Denoising method and system suitable for infant brain function magnetic resonance image data, medium and terminal

Abstract

The invention provides a denoising method and system suitable for infant brain function magnetic resonance image data, a medium and a terminal, and relates to the field of infant brain function magnetic resonance image data denoising. The invention provides a denoising method suitable for infant brain function magnetic resonance image data. The denoising method comprises the following steps: acquiring an original function magnetic resonance image of an infant brain; performing noise component removal on the lightly preprocessed functional magnetic resonance image, wherein the noise component removal decomposes the lightly preprocessed functional magnetic resonance image into a time independent component and a space independent component to be processed respectively; noise time point removal is carried out on the deep preprocessing functional magnetic resonance image, reconstruction errors are adopted as an anomaly detection standard, time points possibly polluted by noise are recognized and removed, and the time points possibly polluted by noise are recognized and removed. According to the method, noise can be effectively removed, neural signals are accurately reserved, and the data quality and the denoising effect are improved.

Description

Technical Field

[0001] The present invention relates to the field of denoising of infant brain functional magnetic resonance image data, and in particular to a denoising method, system, medium and terminal suitable for infant brain functional magnetic resonance image data. Background Art

[0002] Functional magnetic resonance imaging (fMRI) is a non-invasive brain imaging technique widely used to study brain development in infants and young children. By monitoring dynamic changes in the blood oxygenation level-dependent (BOLD) signal, fMRI can reflect changes in cerebral blood oxygenation levels caused by neural activity. However, the acquisition and analysis of fMRI data from infants and young children face numerous challenges, which affect the accuracy and reliability of research results. During fMRI scanning, infants and young children experience frequent and large head movements, which can easily produce motion artifacts. Furthermore, infants and young children have different circadian rhythms than adults, with higher respiratory and heart rates, which can exacerbate physiological noise interference. Hardware noise from the scanning equipment and environmental factors also increase noise complexity, significantly reducing the signal-to-noise ratio. Existing fMRI denoising methods are mostly based on the characteristics of adult data and are difficult to effectively process for infant and young children. Traditional bandpass filtering methods typically use the 0.01-0.1 Hz frequency band in adult studies, but infants and young children have higher physiological signal frequencies, and directly applying these frequency bands can result in the loss of valuable neural signals. Spatial smoothing methods are effective in adults, but in infant and toddler data, due to the small brain volume, excessive smoothing will blur the boundaries of brain regions and affect the analysis of brain development.

[0003] In terms of noise removal, traditional regression denoising methods based on physiological signals are difficult to apply because infants and young children are unable to cooperate with wearing physiological signal acquisition equipment. Head motion parameter and global signal regression methods have limited effectiveness in infant data. Traditional regression methods have difficulty dealing with complex head motion artifacts and may mistakenly delete real neural signals. Methods based on independent component analysis (ICA) have certain advantages, but the functional connectivity patterns in infant data are unstable, and component identification faces great challenges. Methods based on motion indicators usually rely on fixed thresholds or manually set parameters and are difficult to adapt to the variable noise environment in infant fMRI data. As a result, time points containing useful neural signals may be mistakenly deleted, affecting data integrity and analysis accuracy.

[0004] In summary, there is an urgent need for a denoising method that can prevent the accidental deletion of neural signals during denoising and can improve data quality and denoising effect. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of the present invention is to provide a denoising method, system, medium and terminal suitable for infant brain functional magnetic resonance image data, which is used to solve the problem of the prior art of mistakenly deleting neural signals during denoising, affecting data integrity and analysis accuracy.

[0006] The present invention provides a denoising method for infant brain functional magnetic resonance image data, comprising: acquiring original functional magnetic resonance images of the infant brain, performing preprocessing on the original functional magnetic resonance images, including at least head motion correction and distortion correction, to obtain lightly preprocessed functional magnetic resonance images; performing noise component removal on the lightly preprocessed functional magnetic resonance images, wherein the noise component removal decomposes the lightly preprocessed functional magnetic resonance images into time-independent components and space-independent components, and processes them separately, to obtain deep preprocessed functional magnetic resonance images; performing noise time point removal on the deep preprocessed functional magnetic resonance images, and in the noise time point removal stage, using reconstruction error as an abnormality detection criterion to identify and remove time points that may be contaminated by noise.

[0007] In some embodiments of the first aspect of the present invention, the preprocessing of the original functional magnetic resonance image, including at least head motion correction and distortion correction, includes: head motion correction corrects the image by analyzing head motion parameters to eliminate rotation and translation artifacts; wherein the head motion parameters generated during the correction process can be used for subsequent motion evaluation of signal quality; distortion correction compensates for magnetic field inhomogeneity and deformation effects through two EPI images in different directions, and restores the consistency of the image anatomical structure; preprocessing also includes image registration, image registration aligns the original functional magnetic resonance image to the T1 space, and then resamples and integrates the registration parameters at one time to avoid registration errors; preprocessing also includes delinear trending, which uses high-pass filtering to remove low-frequency signal components to reduce the interference of equipment drift and physiological baseline fluctuations on the data.

[0008] In some embodiments of the first aspect of the present invention, the noise component removal decomposes the lightly preprocessed functional magnetic resonance image into time-independent components and spatial-independent components for separate processing, comprising: using independent component analysis to decompose the lightly preprocessed functional magnetic resonance image into statistically independent spatial and temporal components; the spatial component is an activation pattern of a brain region, represented as a spatial map S i The temporal component captures the temporal dynamics of brain activity and is represented as a time series T i ; Space map S i and the time series T i , respectively, through the spatial encoder f s and the time encoder f t to process;

[0009] Time encoder f t Multi-scale analysis is performed using different convolution kernel sizes, as shown in the following formula

[0010] h t =f t (T i θ t )

[0011] Among them, θ t represents a learnable parameter;

[0012] Spatial encoder f s Through convolutional layers, residual blocks and attention mechanisms, the spatial graph S i Extract high-level features as shown in the following formula:

[0013] h s =f s (S i θ s )

[0014] Among them, θ s is a learnable parameter.

[0015] In some embodiments of the first aspect of the present invention, the noise component removal decomposes the lightly preprocessed functional magnetic resonance image into time-independent components and space-independent components and processes them separately, including:

[0016] A multi-loss objective is used for the time-independent component and the space-independent component. The multi-loss objective formula is as follows:

[0017]

[0018] in, is the spatial encoder f s The cross entropy loss, is the time encoder f t The cross entropy loss, It is a joint multimodal cross entropy loss that combines spatial and temporal information.

[0019] In some embodiments of the first aspect of the present invention, the noise time point removal is performed on the deep preprocessed functional magnetic resonance image, and the noise time point removal uses reconstruction error as an abnormality detection criterion to identify and remove time points that may be contaminated by noise, including:

[0020] A noise time point removal model is constructed. The input of the noise time point removal model is a multivariate functional magnetic resonance time series. The multivariate functional magnetic resonance time series is then divided into overlapping windows of size w to capture the temporal dynamics of w consecutive time points. The formula for functional magnetic resonance image data is as follows:

[0021] t=w,w+1,…,T

[0022] in, represents the extracted brain region time series matrix, T is the number of time points, which represents the length of the time series in the functional magnetic resonance image data, d is the number of regions of interest, and w is the number of time points in the overlapping window, which represents the number of consecutive time points contained in each overlapping window;

[0023] In some embodiments of the first aspect of the present invention, the noise time point removal is performed on the deep preprocessed functional magnetic resonance image, and the noise time point removal uses reconstruction error as an abnormality detection criterion to identify and remove time points that may be contaminated by noise, including:

[0024] The time series of deep preprocessed functional magnetic resonance images is input into the noise time point removal model to obtain a reconstruction window. The noise is determined and removed based on the reconstruction error between the original window and the reconstructed window. The formula is as follows:

[0025]

[0026] Among them, L rec represents the reconstruction error, x i,j and represents the original value and the reconstructed value in the i-th time step and the j-th region of interest, d is the number of regions of interest, and w is the number of time points in the overlapping window.

[0027] To achieve the above-mentioned objectives and other related objectives, the second aspect of the present invention provides a denoising system for functional magnetic resonance image data of infant brains, comprising a data preprocessing module for acquiring original functional magnetic resonance images of infant brains, and performing preprocessing on the original functional magnetic resonance images, including at least head motion correction and distortion correction, to obtain lightly preprocessed functional magnetic resonance images; a noise component removal module for performing noise component removal on the lightly preprocessed functional magnetic resonance images, wherein the noise component removal decomposes the lightly preprocessed functional magnetic resonance images into time-independent components and spatial-independent components and processes them separately to obtain deep-preprocessed functional magnetic resonance images; and a noise time point removal module for performing noise time point removal on the deep-preprocessed functional magnetic resonance images, wherein the noise time point removal uses reconstruction error as an abnormality detection criterion to identify and remove time points that may be contaminated by noise.

[0028] To achieve the above-mentioned and other related purposes, the third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, a denoising method for functional magnetic resonance imaging data of the brain of infants and young children is implemented.

[0029] To achieve the above-mentioned objectives and other related objectives, the fourth aspect of the present invention provides a computer program product, which includes computer program code. When the computer program code is run on a computer, the computer implements a denoising method suitable for functional magnetic resonance image data of the brain of infants and young children.

[0030] To achieve the above-mentioned objectives and other related objectives, the fourth aspect of the present invention provides a computer device comprising a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement a denoising method applicable to functional magnetic resonance image data of the infant brain.

[0031] Beneficial effects

[0032] The denoising method for functional magnetic resonance imaging data of infant brains provided by the present invention can effectively remove noise, accurately preserve neural signals, and improve data quality and denoising effects. Its deep learning-based method is adaptive and accurate, does not require excessive parameter setting, and can adapt to different scanning conditions and noise conditions. This provides high-quality data support for infant brain connectome research, brain development analysis, and early diagnosis of related diseases, helps to gain a deeper understanding of the infant brain development mechanism, promotes the development of clinical diagnosis and treatment, and has important scientific research value and clinical application significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 FIG4 is a flow chart of a denoising method for infant brain functional magnetic resonance image data according to an embodiment of the present invention.

[0034] Figure 2 FIG4 is a flow chart of a denoising method for infant brain functional magnetic resonance image data according to an embodiment of the present invention.

[0035] Figure 3 FIG4 is a schematic diagram of a process for preprocessing original functional magnetic resonance images in one embodiment of the present invention.

[0036] Figure 4 FIG. 4 is a flow chart of noise component removal according to an embodiment of the present invention.

[0037] Figure 5 FIG. 4 is a flow chart of noise component removal according to an embodiment of the present invention.

[0038] Figure 6 FIG. 4 is a flow chart of noise time point removal in one embodiment of the present invention.

[0039] Figure 7 FIG. 4 is a flow chart of noise time point removal in one embodiment of the present invention.

[0040] Figure 8 FIG. 1 is a schematic structural diagram of a computer device according to an embodiment of the present invention.

[0041] Figure 9 FIG. 1 is a schematic structural diagram of a system for denoising functional magnetic resonance image data of infant brains according to an embodiment of the present invention.

[0042] Figure 10 This is a comparison chart of retest reliability of different noise time point removal methods in one embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0044] Before further explaining the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are subject to the following interpretations:

[0045] <1> Functional magnetic resonance imaging (fMRI): A technique that reflects brain neural activity by detecting changes in blood oxygen levels.

[0046] <2> Echo planar imaging (EPI): A fast magnetic resonance imaging technique used in fMRI. It uses a single radiofrequency pulse to collect multiple echoes, enabling rapid acquisition of brain images. However, it is susceptible to distortion caused by magnetic field inhomogeneities.

[0047] <3> Single-band reference image: An image obtained by selecting a specific band in multi-band imaging, often used as a benchmark for subsequent analysis.

[0048] <4> Field map: A magnetic field distribution image used for magnetic resonance imaging distortion correction, which intuitively presents magnetic field inhomogeneities.

[0049] <5> Independent Component Analysis (ICA): Belongs to the field of signal processing technology and can decompose fMRI data into independent spatial components and time series to separate signals from noise.

[0050] <6> Head motion correction: For fMRI data, since head movement during scanning produces artifacts, the image is spatially transformed by calculating head motion parameters to eliminate motion artifacts.

[0051] <7> Distortion Correction: To address image distortion caused by magnetic field inhomogeneity in magnetic resonance imaging, the original image is corrected to ensure the accuracy of the image's spatial information.

[0052] <8> Image registration: The process of spatially aligning fMRI data with individual T1-weighted images or a standard spatial template to ensure that the same anatomical structures correspond between different images, facilitating subsequent comparative analysis and group studies.

[0053] <9> De-linearization trend: Equipment instability and physiological factors will introduce low-frequency drift. High-pass filtering filters out low-frequency signals by setting the cutoff frequency.

[0054] <10> Functional connectivity: reflects the information transmission and collaborative working mechanism between brain regions.

[0055] <11> FSL: A set of tools for processing and analyzing neuroimaging data, including MCFLIRT, TOPUP, FLIRT and other tools, which can perform operations such as head motion correction, distortion correction, and image registration.

[0056] <12> ANTs: Tools for image registration and analysis.

[0057] <13> Brain atlas: A standardized template for partitioning and labeling the brain, containing information such as brain structure and function, providing a unified reference standard for brain image registration, regional positioning and functional analysis.

[0058] Figure 1 and Figure 2 The following is a flow chart of a system for denoising functional magnetic resonance imaging data of infant brains according to an embodiment of the present invention. The method in this embodiment mainly includes the following steps:

[0059] Step S1: acquiring an original functional magnetic resonance image of the infant's brain, and performing preprocessing on the original functional magnetic resonance image including at least head motion correction and distortion correction to obtain a lightly preprocessed functional magnetic resonance image;

[0060] Step S2: performing noise component removal on the lightly preprocessed functional magnetic resonance image, wherein the noise component removal decomposes the lightly preprocessed functional magnetic resonance image into time-independent components and space-independent components and processes them separately to obtain a deep preprocessed functional magnetic resonance image;

[0061] Step S3: performing noise time point removal on the deep preprocessed functional magnetic resonance image. In the noise time point removal stage, the reconstruction error is used as an abnormality detection criterion to identify and remove time points that may be contaminated by noise.

[0062] In an embodiment of the present invention, Figure 3The following is a schematic diagram of the process of preprocessing the original functional magnetic resonance image in an embodiment of the present invention, wherein step S1 can be divided into the following steps:

[0063] Step S1a: Head motion correction corrects the image by analyzing head motion parameters to eliminate rotation and translation artifacts. The corrected parameters are used for subsequent evaluation of the impact of motion on the signal.

[0064] In a feasible real-time approach, head motion correction is performed using the MCFLIRT tool in FSL software, a suite of tools for neuroimaging data processing and analysis. MCFLIRT employs advanced linear image registration algorithms to accurately analyze head motion parameters. By applying a rigid transformation to each frame, it effectively eliminates motion artifacts caused by head translation and rotation. After correction, previously blurred brain tissue edges in the image become clear, and motion artifacts are significantly suppressed. The corrected head motion parameters, which include the three-dimensional translation and rotation of each frame relative to the reference frame, are used to subsequently assess the impact of motion on signal quality and serve as a key indicator for determining data reliability and usability.

[0065] Step S1b: Distortion correction uses two EPI images in different directions to compensate for magnetic field inhomogeneity and deformation effects and restore the consistency of the image anatomical structure.

[0066] In a feasible real-time approach, distortion correction is performed by acquiring EPI images in both positive and negative phase encoding directions and compensating for magnetic field distortion using the TOPUP tool in FSL software. The TOPUP tool models and analyzes image distortion caused by magnetic field inhomogeneities based on these acquired EPI images in both positive and negative phase encoding directions. It then uses a complex algorithm to calculate distortion correction parameters and geometrically correct the images. Comparing the images before and after correction, the geometric distortion previously present in deep brain structures is significantly reduced, significantly improving the geometric accuracy of the images and enabling a more accurate display of brain tissue structure.

[0067] Step S1c: Preprocessing also includes image registration. In the image registration stage, the original functional magnetic resonance image is registered to the individual T1-weighted image space, and the registration parameters are resampled and integrated at one time to avoid misregistration.

[0068] In one feasible implementation, the original fMRI image is first registered to the spatial coordinate system of the individual's T1-weighted structural image, known as anatomical space. This space has high spatial resolution and anatomical information, serving as the standard spatial reference for subsequent analysis. The SyN algorithm within ANTs software is then used to align the original fMRI image to the same subject's T1-weighted structural image. This T1-weighted image, with its high spatial resolution and clear anatomical information, serves as the spatial reference for registration. During the registration process, normalized mutual information (NMI) is used as a similarity metric to measure the degree of alignment between the functional and structural images, thereby improving registration accuracy. By optimizing transformation parameters, the SyN algorithm enables nonlinear deformation registration, accurately correcting for structural differences or distortions. After registration, all transformation parameters (such as motion correction, distortion correction, and transformation to T1 space) are combined and applied in a single resampling operation, avoiding image quality degradation caused by multiple interpolations. This strategy effectively maintains spatial consistency and anatomical fidelity, providing high-quality data for subsequent brain region mapping and functional analysis.

[0069] Step S1d: Preprocessing also includes de-linear trending, which uses high-pass filtering to remove low-frequency signal components to reduce the interference of device drift and physiological baseline fluctuations on the data.

[0070] In one feasible implementation, the de-linear trending operation is implemented using the fslmaths tool in the FSL software. The fslmaths tool supports high-pass filtering operations. By setting reasonable parameters, it can effectively filter out low-frequency drift and baseline noise. The cutoff frequency is set to 0.0008 Hz to remove low-frequency interference signals generated by device drift and physiological baseline fluctuations.

[0071] In an embodiment of the present invention, Figure 4 and Figure 5 The following figure shows a schematic diagram of the noise component removal process, where step S2 can be divided into the following steps:

[0072] Step S2a: Use independent component analysis to decompose the lightly preprocessed functional magnetic resonance image into statistically independent spatial and temporal components; the spatial component is the activation pattern of brain regions, represented as a spatial map S i The temporal component captures the temporal dynamics of brain activity and is represented as a time series T i ; Space map S i and the time series T i , respectively, through the spatial encoder f s and the time encoder f t Specifically, the activation pattern between brain regions represented by the spatial component can be understood as the distribution of activation intensity of different brain regions under specific independent components, forming a static spatial map Si , which can be seen as the spatial localization of specific functional patterns in the brain. For example, a spatial component related to vision may show high activation in the occipital area (visual cortex). The inter-component represents the activation intensity of the functional pattern over time, forming a time series T i , reflecting the functional state of the brain Dynamic evolution during the entire scan .

[0073] Time encoder f t Use different convolution kernel sizes to perform multi-scale analysis and obtain the temporal feature representation h t , the time feature represents h t As shown in the following formula:

[0074] h t =f t (T i θ t )

[0075] Among them, θ t are learnable parameters.

[0076] Spatial encoder f s Through convolutional layers, residual blocks and attention mechanisms, the spatial graph S i Extract high-level features from the temporal feature representation h s , the time feature represents h s As shown in the following formula:

[0077] h s =f s (S i θ s )

[0078] Among them, θ s is a learnable parameter. Specifically, θ s is the spatial encoder f s The set of learnable parameters in all convolutional layers, residual structures, and attention mechanisms in the network, covering variables such as weights and biases that are automatically optimized during network training.

[0079] It is understandable that the convolutional layer is used to extract local spatial structural information and perceive spatial patterns between different brain regions through multiple learnable convolution kernels. The residual block introduces a skip connection mechanism. By introducing a short-circuit connection between the input and output, the model can gradually superimpose deeper nonlinear information while maintaining the original spatial features, effectively alleviating the gradient vanishing problem in deep networks, improving feature expression capabilities, helping to capture deep nonlinear relationships, and enhancing the model's ability to express complex brain region co-activation structures. The attention mechanism dynamically allocates weights to enable the model to focus on more discriminative key brain regions, while suppressing interference from background or low-correlation areas, thereby extracting more robust and semantically informative spatial feature representations.

[0080] Step S2b: A multi-loss target is used for the time-independent component and the space-independent component. The multi-loss target formula is shown in the following formula:

[0081]

[0082] in, is the spatial encoder f s The cross entropy loss, is the time encoder f t The cross entropy loss, It is a joint multimodal cross entropy loss that combines spatial and temporal information.

[0083] In order to improve the joint discrimination ability of spatial and temporal components, this paper introduces a multi-level gradient modulation mechanism to dynamically adjust the learning weights of each modality classification task during the joint training process. i ) is consistently lower than the classification accuracy of the time modality (such as the time series T i ), the system will automatically increase the weight ratio of the spatial modality loss term in the total loss function to enhance the spatial encoder's attention to spatial features. This regulation mechanism monitors the loss convergence of each subtask and adjusts the balance coefficient of the three losses in real time, thereby avoiding the problem of one modality dominating the training process and causing insufficient learning of the other modality. In specific implementation, the system adjusts the amplification factor of the corresponding gradient direction according to the convergence rate and current accuracy of the cross-entropy loss of each modality, realizes the adaptive enhancement of "soft weights", and enhances the overall robustness and generalization ability of the model. Finally, on the dataset of infant functional magnetic resonance images, the accuracy of noise component recognition reached 98.2%, effectively removing noise from the data.

[0084] In an embodiment of the present invention, Figure 6 and Figure 7 The following figure shows the process flow of noise time point removal, where step S3 can be divided into the following steps:

[0085] Step S3a: Construct a noise time point removal model, wherein the input of the noise time point removal model is a multivariate functional magnetic resonance time series, and then the multivariate functional magnetic resonance time series is divided into overlapping windows of size w to capture the temporal dynamics of w consecutive time points. represents the extracted brain region time series matrix, T is the number of time points, which represents the time series length in the functional magnetic resonance image data, d is the number of regions of interest (ROI), each region of interest corresponds to the BOLD signal of a brain region, w is the number of time points in the overlapping window, which represents the number of consecutive time points contained in each overlapping window. The generated input segment is represented as:

[0086] t=w,w+1,…,T

[0087] For illustration, during the pre-training phase, noise-free data is used as the pre-training dataset to construct the noise time point removal model. The "multivariate" in the multivariate fMRI time series is understood to refer to multiple regions of interest. A region of interest (ROI) consists of a set of specific spatial locations (voxels), i.e., a region within an fMRI image. Thus, in an image formed by continuous fMRI images, by calculating the average BOLD signal intensity for the same ROI at each time point, a time series for that ROI can be obtained, representing the changes in activity in that brain region throughout the scan. Ultimately, multiple time series can be obtained, each reflecting the dynamic characteristics of neural activation over time in the corresponding ROI, which can be used for subsequent functional analysis and anomaly detection.

[0088] In a specific embodiment, the test is performed on a self-built dataset containing noise-free data of 200 infants.

[0089] Step S3b: Input the time series of the deep preprocessed functional magnetic resonance image into the noise time point removal model, and obtain the reconstruction window. The noise is determined and removed based on the reconstruction error between the original window and the reconstructed window. The formula is as follows:

[0090] The training objective of the model is to minimize the mean square error (MSE) between the original window and the reconstructed window:

[0091]

[0092] Among them, L rec Represents the reconstruction error, which is the difference between the original data and the denoised reconstructed data. i,j and Denote the original and reconstructed values ​​at the u-th time step and the j-th region of interest, respectively, d is the number of regions of interest, and w is the number of time points within the overlapping window. By minimizing the reconstruction error on normal data, the model becomes more sensitive to small deviations that represent noise or artifacts, thereby improving the denoising effect and preserving more valuable neural signals.

[0093] For illustration, the original window is the window generated by the noise-free dataset used in pre-training the model, and the reconstructed window is the window generated by the time series of lightly pre-processed functional magnetic resonance images through the noise time point removal model.

[0094] As can be understood, the noise time point removal model extracts temporal features from functional magnetic resonance (fMRI) time series data through a temporal convolutional network (TCN). The graph attention (GAT) module then assigns adaptive attention weights to the connections between ROIs. The MLP acts as a fusion layer, integrating the temporal features extracted by the TCN with the spatial features captured by the GAT. Finally, the MLP is input into a recurrent neural network (RNN) decoder and a series of fully connected layers to generate a reconstruction of the original input window. The RNN decoder processes the latent representation to capture sequential dependencies and generate a reconstructed output.

[0095] Optionally, during the noise time point removal phase, the test fMRI time point is processed by the trained encoder-decoder model. The model reconstructs each time point and calculates the reconstruction error:

[0096]

[0097] The score S represents the deviation between the observed signal and the reconstructed signal, with higher values ​​indicating greater deviation. i,j and denote the original and reconstructed values ​​at the i-th time step and the j-th region of interest, d is the number of regions of interest (ROIs), and w is the number of time points in the overlapping window. The reconstruction error of all time points is expressed as:

[0098] S=[s1,s2,…,s T ]

[0099] High reconstruction error indicates deviations from normal patterns, suggesting the presence of noise or artifacts. The reconstruction error provides an indicator of how well the model’s learned representations align with the observed data, effectively distinguishing between typical neural activity and anomalous patterns.

[0100] In a specific embodiment, for the 80th time point of a certain infant data, the reconstruction error is calculated to be 0.1, through the statistical analysis of the error distribution of clean data in the pre-training phase, the threshold is set to 0.06. If the error exceeds the threshold, it is determined as a noise point and removed.

[0101] Example 1

[0102] To support the aforementioned denoising method for infant and young child functional magnetic resonance imaging (fMRI) brain data, a high-speed NVMe solid-state drive (SSD) was selected as the computer-readable storage medium to ensure efficient and rapid data access during program execution. The storage medium contains the denoising system's program code, which is stored as a binary file to ensure efficient reading and execution on various devices. The program code is written in Python, offering strong scalability and maintainability. Key libraries include computational and scientific libraries such as Pytorch, Numpy, and Scipy, providing robust support for the training and inference of deep learning models. The code's clear and modular structure facilitates future porting and version updates, while also ensuring excellent cross-platform compatibility.

[0103] The electronic terminal utilizes high-performance server hardware, specifically a 12th-generation Intel Core i9-12900K processor. This supports massively parallel computing, significantly accelerating data processing and algorithm execution. Furthermore, the server is equipped with 64GB of memory, capable of processing large-scale infant fMRI datasets and ensuring stable system operation under high load. The server is equipped with an NVIDIA 4090 GPU, offering powerful computing capabilities, making it particularly well-suited for training and inference of deep learning models, accelerating noise removal and feature extraction.

[0104] The operating system used is Ubuntu 20.04, a stable version of the Linux operating system that provides efficient memory management and multi-tasking capabilities, supporting the efficient operation of deep learning frameworks such as Pytorch. In actual applications, users first store infant fMRI data on the server's hard drive and, by running the program code on the storage medium, sequentially call on the server's processor and GPU resources. The system first performs data acquisition and preprocessing, including standard processing steps such as motion correction, spatial registration, and temporal filtering. Then, through the noise component removal module and the noise time point removal module, artifacts and noise components in the data are automatically removed, and finally denoised high-quality data is output. The data after this denoising process can be directly used for infant brain connectome analysis, brain development trajectory research, and early diagnosis of related diseases.

[0105] It should be understood that the specific process of each module executing the above-mentioned corresponding steps has been described in detail in the above-mentioned method embodiment and will not be repeated here for the sake of brevity. It should also be understood that the division of modules in the embodiments of the present invention is schematic and is only a logical functional division. In actual implementation, there may be other division methods. In addition, the functional modules in the various embodiments of the present invention may be integrated into a processor, or may exist separately physically, or two or more modules may be integrated into a single module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0106] Figure 8 This is a schematic block diagram of a system for denoising infant brain functional magnetic resonance image data provided by an embodiment of the present invention. As shown in the figure, the system includes: a data preprocessing module 51, a noise component removal module 52, and a noise time point removal module 53.

[0107] The data preprocessing module 51 is used to acquire raw functional magnetic resonance (fMRI) images of infants' and young children's brains and perform preprocessing on the raw fMRI images, including at least motion correction and distortion correction, to obtain lightly preprocessed fMRI images. The noise component removal module 52 performs noise component removal on the lightly preprocessed fMRI images. This noise component removal decomposes the lightly preprocessed fMRI images into temporally independent components and spatially independent components, and processes each separately to obtain deep-preprocessed fMRI images. The noise time point removal module 53 performs noise time point removal on the deep-preprocessed fMRI images. This noise time point removal uses reconstruction error as an anomaly detection criterion to identify and remove time points that may be contaminated by noise.

[0108] It should be understood that the specific process of each module executing the above-mentioned corresponding steps has been described in detail in the above-mentioned method embodiment, and for the sake of brevity, it will not be repeated here. It should also be understood that the division of modules in the embodiment of the present invention is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, the functional modules in the various embodiments of the present invention can be integrated into a processor, or can be physically placed separately, or two or more modules can be integrated into a module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software functional modules.

[0109] Figure 96 is a schematic block diagram of a computer device provided by an embodiment of the present invention. The computer device includes: at least one processor 601, a memory 602, at least one network interface 603 and a user interface 605. The various components in the device are coupled together through a bus system 604. It can be understood that the bus system 604 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 604 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, Figure 6 Various buses are labeled as bus systems.

[0110] The user interface 605 may include a display, a keyboard, a mouse, a trackball, a click gun, keys, buttons, a touch pad or a touch screen.

[0111] It will be appreciated that the memory 602 may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM) or a programmable read-only memory (PROM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0112] The memory 602 in the embodiment of the present invention is used to store various categories of data to support the operation of the electronic terminal 600. Examples of such data include: any executable program for operating on the electronic terminal 600, such as an operating system 6021 and an application 6022; the operating system 6021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application 6022 can include various applications, such as a media player (Media Player), a browser (Browser), etc., for implementing various application services. The implementation of the fundus color photography restoration method based on the diffusion model provided in the embodiment of the present invention can be included in the application 6022.

[0113] The methods disclosed in the above embodiments of the present invention can be applied to or implemented by processor 601. Processor 601 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in processor 601 or by software instructions. The above processor 601 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 601 can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor 601 can be a microprocessor or any conventional processor. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium located in a memory. The processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0114] In an exemplary embodiment, the electronic terminal 600 may be configured to execute the aforementioned method using one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs).

[0115] According to the method provided by the embodiment of the present invention, the present invention also provides a computer program product, the computer program product comprising: computer program code, when the computer program code is run on a computer, the computer executes Figures 1 to 4 A method for restoring fundus color photographs based on a diffusion model in any of the embodiments shown.

[0116] According to the method provided by an embodiment of the present invention, the present invention also provides a computer-readable storage medium, which stores program code. When the program code is run on a computer, the computer executes the above method.

[0117] As used in this specification, the terms "component," "module," "system," and the like are used to represent computer-related entities, hardware, firmware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, and / or a computer. By way of illustration, both an application running on a computing device and a computing device can be a component. One or more components can reside in a process and / or an execution thread, and a component can be located on a computer and / or distributed between two or more computers. In addition, these components can be executed from various computer-readable media having various data structures stored thereon. Components can communicate, for example, via local and / or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component on a local system, a distributed system, and / or a network, such as the Internet interacting with other systems via signals).

[0118] Those skilled in the art will appreciate that the various illustrative logical blocks and steps described in conjunction with the embodiments disclosed herein can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the present invention.

[0119] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0120] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0121] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0122] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0123] In the above embodiments, the functions of each functional unit can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. A computer program product includes one or more computer instructions (programs). When the computer program instructions (program) are loaded and executed on a computer, the process or function according to the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. Available media may be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., high-density digital video discs (DVDs), or semiconductor media (e.g., solid state disks (SSDs)).

[0124] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other media that can store program codes.

[0125] Figure 10 The figure shows the effect of different noise time point removal methods on test-retest reliability in the embodiments of the present invention. The figure shows four different noise time point removal methods: DVARS, FD, projection denoising, and the denoising method proposed in this invention. The figure compares the effects of each method on test-retest reliability at different noise removal ratios for an infant fMRI dataset.

[0126] The results shown clearly demonstrate that the proposed noise time point removal method can significantly maintain the stability of functional connectivity estimates while removing noise. Specifically, the method improved the test-retest reliability by approximately 2%. This improvement demonstrates that the proposed denoising method can maximize the preservation of the subject's neural signals during the denoising process, avoiding the negative impact of excessive denoising on data consistency and reliability, thereby enhancing the accuracy and reliability of functional connectivity analysis.

[0127] In contrast, traditional noise removal methods, such as DVARS and FD, maintained relatively stable test-retest reliability values ​​at lower removal ratios. However, as the removal ratio increased, the test-retest reliability values ​​showed a gradual upward trend, indicating that data consistency improved after removing more noise. However, the projection denoising method performed worst at higher removal ratios, with its test-retest reliability values ​​showing a significant decrease, indicating that a large amount of valid neural signals may have been lost during the denoising process, affecting the stability and reliability of functional connectivity estimates.

[0128] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

[0129] In summary, the present invention proposes an efficient denoising method, system, medium and terminal for infant brain functional magnetic resonance imaging data. This method combines functional magnetic resonance imaging preprocessing, independent component decomposition and multimodal discrimination strategies, and can effectively remove noise while accurately retaining neural signals, significantly improving data quality and analysis reliability. Compared with traditional methods, the present invention has stronger adaptability and robustness in dealing with motion artifacts, physiological noise and signal variability unique to infants and young children. It can provide high-quality data support for brain function development research and early diagnosis of related diseases, and has significant scientific research value and clinical application prospects.

[0130] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

[0131] The above description is only a preferred embodiment of the present invention. It should be noted that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention. These improvements and substitutions should also be considered as the scope of protection of the present invention.

[0132] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention. These improvements and substitutions should also be regarded as the scope of protection of the present invention.

Claims

1. A denoising method for infant brain functional magnetic resonance imaging data, characterized in that: include: collecting raw functional magnetic resonance images of the infant's brain, and performing preprocessing on the raw functional magnetic resonance images, including at least head motion correction and distortion correction, to obtain lightly preprocessed functional magnetic resonance images; Noise components are removed from the lightly preprocessed functional magnetic resonance images. The noise component removal decomposes the lightly preprocessed functional magnetic resonance images into time-independent components and space-independent components, and processes them separately to obtain deep preprocessed functional magnetic resonance images. Noise time points are removed from deep preprocessed functional magnetic resonance images. In the noise time point removal stage, reconstruction error is used as an abnormality detection criterion to identify and remove time points that may be contaminated by noise.

2. The denoising method for infant brain functional magnetic resonance image data according to claim 1, characterized in that: The preprocessing of the original functional magnetic resonance image, including at least head motion correction and distortion correction, comprises: Head motion correction is based on the original functional magnetic resonance image sequence. The translation and rotation parameters of each frame relative to the reference frame are calculated and used to align the images to eliminate artifacts caused by head motion. The head motion parameters generated during the correction process can be used to evaluate the signal quality of subsequent motion. Distortion correction uses two EPI images in different directions to compensate for magnetic field inhomogeneity and deformation effects and restore the consistency of image anatomical structure; Preprocessing also includes image registration, which registers the original functional magnetic resonance images to the individual T1 space and then resamples and integrates the registration parameters to avoid misregistration; Preprocessing also includes de-linear trending, which uses high-pass filtering to remove low-frequency signal components and reduce the interference of device drift and physiological baseline fluctuations on the data.

3. The denoising method for infant brain functional magnetic resonance image data according to claim 1, characterized in that: The noise component removal decomposes the lightly preprocessed functional magnetic resonance image into time-independent components and space-independent components and processes them separately, including: Independent component analysis was used to decompose the lightly preprocessed functional magnetic resonance images into statistically independent spatial and temporal components; the spatial components were activation patterns of brain regions, represented as spatial maps S i The temporal component captures the temporal dynamics of brain activity and is represented as a time series T i ; Space map S i and the time series T i , respectively, through the spatial encoder f s and the time encoder f t to process; Time encoder f t Use different convolution kernel sizes to perform multi-scale analysis and obtain the temporal feature representation h t , the time feature represents h t As shown in the following formula: h t =f t (T i ;θ t ) Among them, θ t is a learnable parameter; Spatial encoder f s Through convolutional layers, residual blocks and attention mechanisms, the spatial graph S i Extract high-level features from the temporal feature representation h s , the time feature represents h s As shown in the following formula: h s =f s (S i ;θ s ) Among them, θ s are learnable parameters.

4. The denoising method for infant brain functional magnetic resonance image data according to claim 4, characterized in that: The noise component removal decomposes the lightly preprocessed functional magnetic resonance image into time-independent components and space-independent components and processes them separately, including: A multi-loss objective is used for the time-independent component and the space-independent component. The multi-loss objective formula is as follows: in, is the spatial encoder f s The cross entropy loss, is the time encoder f t The cross entropy loss, It is a joint multimodal cross entropy loss that combines spatial and temporal information.

5. The denoising method for infant brain functional magnetic resonance image data according to claim 1, characterized in that: The noise time point removal is performed on the deep preprocessed functional magnetic resonance image. The noise time point removal uses the reconstruction error as the abnormality detection standard to identify and remove the time points that may be contaminated by noise, including: A noise time point removal model is constructed. The input of the noise time point removal model is a multivariate functional magnetic resonance time series. The multivariate functional magnetic resonance time series is then divided into overlapping windows of size w to capture the temporal dynamics of w consecutive time points. The formula for functional magnetic resonance image data is as follows: in, represents the extracted brain region time series matrix, T is the number of time points, which represents the length of the time series in the functional magnetic resonance image data, d is the number of regions of interest, and w is the number of time points in the overlapping window, which represents the number of consecutive time points contained in each overlapping window.

6. The denoising method for infant brain functional magnetic resonance image data according to claim 5, characterized in that: The noise time point removal is performed on the deep preprocessed functional magnetic resonance image. The noise time point removal uses the reconstruction error as the abnormality detection standard to identify and remove the time points that may be contaminated by noise, including: The time series of deep preprocessed functional magnetic resonance images is input into the noise time point removal model to obtain a reconstruction window. The noise is determined and removed based on the reconstruction error between the original window and the reconstructed window. The formula is as follows: Among them, L rec Represents the reconstruction error, which is the difference between the original data and the denoised reconstructed data. i,j and denote the original value and the reconstructed value in the i-th time step and the j-th region of interest, respectively, d is the number of regions of interest, and w is the number of time points in the overlapping window.

7. A denoising system for infant brain functional magnetic resonance image data, characterized in that: include: a data preprocessing module, configured to acquire raw functional magnetic resonance images of the infant's brain, and perform preprocessing on the raw functional magnetic resonance images, including at least head motion correction and distortion correction, to obtain lightly preprocessed functional magnetic resonance images; A noise component removal module removes noise components from the lightly preprocessed functional magnetic resonance image. The noise component removal decomposes the lightly preprocessed functional magnetic resonance image into time-independent components and space-independent components, and processes them separately to obtain a deep preprocessed functional magnetic resonance image. The noise time point removal module removes noise time points from deep preprocessed functional magnetic resonance images. The noise time point removal uses reconstruction error as the abnormality detection criterion to identify and remove time points that may be contaminated by noise.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the denoising method for infant brain functional magnetic resonance image data according to any one of claims 1 to 6 is implemented.

9. A computer program product, characterized in that The computer program product includes computer program code, and when the computer program code is run on a computer, the computer is enabled to implement the denoising method for infant brain functional magnetic resonance image data according to any one of claims 1 to 6.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the denoising method for infant brain functional magnetic resonance image data according to any one of claims 1 to 6.

Images