CN-122023178-A - Denoising method, system, equipment and medium for functional magnetic resonance image data

CN122023178ACN 122023178 ACN122023178 ACN 122023178ACN-122023178-A

Abstract

The invention belongs to the technical field of medical image processing, and particularly relates to a denoising method, a system, equipment and a medium for functional magnetic resonance image data, wherein the denoising method, the system, the equipment and the medium are used for acquiring child brain functional magnetic resonance original time sequence image data, child brain development priori map data, historical same-age-stage denoising gold standard data and physiological motion synchronous acquisition data, completing brain tissue region segmentation and interested region positioning to obtain positioning results of different brain regions, extracting time sequence signal characteristics and space structure characteristics of the different brain regions, constructing a child brain functional magnetic resonance multi-dimensional self-adaptive denoising model, generating brain region noise type thermodynamic diagram, generating brain region differential denoising strategies through a signal-noise separation algorithm, obtaining denoised functional magnetic resonance image data, generating a denoising quality evaluation report and updating parameter weights of the multi-dimensional self-adaptive denoising model. Therefore, the problems of insufficient denoising accuracy and the like in the prior art are solved.

Inventors

ZENG SHUJING
ZHANG PEIRAN
LI WENJUAN

Assignees

西南医科大学附属医院

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (9)

1. A method of denoising functional magnetic resonance image data, comprising: Acquiring child brain function magnetic resonance original time sequence image data, child brain development priori map data, historical age-related denoising gold standard data and physiological motion synchronous acquisition data; According to the child brain function magnetic resonance original time sequence image data and child brain development priori map data, brain tissue region segmentation and interested region positioning are completed, positioning results of different brain regions are obtained, and time sequence signal characteristics and space structure characteristics of the different brain regions are extracted; Based on the historical data of the same age range denoising gold standard, the physiological motion synchronous acquisition data, the time sequence signal characteristics and the space structure characteristics, constructing a children brain function magnetic resonance multidimensional self-adaptive denoising model, combining the positioning results of different brain areas, generating a brain area noise type thermodynamic diagram, obtaining the distinguishing result of nerve signals and noise components of each brain area through a signal-noise separation algorithm, generating a brain distinguishing differential denoising strategy, and executing self-adaptive denoising processing on the function magnetic resonance original time sequence image data to obtain denoised function magnetic resonance image data; And synchronously calculating image signal-to-noise ratio evaluation data and brain region nerve signal fidelity data according to the denoised functional magnetic resonance image data, generating a denoised quality evaluation report and updating the parameter weight of the multidimensional adaptive denoising model.
2. The method for denoising functional magnetic resonance image data according to claim 1, wherein completing brain tissue region segmentation and region of interest localization, obtaining localization results of different brain regions, and extracting time sequence signal features and spatial structure features of different brain regions, comprises: Constructing a brain tissue segmentation and brain region positioning model of the child; based on the child brain tissue segmentation and brain region positioning model, the anatomical structure and the functional partition characteristics of the brain region are combined with the age-matched brain region in the child brain development priori map data, the region segmentation of brain gray matter, cerebrospinal fluid and skull non-brain tissue is completed through the improved 3DU-Net network, the spatial correlation characteristics of each functional brain region are captured by combining with the graph attention network, and the positioning of the region of interest is completed, so that the positioning results of different brain regions are obtained; Based on the positioning results of the different brain regions, extracting time sequence fluctuation characteristics and frequency domain characteristics of functional magnetic resonance signals of each brain region as time sequence signal characteristics, and extracting anatomical structure boundary characteristics, gray distribution characteristics and spatial texture characteristics of the brain region as spatial structure characteristics.
3. The method of denoising functional magnetic resonance image data according to claim 1, wherein generating a brain region noise type thermodynamic diagram comprises: constructing a regional space grid division model adapting to the brain development characteristics of children; Based on the regional space grid division model adapting to the brain development characteristics of the children, combining brain region boundaries and development key region labels in brain development priori map data of the children, noise fluctuation amplitude in time sequence signal characteristics and artifact distribution characteristics in space structure characteristics, and 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 the noise type confidence coefficient to a preset color gradient interval, superposing the noise density value and the noise type confidence coefficient to a three-dimensional brain region template of the children brain development priori map, and generating a brain region noise type thermodynamic diagram marked with the noise types, density levels, confidence coefficient and update time stamps of each grid unit.
4. The method for denoising functional magnetic resonance image data according to claim 1, wherein the distinguishing result of the neural signal and the noise component of each brain region is obtained by a signal-noise separation algorithm, comprising: constructing a brain region specific signal-noise separation model; According to the brain region specific signal-noise separation model, extracting the non-stationary fluctuation characteristic of each brain region time sequence signal by adopting a sliding window method, completing the component separation of the neurophysiologic signal and the motion noise, the physiological noise and the instrument noise by combining independent component analysis with a gradient lifting tree algorithm, and outputting the distinguishing result of the neurophysiologic signal and the noise component of each brain region; Based on the distinguishing result of the nerve signals and the noise components of each brain region, the noise component duty ratio of the corresponding brain region is obtained, and when the duty ratio of the noise component exceeds the noise threshold preset by the corresponding brain region, a brain region self-adaptive weighting denoising mechanism is triggered, so that a targeted denoising parameter correction instruction is generated.
5. The method for denoising functional magnetic resonance image data according to claim 1, wherein synchronously calculating image signal-to-noise ratio evaluation data and brain region nerve signal fidelity data comprises: Constructing a two-dimensional quantitative evaluation model of the magnetic resonance denoising effect of the brain function of the child; Based on the two-dimensional quantitative evaluation model of the magnetic resonance denoising effect of the brain function of the child, by combining denoised functional magnetic resonance image data, original time sequence image data of magnetic resonance of the brain function of the child, historical denoising gold standard data of the same age range and positioning results of different brain regions, calculating signal-to-noise ratios, peak signal-to-noise ratios and structural similarity indexes of the whole brain and each brain region through a medical image signal-to-noise ratio statistical algorithm to obtain image signal-to-noise ratio evaluation data, and simultaneously calculating time sequence correlation, low-frequency oscillation amplitude fidelity and functional connection strength fidelity of nerve signals of each brain region through a neurophysiologic feature matching algorithm to obtain brain region nerve signal fidelity data; And performing brain region matching association on the image signal-to-noise ratio evaluation data and brain region nerve signal fidelity data, and outputting a two-dimensional quantitative evaluation data set with brain region identification.
6. The method of denoising functional magnetic resonance image data according to claim 1, wherein the signal-to-noise separation algorithm formula: ; ; ; ; ; ; ; ; Wherein, the A magnetic resonance original time sequence image data matrix for brain functions of children; a design matrix for containing nerves and noise; Is a regression coefficient matrix; Neural signals in the form of residuals; is the first A brain region spatial mask matrix; is the first A brain region neural signal regression matrix; regression coefficients for neural signals; A regression matrix is referenced for noise; is a noise regression coefficient; is the first Brain region neural signal residual; is a signal mixing matrix; is an independent component set; Is the total noise matrix; Is that Filtering weight vectors at the moment; Is that Moment weight vector; Is a step size parameter; is zero-proof constant; Is that A time error signal; Is that A time-of-day noise reference vector; For an estimated neural signal matrix; to solve the operator optimally; Is regularized weight; a prior space regularization matrix is developed; Is a true neural signal; the wavelet coefficient after threshold processing; Is a wavelet coefficient sign function; is the original wavelet coefficient; is the first Denoising nerve signals in brain regions; is the first Brain region self-adaptive separation operator; A physiological motion noise reference matrix; A priori map for children development; the model weight parameters; Is functional magnetic resonance image data after denoising.
7. A denoising system of functional magnetic resonance image data, comprising: The data acquisition module acquires child brain function magnetic resonance original time sequence image data, child brain development priori map data, historical denoising gold standard data in the same age range and physiological motion synchronous acquisition data; The brain region feature extraction module is used for completing brain tissue region segmentation and region of interest positioning according to the child brain function magnetic resonance original time sequence image data and child brain development priori map data to obtain positioning results of different brain regions and extracting time sequence signal features and space structure features of the different brain regions; The self-adaptive denoising module is used for constructing a children brain function magnetic resonance multi-dimensional self-adaptive denoising model based on the historical same-age-period denoising gold standard data, the physiological motion synchronous acquisition data, the time sequence signal characteristics and the space structure characteristics, generating brain region noise type thermodynamic diagrams by combining the positioning results of different brain regions, obtaining the distinguishing result of nerve signals and noise components of each brain region through a signal-noise separation algorithm, generating a brain distinguishing-level differential denoising strategy, and executing self-adaptive denoising processing on the functional magnetic resonance original time sequence image data to obtain denoised functional magnetic resonance image data; and the evaluation optimization module synchronously calculates image signal-to-noise ratio evaluation data and brain region nerve signal fidelity data according to the denoised functional magnetic resonance image data, generates a denoised quality evaluation report and updates the parameter weight of the multidimensional self-adaptive denoise model.
8. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the method of denoising functional magnetic resonance image data as claimed in any one of claims 1 to 6.
9. A computer readable storage medium having stored thereon a computer program or instructions, which when executed, perform the method of denoising functional magnetic resonance image data according to any one of claims 1-6.

Description

Denoising method, system, equipment and medium for functional magnetic resonance image data Technical Field The invention belongs to the technical field of medical image processing, and particularly relates to a denoising method, a denoising system, denoising equipment and denoising medium for functional magnetic resonance image data. Background Functional magnetic resonance imaging is used as a noninvasive neuroimaging technology, has become a core tool for researching typical and atypical brain development and diagnosing neuro development disorder of children, can effectively capture brain blood oxygen level dependent signals, and reflects spontaneous activity and functional connection characteristics of brain neurons. Aiming at the denoising technology of the child brain fMRI image data, the denoising technology is a key premise for guaranteeing the accuracy of subsequent image analysis, the existing denoising method forms a multidimensional technical system, mainly comprises traditional filtering, head motion parameter regression, independent component analysis, volume censoring and the like, can pertinently remove various noises generated in the scanning process, can inhibit equipment related noises such as nonuniform magnetic field and electromagnetic interference of a scanner, can also relieve head motion artifacts caused by difficulty in focusing of movements and attention of children, and can also weaken signal interference caused by physiological activities such as respiration and heartbeat. The application of the prior art effectively improves the signal-to-noise ratio of the child brain fMRI image, reduces the interference of noise on brain function network analysis and lesion feature identification, provides basic data support for early screening and mechanism exploration of the child brain development track research, epilepsy, attention deficit hyperactivity disorder and other nerve development diseases, and promotes the development of pediatric neuroimaging and the transformation of clinical application. The existing children brain function magnetic resonance image denoising technology has a plurality of difficult problems, and the problems that the existing technology is difficult to fully acquire and combine children brain development priori information, age-segment denoising gold standard data and synchronously acquired physiological motion data, is difficult to develop due to the fact that the existing technology depends on adult data, cannot adapt to special physiological behavior characteristics of children with high respiratory frequency and low-amplitude high-frequency motion of the head, so that accurate distinction of noise and real nerve signals is difficult to realize, accurate denoising is difficult to carry out on brain tissue areas, accurate segmentation and interested area positioning are difficult to carry out, time sequence signal characteristics and space structure characteristics of each brain area are also difficult to effectively extract, noise types of different brain areas are difficult to accurately analyze, a targeted brain distinction level differentiation denoising strategy is further difficult to form, the problem that the noise difference of each brain area is large is difficult to deal with, a multi-dimensional self-adaptive denoising model suitable for children is difficult to be constructed, high-efficient and accurate signal-noise separation cannot be realized due to complex scenes of equipment noise, motion artifacts and physiological noise superposition, denoising accuracy is difficult to synchronously carry out on denoising quality, and reliability of the data of the weight and the real-time computing quality is difficult to update of the data. Disclosure of Invention The application provides a denoising method, a denoising system, denoising equipment and denoising media for functional magnetic resonance image data, which aim to solve the problems of insufficient denoising accuracy and the like in the prior art. The embodiment of the first aspect of the application provides a denoising method for functional magnetic resonance image data, which comprises the following steps of obtaining child brain functional magnetic resonance original time sequence image data, child brain development priori map data, historical same-age-stage denoising gold standard data and physiological motion synchronous acquisition data, completing brain tissue region segmentation and interested region positioning according to the child brain functional magnetic resonance original time sequence image data and the child brain development priori map data, obtaining positioning results of different brain regions, extracting time sequence signal characteristics and space structure characteristics of the different brain regions, constructing a child brain functional magnetic resonance multidimensional adaptive denoising model based on the historical same-age-stage denoising gold standard d