it relies on a neurophysiological feature matching algorithm to accurately assess the fidelity of neural signals in terms of temporal sequence, low-frequency oscillations, and functional connectivity. This allows for brain region-specific, two-dimensional quantitative evaluation of the denoising effect, providing a scientific quantitative basis for parameter weight optimization and denoising strategy iteration in a multi-dimensional adaptive denoising model. It comprehensively ensures the quality of denoised images and the integrity of neural signal physiological characteristics, providing reliable quality evaluation support for clinical diagnosis and developmental research of pediatric brain functional magnetic resonance imaging.

[0065] It should be noted that the output of the two-dimensional quantitative evaluation dataset with brain region labels is achieved by first classifying, matching and associating the image signal-to-noise ratio evaluation data and the brain region neural signal fidelity data according to different brain regions, labeling each type of evaluation result with the corresponding brain region label, and integrating them to form a two-dimensional quantitative evaluation dataset with clear brain region labels and containing quantitative indicators for each brain region.

[0066] For example, based on a dual-dimensional quantitative evaluation model of denoising effects in pediatric brain functional magnetic resonance imaging (fMRI), a comprehensive quantitative analysis was conducted on the imaging and neural signal data of 3-year-old children before and after denoising. Using a medical image signal-to-noise ratio (SNR) statistical algorithm, the average SNR of the whole brain in the original images was 18.2 dB, which increased to 32.5 dB after denoising. The peak SNR of the whole brain improved by 79.3%, and the structural similarity index remained stable above 0.92. Simultaneously, a neurophysiological feature matching algorithm was used to calculate signal indicators for each brain region. The temporal correlation of each core brain region was higher than 0.94, the fidelity of low-frequency oscillation amplitude remained above 95.7%, and the fidelity of functional connectivity strength reached 94.2%. The evaluation report fully annotated the quantitative evaluation data of the whole brain and each functional brain region. The overall denoising quality compliance rate was 96.3%. The evaluation data was fed back to a multi-dimensional adaptive denoising model, enabling dynamic updates and optimization of model parameter weights, further improving the model's adaptability to brain data from children at different developmental stages and its denoising stability.

[0067] The following will illustrate a method for denoising functional magnetic resonance imaging data through a specific embodiment, such as... Figure 2 As shown, it includes: The study simultaneously acquired four core data categories: raw time-series fMRI images of children's brains, prior brain development atlas data, denoised gold standard data from historical age-matched models, and synchronously acquired physiological and motor data. fMRI data came from resting-state scans of 120 children aged 7-10 years, using a 3.0T MRI scanner with a 64-channel head and neck coil, acquired via multi-bandwidth radiofrequency sequences at a resolution of 1.5×1.5×3.0mm, TR=2000ms, TE=30ms, with 240 time points per subject and a single dataset size of 186MB. The prior brain development atlas was an age-specific unbiased template adapted to children aged 4.5-18.5 years, covering 116 functional brain regions at a resolution of 1×1×1mm. It included anatomical boundaries, developmental maturity grading, and functional regional characteristics, incorporating indicators such as cortical thickness and gray matter volume to match the brain development characteristics of children aged 7-10 years. Historical denoising gold standard data came from 800 age-matched children's fMRI denoised samples. Manual ICA validation showed a classification accuracy of 97.5%, including whole-brain and individual brain region denoising baseline parameters. The mean whole-brain SNR was 28.6±4.2, the mean temporal correlation of neural signals in brain regions was 0.89±0.07, and the mean ALFF was 0.018±0.004. Simultaneous physiological motion data were acquired using synchronous monitoring equipment, including a respiratory rate of 18-22 breaths / min at 50Hz, a heart rate of 70-90 beats / min at 100Hz, and head motion parameters FD values ​​of 0.02-0.35mm. Frames with FD>0.15mm accounted for 12.7%, providing a basis for noise type identification.

[0068] After data acquisition, based on children's fMRI brain data and developmental prior maps, a brain tissue segmentation and region of interest (ROI) localization process was initiated to extract temporal signal features and spatial structural features of brain regions. A children's brain tissue segmentation and region of interest localization model was constructed. This model is based on an improved 3DU-Net network and incorporates a graph attention network (GAT). Addressing the characteristics of children's brains, such as blurred gray-white matter boundaries, small brain regions, and strong developmental heterogeneity, the encoder and decoder structures were optimized. The number of skip connection feature fusion layers was increased to reduce gradient vanishing, and a constraint term based on children's brain development features was introduced to improve segmentation and localization accuracy. Combining age-matched brain region anatomical structures and functional partitioning features, standardized fMRI data was input, and gray-white matter, cerebrospinal fluid, and non-brain tissues of the skull were segmented using the improved 3DU-Net. An adaptive threshold iteration algorithm was used to adjust the segmentation threshold. The Dice coefficients for gray-white matter segmentation were 0.89±0.03, cerebrospinal fluid 0.87±0.04, and non-brain tissues of the skull 0.92±0.02, representing an 8.3% improvement in accuracy compared to the traditional 3DU-Net. GAT captures spatial association features of functional brain regions, with attention weighting coefficients ranging from 0.1 to 0.9. It strengthens the spatial association capture of key developmental brain regions such as the prefrontal cortex and hippocampus, completing the localization of 116 ROIs with a localization error not exceeding 0.3 mm and an accuracy of 96.7% ± 1.2%, higher than the traditional method's 89.5% ± 2.1%. Based on the localization results, temporal signal features and spatial structural features of each brain region are extracted. Temporal signal features include time series fluctuation characteristics (standard deviation 0.02-0.15, coefficient of variation 0.08-0.32, peak frequency 0.01-0.1 Hz), and frequency domain features (FFT) extracts the power spectral density of the low-frequency band (0.01-0.08 Hz) at 68.3% ± 5.7% and the mid-frequency band (0.08-0.15 Hz). Spatial structural features include anatomical boundary features (boundary gradient values ​​5-28), gray-level distribution features (mean gray-level 85-168, standard deviation 12-35, histogram peak value 28-76), and spatial texture features (gray-level co-occurrence matrix extraction: contrast 12.5-48.3, correlation 0.52-0.87, entropy 3.2-5.8). All features are normalized to the [0,1] interval for subsequent model construction.

[0069] After feature extraction, a multi-dimensional adaptive denoising model for pediatric brain fMRI was constructed based on historical gold standard denoising data, synchronously acquired physiological motion data, and extracted temporal and spatial features. This model is based on a deep learning framework. The input layer includes three dimensions: temporal signal features, spatial structural features, and physiological motion features. The hidden layers have six layers with 256, 128, 64, 64, 128, and 256 neurons respectively. The output layer contains the denoised brain region signal features. The model uses an adaptive learning rate of 0.001-0.01 and optimizes the loss function through backpropagation. The loss function is a weighted sum of mean squared error (MSE) and structural similarity loss (SSIM) with weights of 0.6 and 0.4, respectively, ensuring that the original features of the neural signals are preserved while denoising. The model was trained using historical denoised gold standard data as labels and physiological motion data as noise constraints. Input was temporal and spatial features, with 500 iterations and a batch size of 32. The training and validation sets were split in an 8:2 ratio. After training, the validation set loss was 0.008±0.002, and the goodness of fit R²=0.97±0.02, indicating good generalization ability. A brain region noise type heatmap was generated based on brain region localization results. A regional spatial grid partitioning model adapted to the characteristics of children's brain development was constructed. Based on the brain region partitioning boundaries of the developmental prior map, the whole brain was divided into 1024×1024×512 grid units (0.5×0.5×0.5mm). The grid for key developmental brain regions such as the prefrontal cortex, hippocampus, and thalamus was refined to 0.25×0.25×0.25mm. Based on a grid model, combined with the temporal signal noise fluctuation amplitude (0.03-0.21) and spatial structure artifact distribution characteristics (artifact pixel ratio (0.05-0.18)), a kernel density estimation algorithm was used to calculate the noise density value and noise type confidence of each grid cell. The Gaussian kernel function bandwidth was 0.3, with noise density values ​​of 0.05-0.82 and confidence levels of 0.75-0.98. The noise density values ​​and confidence levels were mapped to color gradient ranges: low noise blue (0.05-0.25), medium noise yellow (0.25-0.55), and high noise red (0.55-0.82). This was then overlaid onto a 3D brain region template of a priori atlas of child brain development to generate a heatmap. The noise type, physiological noise, instrument noise, density level, confidence level, and update timestamp of each grid cell were labeled. Motion noise is mainly concentrated in the frontal and parietal lobes, with a noise density value of 0.45-0.78 and a confidence level of 0.85-0.98, caused by head shaking in children; physiological noise is located near the cerebrospinal fluid, with a noise density value of 0.32-0.65 and a confidence level of 0.80-0.95, caused by fluctuations in respiratory and heart rates; instrument noise is evenly distributed, with a noise density value of 0.05-0.22 and a confidence level of 0.75-0.88, caused by uneven magnetic fields.

[0070] After heatmap generation, a signal-noise separation algorithm was used to distinguish neural signals and noise components in each brain region. A differentiated denoising strategy based on brain region classification was then implemented for adaptive denoising. A brain region-specific signal-noise separation model was constructed, employing personalized separation strategies tailored to the noise characteristics of different brain regions in children. This was achieved by combining Independent Component Analysis (ICA) and the XGBoost algorithm. The ICA algorithm initially separated signals and noise. With 40 independent components, FastICA was used, achieving a separation accuracy of 97.3% ± 1.1%, close to the 97.9% accuracy of manual ICA classification. The XGBoost algorithm classified the separated components, distinguishing between neurophysiological signals and noise. With 100 decision trees, a learning rate of 0.05, and a maximum depth of 6, the classification accuracy was 96.8% ± 1.3%. A sliding window method was used to extract non-stationary fluctuation features of temporal signals in each brain region. The window size was 20, with a 50% overlap at each time point. Indicators such as variance, kurtosis, and skewness were extracted to aid in signal-noise separation. ICA combined with XGBoost was used to separate neurophysiological signals from motion noise, physiological noise, and instrument noise components. The output showed that the proportion of neural signal components in each brain region was 64.3%-91.8%, and the proportion of noise components was 8.2%-35.7%. Based on the discrimination results, the proportion of noise components in the corresponding brain regions was calculated, and noise thresholds were preset according to the developmental characteristics and functional importance of different brain regions. The noise threshold for the key developmental brain regions, the prefrontal cortex and hippocampus, was 25%, and for ordinary brain regions, it was 30%. When the proportion of noise components in a brain region exceeded the preset threshold, the brain region's adaptive weighted denoising mechanism was triggered, generating denoising parameter correction instructions, including adjusting the denoising intensity range from 0.3-0.8 and the filtering frequency range to 0.01-0.12Hz. Adaptive denoising was performed on the raw fMRI data based on a graded and differentiated denoising strategy for brain regions. For low-noise brain regions (noise density 0.05-0.25), a mild denoising intensity of 0.3-0.4 was used to preserve more neural signal details. For medium-noise brain regions (noise density 0.25-0.55), a moderate denoising intensity of 0.5-0.6 was used to balance noise removal and signal fidelity. For high-noise brain regions (noise density 0.55-0.82), a severe denoising intensity of 0.7-0.8 was used to focus on noise removal, and signal loss was compensated for through signal reconstruction algorithms. After denoising, the data resolution remained at 1.5×1.5×3.0mm, the number of time points remained unchanged, the data size per group was 172MB, and the percentage of invalid noise pixels decreased from 15.6%±3.2% to 3.8%±1.1%, significantly improving the clarity of brain region boundaries.

[0071] After denoising, image signal-to-noise ratio (SNR) assessment data and brain region neural signal fidelity data are simultaneously calculated based on the denoised fMRI data. A denoising quality assessment report is generated, and the parameter weights of the multi-dimensional adaptive denoising model are updated. A two-dimensional quantitative assessment model for the denoising effect of pediatric brain fMRI is constructed, including image quality assessment and neural signal fidelity assessment dimensions, each with a weight of 50%. Combining denoised data, original data, historical denoising gold standard data, and brain region localization results, the SNR assessment data for the whole brain and each brain region is calculated using a medical image SNR statistical algorithm. SNR is calculated using the ratio of signal mean to noise standard deviation, PSNR is calculated using the logarithm of mean squared error, and SSIM is calculated using three-dimensional similarity of brightness, contrast, and structure. Before denoising, the mean SNR of the whole brain was 15.3±3.8, which improved to 32.7±4.5 after denoising, an increase of 113.7%; PSNR improved from 22.6±2.9 to 38.9±3.1, an increase of 72.1%; and SSIM improved from 0.72±0.08 to 0.91±0.04, an increase of 26.4%. Among the brain regions, the SNR of high-noise regions such as the frontal lobe improved from 12.8±3.5 to 30.2±4.2, an increase of 135.9%, while that of low-noise regions such as the occipital lobe improved from 18.7±4.1 to 35.8±4.7, an increase of 91.4%. The SNR of all brain regions reached more than 90% of the historical gold standard data. The temporal correlation of neural signals, the fidelity of low-frequency oscillation amplitude (ALFF), and the fidelity of functional connectivity strength were calculated for each brain region using a neurophysiological feature matching algorithm. The temporal correlation was 0.68±0.11 before denoising and 0.89±0.05 after denoising, with a deviation of no more than 0.02 from the gold standard of 0.89±0.07. The mean ALFF fidelity was 0.92±0.06, reaching 0.95±0.04 in key developmental brain regions. The mean functional connectivity fidelity was 0.88±0.05, with a matching degree of 93.2%±2.3% with the gold standard. Image signal-to-noise ratio assessment data and brain region neural signal fidelity data were matched and correlated by brain region, outputting a two-dimensional quantitative assessment dataset with brain region labels, containing six indicators: SNR, PSNR, SSIM, temporal correlation, ALFF fidelity, and functional connectivity fidelity for 116 brain regions. 89.7% of the indicators had a deviation of ≤5% from the gold standard for excellent indicators, 8.3% had a deviation of 5%-10% for good indicators, 2.0% had a deviation of 10%-15% for acceptable indicators, and there were no unacceptable indicators. A denoising quality assessment report was generated based on this dataset, including basic data information, the denoising process, assessment results for each brain region, a summary of the overall denoising effect, and suggestions for improvement. The overall assessment was excellent. Finally, the weights of the multi-dimensional adaptive denoising model parameters were updated based on the data deviation in the assessment report. The weights for denoising high-noise brain regions were increased by 0.15-0.25, and the weights for signal fidelity preservation in key developmental brain regions were increased by 0.10-0.20. After the update, the model's validation set loss value decreased to 0.006±0.001, and the goodness of fit R² improved to 0.98±0.01, further enhancing the generalization ability.

[0072] This denoising practice successfully processed fMRI brain data from 120 children aged 7-10 years, with an average processing time of 45±8 minutes per case. This included 15±4 minutes for brain region segmentation and feature extraction, 14±2 minutes for model denoising and signal separation, and 4±1 minutes for quality assessment and model updates. Brain development analysis validated the denoised data, demonstrating accurate extraction of neural signal features and a clear representation of the developmental patterns of brain functional connectivity in children. The correlation between the strength of functional connectivity between the prefrontal cortex and hippocampus and children's cognitive abilities reached 0.76±0.08, providing high-quality imaging data support for the assessment of children's brain development. Compared to traditional unified denoising methods, this denoising method achieved a 35.6% higher signal-to-noise ratio, a 12.3% higher neural signal fidelity, and a 28.9% higher processing efficiency. It effectively addresses the challenges of severe noise interference and the difficulty in balancing denoising and signal fidelity in children's fMRI brain data, making it particularly suitable for school-aged children. It can be widely applied in fields such as children's brain development research and early screening for neurodevelopmental disorders. This method can be further extended to children of different ages. By adjusting the model parameters and prior maps to adapt to the brain development characteristics of each age group, the denoising method can be made more universal and applicable.

[0073] In summary, this invention fully utilizes age-matched brain development atlases and multi-dimensional adaptive algorithms to accurately solve the noise interference problems caused by large motion artifacts and strong brain structural heterogeneity in children. By generating noise heatmaps and implementing brain-differentiated denoising strategies, it significantly improves the image signal-to-noise ratio and neural signal fidelity, and provides comprehensive quantitative evaluation indicators, laying a high-quality data foundation for subsequent brain functional connectivity analysis and developmental research. Furthermore, its modular design framework has good scalability and can be extended to children of different ages and clinical applications, providing reliable technical support for pediatric brain development research and early screening of neurodevelopmental disorders.

[0074] Figure 3 This is a schematic diagram of the structure of a denoising system for functional magnetic resonance image data according to an embodiment of this application.

[0075] like Figure 3 As shown, the denoising system 10 for functional magnetic resonance imaging data includes: a data acquisition module 100, a brain region feature extraction module 200, an adaptive denoising module 300, and an evaluation and optimization module 400.

[0076] The system includes: a data acquisition module 100, which acquires raw time-series images of children's functional magnetic resonance imaging (fMRI), prior brain development atlas data, denoised gold standard data of the same age group, and synchronously acquired physiological and motor data; a brain region feature extraction module 200, which, based on the raw time-series images of children's fMRI and prior brain development atlas data, performs brain tissue region segmentation and region of interest localization, obtains the localization results of different brain regions, and extracts the temporal signal features and spatial structure features of different brain regions; and an adaptive denoising module 300, which, based on the denoised gold standard data of the same age group, synchronously acquired physiological and motor data, and temporal signal features and spatial structure... The system features a multi-dimensional adaptive denoising model for functional magnetic resonance imaging (fMRI) of children's brains. Combining the localization results of different brain regions, it generates a heatmap of brain region noise types. Through a signal-noise separation algorithm, it obtains the distinction between neural signals and noise components in each brain region, generates a differentiated denoising strategy for brain region differentiation, and performs adaptive denoising processing on the original time-series fMRI images to obtain denoised fMRI image data. An evaluation and optimization module 400, based on the denoised fMRI image data, simultaneously calculates image signal-to-noise ratio evaluation data and brain region neural signal fidelity data, generates a denoising quality evaluation report, and updates the parameter weights of the multi-dimensional adaptive denoising model.

[0077] It should be noted that the foregoing explanation of the denoising method embodiment for functional magnetic resonance image data also applies to the denoising system for functional magnetic resonance image data in this embodiment, and will not be repeated here.

[0078] The denoising system for functional magnetic resonance imaging (fMRI) data proposed in this application integrates prior brain development atlases, age-appropriate denoising gold standards, and synchronously acquired physiological motion data to accurately segment brain tissue regions and locate regions of interest, efficiently extracting temporal and spatial structural features of brain regions. It then constructs a multi-dimensional adaptive denoising model, generates a heatmap of brain region noise types, and uses a signal-noise separation algorithm to form a graded differentiated denoising strategy, performing adaptive denoising processing on the original data. Simultaneously, it evaluates the image signal-to-noise ratio and the fidelity of neural signals in brain regions, generates a quality assessment report, and dynamically updates the model parameter weights. This significantly improves the accuracy of denoising and the fidelity of neural signals in children's fMRI data, better adapts to the physiological characteristics of children's brain development, effectively suppresses various noise interferences, provides high-quality data support for the analysis of children's brain functional imaging data, and helps improve the accuracy and reliability of research related to children's brain development. Therefore, it solves the problem of insufficient denoising accuracy in existing technologies.

[0079] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 401, the processor 402, and the computer program stored on the memory 401 and capable of running on the processor 402.

[0080] When the processor 402 executes the program, it implements the method for denoising functional magnetic resonance image data provided in the above embodiments.

[0081] Furthermore, electronic devices also include: Communication interface 403 is used for communication between memory 401 and processor 402.

[0082] The memory 401 is used to store computer programs that can run on the processor 402.

[0083] The memory 401 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0084] If the memory 401, processor 402, and communication interface 403 are implemented independently, then the communication interface 403, memory 401, and processor 402 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0085] Optionally, in a specific implementation, if the memory 401, processor 402, and communication interface 403 are integrated on a single chip, then the memory 401, processor 402, and communication interface 403 can communicate with each other through an internal interface.

[0086] Processor 402 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0087] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for denoising functional magnetic resonance image data.

[0088] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that 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. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0089] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0090] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0091] 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 suitable instructions. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: 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.

[0092] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0093] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for denoising functional magnetic resonance imaging data, characterized in that, include: Acquire raw time-series images of children's brain functional magnetic resonance imaging, prior brain development atlas data, denoised gold standard data of historical age groups, and synchronously acquired physiological and motor data; Based on the original temporal image data of the child's functional magnetic resonance imaging and the prior atlas data of the child's brain development, the brain tissue region segmentation and region of interest localization were completed, the localization results of different brain regions were obtained, and the temporal signal features and spatial structural features of different brain regions were extracted. Based on the historical gold standard data for denoising at the same age, the synchronously acquired physiological motion data, and the temporal signal features and spatial structural features, a multi-dimensional adaptive denoising model for functional magnetic resonance imaging (fMRI) of children's brain is constructed. Combined with the localization results of different brain regions, a heatmap of brain region noise types is generated. Through a signal-noise separation algorithm, the distinction between neural signals and noise components in each brain region is obtained, a brain region differentiation-based denoising strategy is generated, and adaptive denoising processing is performed on the original temporal image data of the fMRI to obtain denoised fMRI image data. Based on the denoised functional magnetic resonance imaging data, image signal-to-noise ratio evaluation data and brain region neural signal fidelity data are calculated simultaneously to generate a denoising quality evaluation report and update the parameter weights of the multi-dimensional adaptive denoising model.

2. The method for denoising functional magnetic resonance imaging data according to claim 1, characterized in that, The brain tissue regions were segmented and regions of interest were located, resulting in the localization of different brain regions. Temporal signal features and spatial structural features of different brain regions were extracted, including: Constructing a model for brain tissue segmentation and brain region localization in children; Based on the aforementioned model for segmenting and locating brain regions in children, and combined with the age-matched anatomical structure and functional zoning features of brain regions in the prior atlas data of children's brain development, the improved 3DU-Net network is used to segment the gray and white matter of the brain, cerebrospinal fluid, and non-brain tissues of the skull. The graph attention network is then used to capture the spatial correlation features of each functional brain region, thereby completing the localization of regions of interest and obtaining the localization results of different brain regions. Based on the localization results of the different brain regions, the time-series fluctuation features and frequency domain features of the functional magnetic resonance signals of each brain region are extracted as time-series signal features, and the anatomical boundary features, gray-scale distribution features, and spatial texture features of the brain regions are extracted as spatial structure features.

3. The method for denoising functional magnetic resonance imaging data according to claim 1, characterized in that, Generate a heatmap of brain region noise types, including: Construct a regional spatial grid partitioning model that adapts to the characteristics of children's brain development; Based on the regional spatial grid partitioning model adapted to the characteristics of children's brain development, combined with the brain region partitioning boundaries and key developmental region annotations in the prior atlas data of children's brain development, as well as the noise fluctuation amplitude in the temporal signal characteristics and the artifact distribution characteristics in the spatial structure characteristics, the noise density value and noise type confidence of each grid cell are calculated by the kernel density estimation algorithm. The noise density value and noise type confidence level are mapped to a preset color gradient range and superimposed on a three-dimensional brain region template of a priori map of children's brain development to generate a brain region noise type heatmap labeled with the noise type, density level, confidence level and update timestamp of each grid unit.

4. The method for denoising functional magnetic resonance imaging data according to claim 1, characterized in that, The signal-noise separation algorithm is used to obtain the distinction between neural signals and noise components in each brain region, including: Construct brain region-specific signal-noise separation models; Based on the brain region-specific signal-noise separation model, the non-stationary fluctuation features of the temporal signals of each brain region are extracted using the sliding window method. By combining independent component analysis with gradient boosting tree algorithm, the components of neurophysiological signals, motion noise, physiological noise, and instrument noise are separated, and the results of distinguishing the neural signals and noise components of each brain region are output. Based on the differentiation results of neural signals and noise components in each brain region, the proportion of noise components in the corresponding brain region is obtained. When the proportion of noise components exceeds the preset noise threshold of the corresponding brain region, the adaptive weighted denoising mechanism of the brain region is triggered to generate targeted denoising parameter correction instructions.

5. The method for denoising functional magnetic resonance imaging data according to claim 1, characterized in that, Simultaneously calculate image signal-to-noise ratio assessment data and brain region neural signal fidelity data, including: Construct a two-dimensional quantitative evaluation model for the noise reduction effect of functional magnetic resonance imaging (fMRI) in children's brains; Based on the dual-dimensional quantitative evaluation model of the denoising effect of functional magnetic resonance imaging (fMRI) of children's brain, combined with denoised fMRI image data, original temporal fMRI image data of children's brain, historical gold standard data of denoising for the same age group, and localization results of different brain regions, the signal-to-noise ratio (SNR), peak SNR, and structural similarity index of the whole brain and each brain region are calculated by medical image SNR statistical algorithm to obtain image SNR evaluation data. At the same time, the temporal correlation, low-frequency oscillation amplitude fidelity, and functional connectivity strength fidelity of neural signals in each brain region are calculated by neurophysiological feature matching algorithm to obtain neural signal fidelity data of brain regions. The image signal-to-noise ratio evaluation data and the brain region neural signal fidelity data are matched and associated by brain region to output a two-dimensional quantitative evaluation dataset with brain region labels.

6. The method for denoising functional magnetic resonance imaging data according to claim 1, characterized in that, The signal-noise separation algorithm formula is as follows: ； ； ； ； ； ； ； ； in, A matrix of raw time-series images from functional magnetic resonance imaging (fMRI) of the brain in children; A design matrix that includes both neural networks and noise; This is the regression coefficient matrix; The neural signal is in residual form; For the first Brain region spatial mask matrix; For the first Brain region neural signal regression matrix; These are the regression coefficients of the neural signal; This is the noise reference regression matrix; The noise regression coefficient; For the first residual neural signals in brain regions; It is a signal mixing matrix; It is a set of independent components; This is the total noise matrix; for Time-major filter weight vector; for Time-weight vector; This is the step size parameter; To prevent zero constant; for Timing error signal; for Time-of-flight noise reference vector; The estimated neural signal matrix; To find the optimal solution operator; For regularization weights; For the development of the prior space regularization matrix; These are real neural signals; These are the wavelet coefficients after thresholding. The sign function for wavelet coefficients; These are the original wavelet coefficients; For the first Noise-reducing neural signals in brain regions; For the first Adaptive brain region separation operator; This is the physiological motion noise reference matrix; A priori atlas of child development; These are the model weight parameters; This is the denoised functional magnetic resonance imaging data.

7. A denoising system for functional magnetic resonance imaging data, characterized in that, include: The data acquisition module acquires raw time-series images of children's functional magnetic resonance imaging (fMRI), prior brain development atlas data, denoised gold standard data of historical age groups, and synchronously acquired physiological and motor data. The brain region feature extraction module, based on the original temporal image data of the child's functional magnetic resonance imaging and the prior brain development atlas data, completes the segmentation of brain tissue regions and the localization of regions of interest, obtains the localization results of different brain regions, and extracts the temporal signal features and spatial structural features of different brain regions; The adaptive denoising module, based on the historical gold standard data for denoising at the same age, synchronously acquired physiological motion data, and the temporal signal features and spatial structural features, constructs a multi-dimensional adaptive denoising model for functional magnetic resonance imaging (fMRI) of children's brains. Combining the localization results of different brain regions, it generates a heatmap of brain region noise types. Through a signal-noise separation algorithm, it obtains the distinction results between neural signals and noise components in each brain region, generates a brain region-specific hierarchical differential denoising strategy, and performs adaptive denoising processing on the original temporal image data of the fMRI to obtain denoised fMRI image data. The evaluation and optimization module simultaneously calculates image signal-to-noise ratio evaluation data and brain region neural signal fidelity data based on the denoised functional magnetic resonance imaging data, generates a denoising quality evaluation report, and updates the parameter weights of the multi-dimensional adaptive denoising model.

8. An electronic device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the method for denoising functional magnetic resonance image data as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When a computer program or instruction is executed, it implements the method for denoising functional magnetic resonance image data as described in any one of claims 1-6.

Images