A method and system for predicting brain age of multiple fetuses

Abstract

The present invention provides a method and system for predicting the brain age of multiple fetuses, which relates to the technical field of image data processing, including: original image data; performing image segmentation and motion artifact correction processing according to the original image data to obtain separated images; reconstructing mutually occluded areas in the separated images to generate reconstructed images; performing feature extraction processing according to the reconstructed images to obtain brain development feature data of each fetus, and performing difference analysis of relative development levels based on the brain development feature data to generate development difference data; performing fusion processing according to the brain development feature data and the development difference data to evaluate the actual brain age of each fetus and output the brain age prediction result. The present invention solves the problems of image blur and structural overlap caused by mutual occlusion, fetal movement and signal interference in multiple fetuses in the mother's body through image segmentation, motion correction and signal interference compensation technology, and significantly improves the accuracy of brain age prediction of multiple fetuses.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and in particular to a method and system for predicting the brain age of multiple fetuses. Background Art

[0002] At present, the monitoring of fetal brain development mainly relies on magnetic resonance imaging (MRI), which can provide high-resolution brain structure images under non-invasive and radiation-free conditions. However, in the case of multiple pregnancies, existing technologies have significant technical bottlenecks and deficiencies in brain development detection. First, due to the close position of multiple fetuses in the mother's body, there is usually a problem of mutual occlusion between fetuses, which makes it difficult for traditional image segmentation algorithms to effectively separate the independent brain structure of each fetus, resulting in overlapping brain tissue. Secondly, in the environment of multiple fetuses, frequent fetal movements and irregular movements between fetuses will produce significant motion artifacts, further exacerbating the ambiguity and instability of the image. Traditional motion correction technology can usually only perform simple corrections on a single fetus, but lacks flexibility for the complex movement patterns of multiple fetuses, and it is difficult to synchronously process motion artifacts of multiple fetuses.

[0003] Based on the shortcomings of the above-mentioned existing methods, there is an urgent need for a method and system for predicting the brain age of fetuses in multiple births. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for predicting the brain age of multiple fetuses to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present application provides a method for predicting the brain age of multiple fetuses, comprising:

[0006] Acquiring original image data, wherein the original image data is magnetic resonance image data collected during the pregnancy of a pregnant woman with multiple births;

[0007] Performing image segmentation and motion artifact correction processing according to the original image data to obtain a separated image;

[0008] Reconstructing mutually occluded areas in the separated images, separating and compensating for signal interference between adjacent fetuses using a non-local means algorithm, and restoring and reconstructing the images in combination with the positional relationship of each fetus in three-dimensional space to generate a reconstructed image;

[0009] Performing feature extraction processing according to the reconstructed image to obtain brain development feature data of each fetus, and performing difference analysis of relative development levels among multiple fetuses based on the brain development feature data to generate development difference data;

[0010] The brain development characteristic data and the development difference data are fused and processed, and a regression analysis model is constructed based on the fused data set to evaluate the actual brain age of each fetus and output a brain age prediction result.

[0011] In a second aspect, the present application also provides a system for predicting brain age of multiple fetuses, including:

[0012] An acquisition module, used to acquire original image data, wherein the original image data is magnetic resonance image data collected during the pregnancy of a pregnant woman with multiple births;

[0013] A segmentation module, used for performing image segmentation and motion artifact correction processing according to the original image data to obtain a separated image;

[0014] A restoration module, used to reconstruct mutually occluded areas in the separated images, separate and compensate for signal interference between adjacent fetuses through a non-local means algorithm, and restore and reconstruct the images in combination with the positional relationship of each fetus in three-dimensional space to generate a reconstructed image;

[0015] An extraction module, configured to perform feature extraction processing based on the reconstructed image to obtain brain development feature data of each fetus, and to perform difference analysis of relative development levels among multiple fetuses based on the brain development feature data to generate development difference data;

[0016] The prediction module is used to perform fusion processing based on the brain development feature data and the development difference data, and to build a regression analysis model based on the fused data set to evaluate the actual brain age of each fetus and output a brain age prediction result.

[0017] The beneficial effects of the present invention are:

[0018] The present invention solves the problems of image blur and structural overlap caused by mutual occlusion, fetal movement and signal interference of multiple fetuses in the mother's body through precise image segmentation, motion correction and signal interference compensation technology, and significantly improves the separation accuracy of fetal brain magnetic resonance images; the present invention deeply extracts brain features such as gray matter, white matter, cortical thickness, etc., and conducts quantitative analysis in combination with the impact of maternal resource competition on fetal development, so as to better identify developmental differences between fetuses.

[0019] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 A schematic flow chart of a method for predicting fetal brain age in multiple births according to an embodiment of the present invention;

[0022] Figure 2 Schematic diagram of the structure of the system for predicting brain age of multiple fetuses described in an embodiment of the present invention.

[0023] Markings in the figure: 1. Acquisition module; 2. Segmentation module; 21. First segmentation unit; 22. Second segmentation unit; 23. First correction unit; 24. First processing unit; 3. Repair module; 31. First analysis unit; 32. First decomposition unit; 33. First compensation unit; 34. First repair unit; 4. Extraction module; 41. Third segmentation unit; 42. Second analysis unit; 43. First simulation unit; 44. Third analysis unit; 5. Prediction module; 51. Second simulation unit; 52. Fourth analysis unit; 53. First fusion unit; 54. First prediction unit. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0026] Embodiment 1:

[0027] This embodiment provides a method for predicting the brain age of fetuses in multiple births.

[0028] See also Figure 1 , the figure shows that the method includes steps S100 to S500.

[0029] Step S100, obtaining original image data, where the original image data is magnetic resonance image data collected during the pregnancy of a pregnant woman with multiple births;

[0030] Understandably, magnetic resonance imaging (MRI) has become the main tool for fetal brain monitoring because of its non-invasive, radiation-free, and high-resolution image acquisition characteristics. For pregnant women, especially in the case of multiple births, magnetic resonance imaging can effectively capture the brain structure of each fetus and provide raw data for further developmental analysis. This imaging technology uses strong magnetic fields and radio frequency pulses to generate detailed brain images, which can reveal important information such as the anatomical characteristics of the fetal brain and differences in tissue density.

[0031] Step S200, performing image segmentation and motion artifact correction processing according to the original image data to obtain a separated image;

[0032] It is understandable that in the scenario of multiple births, the challenge is that the brain structures of multiple fetuses will overlap or occlude each other, so a more sophisticated segmentation algorithm is needed. Conventional segmentation methods, such as edge detection-based techniques, cannot effectively deal with the complex structural interference in multiple births. This step can more accurately separate the brain area of ​​each fetus by analyzing the density difference between the gray matter and the surrounding tissue in the image. In addition, although the magnetic resonance image is static, the artifacts caused by fetal movement will blur the boundaries of certain brain structures. This step further uses motion correction technology to identify the displacement artifacts caused by fetal movement by analyzing the local block consistency and spatial correlation in the image, and make corresponding corrections to ensure the accuracy of the image.

[0033] Step S300, reconstructing mutually occluded areas in the separated images, separating and compensating for signal interference between adjacent fetuses using a non-local means algorithm, and restoring and reconstructing the images in combination with the positional relationship of each fetus in three-dimensional space to generate a reconstructed image;

[0034] It can be understood that, based on the preliminary separation of the magnetic resonance image data, this step uses the non-local mean algorithm to remove and compensate for the signal interference of adjacent fetuses. The uniqueness of the non-local mean algorithm is that it can filter out the interference caused by the overlap of adjacent fetal signals by comparing and weighted averaging similar local blocks in the image. Unlike traditional filtering methods based on local information, the non-local mean algorithm can better preserve the details in the brain structure and avoid losing important anatomical information in the process of removing interference. Through this algorithm, signal interference is effectively compensated and image details are preserved. At the same time, this step restores the image details in combination with the three-dimensional spatial position of the fetus. By aligning the separated image with the anatomical structure of the fetus in three-dimensional space, the restoration effect of the occluded area can be further enhanced, ensuring that the final reconstructed image has higher accuracy and completeness.

[0035] Step S400, performing feature extraction processing based on the reconstructed image to obtain brain development feature data of each fetus, and performing difference analysis of relative development levels among multiple fetuses based on the brain development feature data to generate development difference data;

[0036] It is understandable that feature extraction and developmental difference analysis effectively quantified the brain development of each fetus and revealed the developmental differences between multiple births. This not only helps to accurately assess the individual development of each fetus, but also identifies potential developmental abnormalities or delays, ensuring comprehensive developmental monitoring in a multiple birth environment.

[0037] Step S500: perform fusion processing on the brain development feature data and the development difference data, and build a regression analysis model based on the fused data set to evaluate the actual brain age of each fetus and output a brain age prediction result.

[0038] It is understandable that when performing regression analysis, a suitable regression model is selected, preferably linear regression, nonlinear regression or a regression method based on machine learning, to establish a mapping relationship between fetal brain age and developmental characteristic data. The brain characteristics of each fetus will affect its developmental process, and the developmental differences between multiple fetuses, such as uneven development due to resource competition, also need to be considered in the analysis. The regression model predicts the actual brain age of each fetus by integrating these complex variables, which significantly improves the accuracy of the prediction. Especially in the case of multiple births, it can accurately assess the development level of each fetus, identify potential developmental delays or abnormalities, and provide an important basis for clinical intervention.

[0039] Further, step S200 includes step S210 to step S240.

[0040] Step S210, performing preliminary segmentation processing based on the original image data, analyzing the local gradient changes between the fetal brain and surrounding tissues to distinguish the brain structures of multiple fetuses, defining initial segmentation boundaries, and generating a preliminary brain contour image of each fetus;

[0041] It should be noted that pre-segmentation is the first step in image segmentation, and the focus is on providing a good initial contour framework for subsequent fine segmentation. In this process, there are significant differences in the density and signal intensity of the fetal brain and surrounding tissues, and these differences are manifested as gradient changes in the image. By calculating these local gradients, the boundaries between the fetal brain and maternal tissue or adjacent fetuses can be identified. In general, the segmentation algorithm based on gradient changes can quickly delineate the boundary area to ensure that the main structure of the fetal brain can be captured. Due to the close position of the fetuses in multiple pregnancies, the brain structures of adjacent fetuses may overlap in the image. Therefore, sensitive detection of gradient changes can effectively distinguish the independent brain structures of multiple fetuses. The main effect of pre-segmentation is to form a rough segmentation result by preliminarily identifying the brain contour of each fetus, laying the foundation for subsequent precise segmentation. Since this step mainly provides preliminary contours, the algorithm will maximize the accuracy of the boundary while ensuring the segmentation speed. At the same time, this pre-segmentation can effectively reduce the complexity of subsequent processing, because the preliminary segmentation boundary helps to quickly exclude irrelevant areas. Specifically, this embodiment uses a combination of gradient and Laplace operator to process local image changes to ensure that the brain boundary can be detected. The Laplace operator helps detect changes in the second-order derivative, thereby capturing the image edge more accurately. The calculation formula is as follows:

[0042] ;

[0043] In order to reduce noise, the convolution operation is introduced to smooth the image. This process is completed by convolution with a Gaussian kernel. Convolution helps to eliminate unnecessary noise interference and make the boundaries clearer. The formula is as follows:

[0044] ;

[0045] Furthermore, by combining the gradient change with the Laplace operator results, multiple threshold functions are added, and the boundary is strengthened through piecewise functions. This method can automatically adjust the sensitivity of boundary detection according to different boundary strengths. The formula is as follows:

[0046] ;

[0047] in, and represent the abscissa and ordinate in the magnetic resonance image, respectively; Indicated in coordinates The result of combining the gradient and Laplacian operator at ; Indicated in coordinates The pixel intensity value at ; Indicated in coordinates The image is smoothed by the convolution operation; represents Gaussian kernel; represents the fine edge image obtained by piecewise function; represents an adaptive function based on local gradient changes, used to adjust the boundary strength within the threshold range; and Represent the upper and lower thresholds respectively.

[0048] Step S220, segmenting the preliminary brain contour image based on a preset Markov random field, and obtaining an optimized brain image by modeling the probability of spatial correlation between pixels and optimizing the boundaries of multiple adjacent tissues;

[0049] Specifically, the Markov random field model considers the correlation between adjacent pixels and treats adjacent pixels as a whole, thereby better describing the continuity and smoothness in the image. In the case of multiple births, the fetal brain structure may have blurred boundaries due to the interference of adjacent tissues. It is difficult for ordinary segmentation methods to distinguish the transition area between adjacent fetal brain tissues, and the introduction of Markov random fields solves this problem. By probabilistically modeling the relationship between each pixel and its surrounding pixels, the Markov random field can balance the spatial continuity and local details of the image from a global perspective, avoiding the problem of abrupt or overly smooth boundaries. The segmentation algorithm of the Markov random field is based on the pixel similarity and spatial correlation in the image, further refining the boundaries of the preliminary segmentation, making the segmentation of the brain contour more accurate. In response to the tissue interference problem of multiple fetuses, the Markov random field can effectively identify the boundaries of adjacent fetuses by integrating the mutual relationship of multiple pixels, and optimize these boundaries using spatial correlation. For example, when the brain structures of adjacent fetuses are close, the Markov random field can identify the difference between the two through a probabilistic model, thereby optimizing the boundaries and reducing the inaccurate segmentation caused by the interference of adjacent tissues.

[0050] Step S230, performing spatial consistency correction processing according to the optimized brain image, evaluating the consistency of the fetal brain structure in multiple scanned images by comparing the similarity measurement of local blocks, and performing error adjustment to obtain an aligned brain image after error correction;

[0051] It should be noted that this step decomposes the optimized brain image into multiple local areas (i.e., "blocks"), and by comparing the features of these local blocks, their similarities at different locations are determined. For example, specific areas of the brain (such as gray matter or white matter) maintain the same morphological and structural features at different locations. By calculating the similarity of these local blocks, those inconsistent areas that may be caused by noise or segmentation errors are identified and adjusted. This process can effectively correct local deformations or inconsistencies caused by image processing or segmentation errors. By measuring the similarity of local blocks, the errors in local areas are corrected, and the consistency of the entire brain image in three-dimensional space is ensured. Especially in the scenario of multiple births, the fetal brain may be locally distorted due to interference from adjacent fetuses or image noise. This correction process can effectively eliminate these errors and make the brain structure more consistent and stable.

[0052] Step S240: reconstruct the fetal brain boundary according to the aligned brain image, preliminarily reconstruct the brain contour by using the convex segmentation technique, and optimize the continuity and smoothness of the boundary by using the boundary constraint method to obtain a separated image.

[0053] It is understandable that when adjacent fetal brain tissues are close, the advantage of convex segmentation technology is that it can identify the boundaries of each fetal brain and ensure that each boundary can remain independent. Specifically, the convex segmentation technology generates a convex outline based on the geometric shape of the brain tissue, and reasonably distinguishes all adjacent tissues so that each fetal brain structure can exist independently. In addition, the convex segmentation technology can also avoid the problem of over-segmentation or unclear boundaries in complex structural environments. At the same time, in order to prevent irregular boundaries caused by adjacent fetuses or fetal movements, this step also uses a boundary control method. The boundary control method constrains the continuity and smoothness of the boundary to ensure that the generated boundary can not only tightly surround the brain tissue, but also remain natural and smooth in shape. In the specific implementation process, the boundary control smoothes the irregular edges through an optimization algorithm to eliminate boundary distortion caused by fetal movement or position changes, thereby generating a boundary that conforms to the biological morphological characteristics.

[0054] Further, step S220 includes step S221 to step S224.

[0055] Step S221, performing initialization processing of the Markov random field according to the preliminary brain contour image, by constructing the state space of each pixel point, and establishing the initial probability distribution between pixels based on the correlation between its gray value and the surrounding pixels, to obtain a preliminary probability map of the pixel state;

[0056] Specifically, first, for each pixel, the state space of each pixel is constructed based on the correlation between its grayscale value and the surrounding pixels. The process of constructing the state space involves encoding the state of each pixel, which usually includes two possible states: belonging to the fetal brain area or belonging to other tissues. This association is handled by using the local smoothness assumption between pixels, that is, adjacent pixels are more likely to have the same or similar states. In order to further capture this correlation, the Markov random field model will count the grayscale value of each pixel and the grayscale value of its neighboring pixels to calculate the conditional probability of each pixel state. These conditional probabilities reflect the probability that the current pixel belongs to a certain state given the state of the neighboring pixels. In this way, a preliminary probability distribution can be generated for each pixel of the entire image. The formula is as follows:

[0057] ;

[0058] in, A serial number indicating the status; Indicates the pixel number; Represents pixels The set of pixels in the field; Represents pixels The state of the neighbor pixel set; Represents pixels In state The probability of , given its adjacent pixels (neighborhood pixels) Status; Indicates the sequence number of neighbor pixels; Represents pixels the state of the brain region or of other tissues; Represents a normalization constant, which is used to ensure that pixels The sum of all state probabilities is 1; is the potential function, representing the pixel Its neighbor pixels In Status The relevance of.

[0059] Step S222, performing maximum a posteriori estimation processing according to the preliminary probability map, by combining local smoothness and global spatial consistency between pixels, using the maximum a posteriori estimation in the Bayesian framework to update the state of each pixel, and obtaining a maximum a posteriori estimation map of the pixel state;

[0060] It is understandable that the preliminary probability map contains the initial probability distribution of each pixel belonging to brain tissue or other tissue. In order to further optimize these probabilities, the maximum a posteriori estimation combines the assumption of local smoothness of pixels, that is, adjacent pixels have a higher probability of being in the same state. At the same time, the global spatial consistency is considered to ensure that the pixel state is consistent with the brain structure of the entire image. The core idea of ​​the maximum a posteriori estimation is to optimize the pixel state through Bayes' theorem to balance the local characteristics of the pixel with the global spatial information. During the calculation process, the state of the pixel depends not only on the intensity value of the current pixel, but also on the influence of its neighboring pixels, ensuring that the image segmentation result conforms to the characteristics of the brain structure both locally and globally.

[0061] Step S223, performing tissue interference processing according to the maximum a posteriori estimation map, identifying the tissue overlap between adjacent fetuses by introducing constraints on adjacent regions of multiple fetuses, and marking interfering pixels based on the spatial continuity of the boundaries to obtain an interference region marking map;

[0062] It is understandable that in order to deal with the situation where the brain structures of adjacent fetuses may overlap or interfere with each other, it is necessary to introduce spatial constraints on adjacent regions of multiple fetuses. These constraints determine possible overlapping areas by analyzing the spatial relationship between different fetal brains. The overlapping parts usually appear as uncertainty in the pixel state, so these pixels need to be specially processed. Next, the spatial continuity of the boundary is used as the key criterion for interference detection. At the boundary where the fetal brain region borders the adjacent tissue, the spatial distribution and edge characteristics of the pixels are analyzed to ensure the natural smoothness and coherence of the boundary. Those pixels that show abnormalities in boundary continuity (such as sudden boundary discontinuities or mutations) will be marked as interference pixels. Finally, an interference area marking map is formed, which clearly indicates which pixels are in the interference area between the fetal brain and the adjacent tissue, providing a clear processing target for subsequent boundary optimization and tissue separation.

[0063] Step S224: perform boundary optimization processing according to the interference region marking map, and adjust the probability distribution between pixels in the marked interference region by combining Gaussian smoothing and gradient enhancement algorithms to obtain an optimized brain image.

[0064] It is understandable that Gaussian smoothing helps reduce the noise in the image and makes the intensity changes between pixels smoother, especially at the junction of adjacent tissues, which can effectively remove random interference signals. This step uses the convolution operation to locally smooth the pixels in the interference area, making the pixel intensity in these areas more uniform and reducing the drastic changes caused by interference. Next, the gradient enhancement algorithm is used to restore and enhance the boundary information. Gradient enhancement enhances edge features by calculating the gradient strength of pixels in the image. After Gaussian smoothing, blurred edges can be made clear again through gradient enhancement. The core idea of ​​the gradient enhancement algorithm is to enhance the areas with larger pixel gradients so that the edges in the image can be clearly reflected in the interference area, ensuring that the boundaries between brain tissue and adjacent tissues are clear.

[0065] Further, step S300 includes step S310 to step S340.

[0066] Step S310: Analyze the local brightness difference between adjacent fetal brain signals according to the separated images, and identify the occlusion area caused by signal intensity mutation and spatial discontinuity in combination with spatial continuity detection, and obtain a marked adjacent occlusion area map;

[0067] Specifically, first, different areas of the fetal brain have different signal intensities (i.e., brightness) in magnetic resonance images, and structures such as gray matter, white matter, and meninges will show specific brightness distributions in the images. When the brain signals of adjacent fetuses interfere with each other, it usually causes abnormal changes in the brightness of these areas, such as sudden brightness enhancement or reduction. This abnormal local brightness change can be used as a preliminary basis for identifying occluded areas. By analyzing the brightness gradient changes in adjacent areas, areas where signal overlap or interference may occur can be quickly located. Then, spatial continuity detection is used to further confirm the occluded area. Under normal circumstances, the signal of the fetal brain is continuous in space, that is, the signal intensity of adjacent pixels should transition smoothly. When spatial discontinuity occurs, the signal intensity suddenly jumps, indicating that there may be occlusion or interference. By detecting these spatially discontinuous areas, the occluded areas caused by interference from adjacent fetal signals can be more accurately identified, and combined with brightness difference analysis, the detection accuracy can be improved. By combining these two methods, a labeled adjacent occluded area map is finally generated, which can clearly identify the areas in the fetal brain image that are occluded and signal interfered.

[0068] Step S320: performing local feature decomposition processing according to the marked adjacent occlusion region map, analyzing the sources of different signals in the occlusion region through gradient flow, distinguishing the overlapping signals of the fetal brain and adjacent tissues, and obtaining a feature map of the occlusion region after signal separation processing;

[0069] It should be noted that gradient flow is a tool for analyzing the direction and intensity of signal changes in an image, which can effectively identify the source of the signal. In the occluded area, signals from different fetal brains or adjacent tissues will show different gradient changes. For example, signals from the fetal brain usually have a specific directionality, while signals from adjacent tissues may show different gradient characteristics in different directions. By analyzing the gradient flow of the signal, signals from these different sources can be distinguished. This process is equivalent to deconstructing the signal source of the image and identifying which signals belong to the fetal brain structure and which signals come from adjacent interfering tissues.

[0070] The decomposition of local features further refines this process. It decomposes the occluded area into smaller local blocks and analyzes the signal features in each local block. These local features include brightness, gradient changes, texture, etc. Through feature decomposition, overlapping signals in the occluded area can be processed more carefully to ensure that each signal source is processed independently. Decomposition processing helps to remove interfering signals from adjacent tissues, making the boundaries of the fetal brain clearer.

[0071] After this process, a feature map of the occluded area after signal separation is generated, which clearly distinguishes the fetal brain signal from the signal of the adjacent tissue and successfully separates the overlapping parts. This step not only restores the signal integrity of the occluded area, but also ensures that in a multiple fetus environment, the brain structure of each fetus can be independently extracted and analyzed.

[0072] Step S330, performing tissue morphology compensation processing according to the characteristic map of the occluded area, using the brain morphology structure of the unoccluded area to perform multi-resolution morphology inference on the occluded area, gradually compensating for the missing morphological information of the occluded area, and obtaining a preliminary repaired image;

[0073] It is understandable that, due to the certain symmetry and biological regularity of brain structure, the unobstructed brain area can provide a reference for inferring the occluded area. In the specific processing process, the morphological characteristics of the unobstructed area, such as the distribution of gray matter and white matter, the complexity of the sulci, etc., will be extracted and used to guide the compensation of the occluded area. Multi-resolution morphological inference can gradually restore the lost morphological information at different resolution levels. First, the overall structure is inferred and compensated at a lower resolution level to restore the basic morphological outline. Then, as the resolution increases, the compensation of the details is increased layer by layer, and finally the tiny morphological details are supplemented at the highest resolution level. This layer-by-layer inference method ensures the accuracy and continuity of morphological compensation and avoids the morphological distortion problem that may occur during one-time recovery. Multi-resolution processing can better capture the overall characteristics and local details of brain morphology, and is especially suitable for the recovery of complex brain structures.

[0074] Step S340: Perform morphological restoration processing based on the preliminary restoration image, reconstruct the three-dimensional structure of the occluded area by using the three-dimensional morphological information of the unoccluded area and the known anatomical features of the fetal brain, and perform detail restoration under morphological constraints to obtain the final reconstructed image.

[0075] It should be noted that morphological constraints avoid abrupt morphological transitions or irregular boundaries in the repair results by limiting the smoothness and continuity of the boundaries during the repair process. In specific implementation, morphological constraints will perform detailed repairs on each local structure in the repair area so that the boundaries, surface curvatures, and tissue structures are consistent with the structures in the unobstructed areas. Detail repair further optimizes the microscopic features of the occluded area, so that the repaired brain morphology is not only consistent in the macroscopic sense, but also consistent with the actual anatomical features in detail. The final repaired image has a complete three-dimensional structure, and the morphology of the occluded area is not only reconstructed, but also seamlessly connected with the surrounding undisturbed areas. After morphological repair, each fetal brain structure in the image can be accurately presented, and the structural continuity and morphological consistency are guaranteed.

[0076] Further, step S310 includes step S311 to step S314.

[0077] Step S311, performing multi-scale structure tensor analysis processing on brightness features according to the separated image, and obtaining a multi-scale brightness change map by using the multi-scale structure tensor to capture brightness changes between adjacent pixels at different scales;

[0078] It is understandable that during the processing, the multi-scale structure tensor analyzes the brightness gradient and calculates the brightness direction change at different scales. In this way, the boundaries at multiple scales can be identified, especially in the brightness transition area, such as the junction between the fetal brain and adjacent tissues. The resulting multi-scale brightness change map can more accurately reflect the complex signal interference area.

[0079] Step S312, performing feature association network processing according to the multi-scale brightness change map, calculating the non-local brightness similarity between different regions by constructing a non-local feature association network, identifying possible occlusion regions, and obtaining a non-local occlusion association map;

[0080] It should be noted that the non-local feature association network can capture the long-distance dependencies between the fetal brain and adjacent tissues by comparing the similarities of distant pixels in the image, especially in complex multiple birth scenes. Unlike local methods, this non-local algorithm can analyze the brightness similarity of adjacent regions over a larger range, thereby accurately capturing regional inconsistencies caused by occlusion or interference. This step constructs a brightness similarity network between different regions, identifies those regions with similar brightness of distant pixels but different spatial distributions, and generates a non-local occlusion association map.

[0081] Step S313: performing local complexity enhancement processing according to the non-local occlusion association graph, identifying occlusion regions with high complexity by enhancing the complexity of the occlusion regions, and obtaining an occlusion complexity graph;

[0082] Preferably, this step uses local fractal dimension analysis to enhance the complexity of the occluded area. Local fractal dimension analysis reflects the geometric complexity of the area around the pixel by calculating the local fractal dimension of the image. In the scenario of multiple births, the junction of the brain tissues of adjacent fetuses usually has a higher geometric complexity. These complex areas can be further enhanced by the local fractal dimension analysis method, making the occluded area caused by signal interference clearer. This method highlights the boundary features of the occluded area by enhancing the local complexity and generates an occlusion complexity map.

[0083] Step S314: perform heterogeneous pixel classification processing according to the occlusion complexity map, determine the boundary of the occlusion area by classifying the pixels in the occlusion area, and obtain a marked adjacent occlusion area map.

[0084] Preferably, this step uses multi-task sparse coding to classify the pixels in the occluded area. Multi-task sparse coding is a technology that uses a multi-task learning framework to separate different signal sources in an image. Multi-task sparse coding can effectively classify the pixels in the occluded area by establishing multiple tasks in the sparse coding framework and treating different occluded areas as independent tasks, thereby distinguishing the signals belonging to different fetuses. Finally, a labeled adjacent occluded area map is generated, so that the occluded area in complex scenes can be accurately segmented.

[0085] Further, step S400 includes step S410 to step S440.

[0086] Step S410, performing adaptive morphological segmentation processing according to the reconstructed image, by analyzing the spatial limitations of multiple fetuses in the uterus and the asymmetry of the brain structure morphology, dynamically adjusting the segmentation threshold to extract the gray matter area and the white matter area, and calculating the gray matter volume data and the white matter volume data of each fetus;

[0087] It should be noted that, first of all, adaptive morphological segmentation dynamically adjusts the segmentation threshold to adapt to the differences in brain structure of each fetus in a specific spatial environment. Traditional segmentation methods are usually based on fixed thresholds. In the case of multiple fetuses, the limited space between fetuses will lead to differences in brain tissue density and morphology. The use of fixed thresholds will lead to missegmentation or unclear boundaries. The adaptive method adjusts the segmentation threshold in real time by analyzing the local gradient changes and spatial distribution of fetal brain tissue to better capture the boundaries of gray matter and white matter. This dynamic adjustment process enables the segmentation process to adaptively handle morphological differences and local irregularities between different fetuses to ensure the accuracy of segmentation. During the extraction process, gray matter and white matter regions are segmented by analyzing different densities and signal intensities. Gray matter and white matter appear differently in magnetic resonance images, and the differences in signal intensity and contrast allow them to be effectively distinguished. Using adaptive threshold adjustment, these two key tissues can be accurately distinguished in the case of asymmetric morphology or complex structure.

[0088] Step S420, performing developmental delay analysis based on brain gray matter regions, establishing a model of cortical thickness changes during normal development to identify abnormal thickness regions, and correcting the abnormalities through dynamic compensation to generate cortical thickness distribution data for each fetus;

[0089] It is understandable that developmental delay analysis compares the cerebral cortical thickness of each fetus with the normal value of the standard development stage through a detailed analysis of the gray matter area of ​​the brain. This process integrates a large number of cortical thickness change data of normal fetuses at different stages, thereby constructing a standard cortical thickness trend curve that changes dynamically over time. By comparing the actual cortical thickness of each fetus with this standard curve, it is possible to quickly identify which areas have developmental delays or abnormalities. Once these thickness deviation areas are identified, dynamic compensation is required. The purpose of dynamic compensation is to estimate the potential normal development trend of the area based on the current cortical thickness data of the fetus, and to make appropriate corrections to the abnormal areas. This compensation is not simply to increase the thickness, but to determine whether the area is likely to catch up with the normal development progress in the subsequent development process by analyzing the spatial distribution and development pattern of cortical thickness. Dynamic compensation is usually achieved by adjusting the parameters of the cortical thickness distribution so that the deviation area is reasonably supplemented in the model to ensure that the compensated thickness is consistent with the normal development trajectory. The final cortical thickness distribution data can accurately describe the cortical thickness of each fetus in different brain regions.

[0090] Step S430, performing simulation processing according to the cortical thickness distribution data, simulating the developmental competition of multiple fetuses in the case of sharing maternal resources, and using fractal geometry to analyze the efficiency of resource allocation between multiple fetuses, evaluating the developmental complexity of brain sulci and gyri, and obtaining fractal dimension data of each fetus;

[0091] Specifically, multiple fetuses share the mother's nutrition and blood supply at the same time, and face unequal resource distribution during development. This resource competition may affect the speed and pattern of fetal brain development. By simulating the distribution of maternal resources among multiple fetuses, the relative advantages or disadvantages of each fetus in the competition for resources can be reflected. For example, some fetuses may obtain more resources due to their positional advantages, thus developing faster, while other fetuses may be affected and develop relatively slowly. The simulation process analyzes how the thickness of the cerebral cortex of each fetus is affected by resource competition based on the cortical thickness distribution data, revealing the importance of resource allocation in the development of multiple fetuses. Then, fractal geometry analysis is used to further evaluate the complexity of fetal brain development. The complexity of the sulci reflects the structure and developmental maturity of the brain, and fractal geometry is a mathematical tool used to describe complex and irregular shapes, which can effectively capture the morphological characteristics of the sulci. Through fractal geometry analysis, the fractal dimension data of different regions of each fetal brain can be calculated, which can quantify the complexity of the fetal brain structure. For example, the higher the fractal dimension, the more complex the folding of the sulci, reflecting that the brain is more mature. Finally, the fractal dimension data provided a quantitative indicator of the complexity of each fetus' brain structure. This data can be used to clearly assess the developmental differences between multiple fetuses, especially in the case of developmental competition caused by shared resources, which fetuses have more complex development and which fetuses may be underdeveloped due to resource limitations.

[0092] Step S440: Perform data fusion and correlation analysis based on the gray matter volume data, white matter volume data, cortical thickness distribution data, and fractal dimension data, and obtain developmental difference data by analyzing the developmental interactions between fetuses.

[0093] It should be noted that this step can deeply evaluate the development of multiple fetuses through data fusion and association analysis, revealing complex developmental mechanisms and potential risks. The ratio of gray matter volume to white matter volume may reveal the imbalance in the development of neural circuits in some fetuses. For example, if the gray matter volume of a fetus is significantly larger than the white matter volume, it may indicate excessive growth in dense neuron areas, which in turn affects the coordination of brain function, which may be due to the concentrated allocation of maternal resources. The white matter volume of another fetus is larger, which may show the early maturation of nerve fiber connections, but the cortical thickness is thinner. This situation may indicate that the development of the fetus's cerebral cortex is delayed, resulting in incomplete development of cognitive function. In addition, the complexity of fractal dimension data can reveal the degree of folding and structural complexity of brain sulci. For fetuses with lower fractal dimensions, although they may have a larger cortical thickness, their complexity of brain sulci is poor, which may indicate that the development of brain functional partitions is not yet perfect, and there are hidden dangers in the development of future cognitive abilities. Although fetuses with higher fractal dimensions are smaller in size, they may show a high degree of complexity in the development of brain sulci, which indicates their potential developmental advantages in neural networks. These complex interactions can reveal the impact of resource allocation, fetal position and maternal status on the different developmental paths of each fetus through data fusion. The generated multiple fetal developmental difference data provide an important reference for identifying fetuses at risk of future developmental delays.

[0094] Further, step S500 includes step S510 to step S440.

[0095] Step S510: Based on the developmental difference data, simulate the competition process of each fetus in blood supply and nutrient distribution, quantitatively evaluate the specific impact of resource competition on the development of each fetus, and generate developmental competition impact data for each fetus;

[0096] It should be noted that resource competition is not only based on the single blood and nutrient supply, but also involves differences in the fetus's absorption and utilization efficiency of resources, which may be affected by complex factors such as fetal weight, development speed, placental position, and maternal physiological state. For some fetuses, although they obtain more nutrition, their developmental mechanisms may have low resource utilization efficiency, resulting in actual development that is not as good as fetuses with less resource acquisition but high utilization rate. For example, a fetus close to the main blood supply channel of the placenta should theoretically obtain more blood and nutrient supply, but if this fetus has low absorption efficiency of certain key nutrients (such as oxygen and glucose) during brain development due to genes or biological environment, simulation processing can reveal this potential developmental bottleneck. Conversely, another fetus far away from the main blood supply source, although the amount of resources obtained is small, its high absorption and metabolic efficiency of key nutrients can make up for the lack of resources, and it shows a more balanced development in the development of gray matter, white matter and cortical thickness of the brain. In addition, the dynamics of resource competition is also a complex factor. Simulation should not only capture the resource allocation of each fetus at a specific moment, but also analyze the dynamic changes of resource supply throughout pregnancy. For example, as pregnancy enters the late stage, the needs of the fetus and the mother's supply capacity may change significantly. The fetus with relatively uneven resource distribution in the early stage may obtain more nutrition through the maternal compensatory mechanism in the later stage, thereby accelerating its development. This dynamic competition pattern can be quantified more deeply through simulation, revealing the nonlinear changes in fetal development. The developmental competition impact data generated by simulation processing not only quantifies the relative differences in resource acquisition for each fetus, but also reveals the long-term effects of resource utilization efficiency and nutrient absorption mechanisms on brain development in competition.

[0097] Step S520, analyzing the changes in brain function priority caused by resource and physiological demand limitations during the development process according to the brain development characteristic data, determining the function priority of each brain region, and generating function priority data;

[0098] It is understandable that, first of all, there are significant differences in the development speed of different brain regions, which is directly related to the physiological needs of fetal growth and the supply of maternal resources. For example, the fetal cerebral cortex (especially the brainstem and hypothalamus regions responsible for basic life maintenance) may occupy functional priority at an earlier stage of development because these regions are responsible for controlling the fetus's breathing, heartbeat and basic metabolic functions. The prefrontal cortex and hippocampus responsible for higher-level cognitive functions may have a relatively low priority when resources are tight because these brain regions are mainly gradually improved during later development. Therefore, functional priority analysis needs to combine the developmental characteristics of different brain regions with the physiological needs of the fetus to determine the development priority of each region. In addition, functional priority is also affected by resource allocation and absorption efficiency. For some fetuses, although they are at a disadvantage in the overall resource competition, some of their specific brain regions (such as the sensory cortex or motor cortex) may still be given priority due to functional urgency. This priority ranking is determined by the need for the fetus to respond to the external environment during development. For example, in the early stages of development, the brain regions responsible for perceiving the maternal environment and performing simple movements may obtain more resources in order to adapt to the complex environment in the mother as quickly as possible. This analysis can accurately infer the functional priorities of brain development by modeling the match between resource acquisition and functional requirements of each brain region. Through functional priority analysis, the functional priority data obtained can quantify the development order of each fetal brain region, revealing which brain regions are given priority development when resources are tight in the early stage and which brain regions may be delayed in development. For example, the occipital lobe (visual cortex) of a fetus may mature earlier due to priority allocation of resources, while the prefrontal region responsible for complex cognition and emotional processing is temporarily in a secondary position due to lack of resources. This analysis can effectively help explain the developmental strategy of the fetus when resources are limited, and reveal which functional areas may be limited by resource competition.

[0099] Step S530, performing data fusion processing according to the development competition impact data and the functional priority data, comprehensively considering the comprehensive effects of resource competition and functional priority on development, and generating a comprehensive feature data set;

[0100] It is understandable that during the fusion process, by comprehensively considering the distribution of these weights, it is ensured that the integrated dataset fully reflects the fetal brain development. This dataset provides deeper developmental information, which can more accurately reveal the complex interaction between resource competition and functional development of brain regions, and provide reliable data support for subsequent brain age prediction.

[0101] Step S540: Utilize the comprehensive feature data set and input it into a preset regression model for analysis to predict the actual brain age of each fetus and output the brain age prediction result.

[0102] It should be noted that the comprehensive feature data set contains multiple important developmental indicators, such as gray matter volume, white matter volume, cortical thickness, fractal dimension, developmental competition impact data, and functional priority data. These data together reflect the developmental status of the fetus in a multiple birth environment and its mutual competition and priority development in sharing maternal resources. In order to deal with these complex multidimensional features, the support vector regression (SVR) model is preferably used. The core of the model is the radial basis function (RBF) kernel, which can map low-dimensional input data to high-dimensional space, thereby capturing the nonlinear relationship between different fetal developmental characteristics and brain age. The kernel function formula is:

[0103] ;

[0104] in, and Respectively represent the serial numbers of two different data points (or samples) in the feature space; and Respectively represent Samples and The feature vector of samples; It is the parameter of the kernel function, which controls the width of the kernel function and determines the sensitivity of the model to similar points; Indicates that at the data point and The similarity function between .

[0105] The radial basis kernel function is obtained by Parameters are used to control the sensitivity of the mapping, ensuring that the model can flexibly handle complex dependencies in fetal development data. When dealing with significant developmental differences between fetuses in a multiple birth environment, weighted support vector regression is introduced so that fetuses with intense resource competition or high functional priority receive higher weights and thus have greater influence in the model. Specifically, the weights are based on the fetal development competition impact data ( ) and feature priority data ( ) to ensure that the weight of each fetus in the model can reflect the complexity of its developmental status. This weight adjustment is reflected in the optimization objective function of the model through the following formula:

[0106] ;

[0107] in, represents the weight vector; Represents the bias term, which is a constant used to adjust the predicted value of the regression model; represents the regularization parameter; Indicates the serial number of the sample; Indicates the total number of samples; Indicates the fetal serial number; Indicates the total number of fetuses; represents the weighting coefficient; Indicates The weights of the input features; Indicates the sequence number of the input feature; and represents the slack variable; and Indicates Slack variables for samples; Indicates The weighting factor for each fetus; Represents a composite function, used to describe the Interactive effects between resource competition and functional priority in each fetus.

[0108] During the optimization process, the model inputs these feature data into the regression model to calculate the optimal weight configuration, so that fetuses with resource allocation advantages or high developmental priorities can obtain more accurate results in brain age prediction. The goal of regression analysis is to minimize errors while capturing the potential relationship between developmental characteristics and brain age. Through the learning process of the model, it can not only identify the dominant areas in development, but also predict developmental problems caused by resource competition or priority differences. Ultimately, the brain age prediction data output by the support vector regression model provides a quantitative basis for the development level of each fetus, which can help clinicians detect potential developmental delays or abnormalities early. This step ensures the accuracy and effectiveness of brain age prediction by comprehensively considering the complex effects of resource competition and functional priority among multiple fetuses.

[0109] Embodiment 2:

[0110] like Figure 2 As shown, this embodiment provides a system for predicting brain age of multiple fetuses, the system comprising:

[0111] An acquisition module 1 is used to acquire original image data, where the original image data is magnetic resonance image data collected during the pregnancy of a pregnant woman with multiple births;

[0112] Segmentation module 2, used for performing image segmentation and motion artifact correction processing according to the original image data to obtain a separated image;

[0113] Restoration module 3 is used to reconstruct mutually occluded areas in the separated images, separate and compensate for signal interference between adjacent fetuses through a non-local mean algorithm, and restore and reconstruct the images in combination with the positional relationship of each fetus in three-dimensional space to generate a reconstructed image;

[0114] Extraction module 4, used for performing feature extraction processing according to the reconstructed image to obtain brain development feature data of each fetus, and performing difference analysis of relative development levels among multiple fetuses based on the brain development feature data to generate development difference data;

[0115] Prediction module 5 is used to perform fusion processing based on brain development feature data and development difference data, and to build a regression analysis model based on the fused data set to evaluate the actual brain age of each fetus and output the brain age prediction result.

[0116] In a specific embodiment disclosed in the present application, the segmentation module 2 includes:

[0117] The first segmentation unit 21 performs preliminary segmentation processing based on the original image data, analyzes the local gradient changes between the fetal brain and the surrounding tissues to distinguish the brain structures of multiple fetuses, defines the initial segmentation boundary, and generates a preliminary brain contour image of each fetus;

[0118] The second segmentation unit 22 performs segmentation processing on the preliminary brain contour image based on a preset Markov random field, and obtains an optimized brain image by modeling the probability of spatial correlation between pixels and optimizing the boundaries of multiple adjacent tissues;

[0119] A first correction unit 23 is used to perform spatial consistency correction processing according to the optimized brain image, evaluate the consistency of the fetal brain structure in multiple scanned images by comparing the similarity measurement of local blocks, and perform error adjustment to obtain an aligned brain image after error correction;

[0120] The first processing unit 24 is used to perform fetal brain boundary reconstruction processing according to the aligned brain image, preliminarily reconstruct the brain contour through the convex segmentation technology, and optimize the continuity and smoothness of the boundary by using the boundary constraint method to obtain a separated image.

[0121] In a specific embodiment disclosed in the present application, the repair module 3 includes:

[0122] The first analysis unit 31 is used to analyze the local brightness difference between adjacent fetal brain signals according to the separated images, and identify the occlusion area caused by the signal intensity mutation and spatial discontinuity in combination with the spatial continuity detection, and obtain a marked adjacent occlusion area map;

[0123] The first decomposition unit 32 is used to perform local feature decomposition processing according to the marked adjacent occlusion region map, analyze the sources of different signals in the occlusion region through gradient flow, distinguish the overlapping signals of the fetal brain and adjacent tissues, and obtain the occlusion region feature map after signal separation processing;

[0124] The first compensation unit 33 is used to perform tissue morphology compensation processing according to the characteristic map of the occluded area, use the brain morphology structure of the unoccluded area to perform multi-resolution morphology inference on the occluded area, gradually compensate for the missing morphological information of the occluded area, and obtain a preliminary repaired image;

[0125] The first restoration unit 34 is used to perform morphological restoration processing according to the preliminary restoration image, reconstruct the three-dimensional structure of the occluded area by using the three-dimensional morphological information of the unoccluded area and the known anatomical features of the fetal brain, and perform detail restoration under morphological constraints to obtain a final reconstructed image.

[0126] In a specific embodiment disclosed in the present application, the extraction module 4 includes:

[0127] The third segmentation unit 41 is used to perform adaptive morphological segmentation processing according to the reconstructed image, and dynamically adjust the segmentation threshold to extract the gray matter area and the white matter area of ​​the brain respectively by analyzing the spatial limitation of the multiple fetuses in the uterus and the asymmetry of the brain structure morphology, and calculate the gray matter volume data and the white matter volume data of each fetus;

[0128] A second analysis unit 42 is used to perform developmental delay analysis based on brain gray matter regions, establish a model of cortical thickness changes during normal development to identify abnormal thickness regions, and correct the abnormalities through dynamic compensation to generate cortical thickness distribution data for each fetus;

[0129] The first simulation unit 43 is used to perform simulation processing according to the cortical thickness distribution data, simulate the developmental competition of multiple fetuses under the condition of sharing maternal resources, and use fractal geometry to analyze the resource allocation efficiency between multiple fetuses, evaluate the developmental complexity of brain sulci, and obtain fractal dimension data of each fetus;

[0130] The third analysis unit 44 is used to perform data fusion and correlation analysis processing based on the gray matter volume data, white matter volume data, cortical thickness distribution data and fractal dimension data, and obtain developmental difference data by analyzing the developmental interactions between fetuses.

[0131] In a specific implementation disclosed in the present application, the prediction module 5 includes:

[0132] The second simulation unit 51 simulates the competition process of each fetus in blood supply and nutrient distribution based on the developmental difference data, quantitatively evaluates the specific impact of resource competition on the development of each fetus, and generates developmental competition impact data for each fetus;

[0133] A fourth analysis unit 52 is used to analyze the changes in brain function priority caused by resource and physiological demand limitations during the development process according to the brain development characteristic data, determine the function priority of each brain region, and generate function priority data;

[0134] A first fusion unit 53 is used to perform data fusion processing according to the development competition impact data and the functional priority data, comprehensively consider the comprehensive effects of resource competition and functional priority on development, and generate a comprehensive feature data set;

[0135] The first prediction unit 54 is used to use the comprehensive feature data set, input it into the preset regression model for analysis, predict the actual brain age of each fetus, and output the brain age prediction result. The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting brain age of multiple fetuses, characterized in that: include: Acquiring original image data, wherein the original image data is magnetic resonance image data collected during the pregnancy of a pregnant woman with multiple births; Performing image segmentation and motion artifact correction processing according to the original image data to obtain a separated image; Reconstructing mutually occluded areas in the separated images, separating and compensating for signal interference between adjacent fetuses using a non-local means algorithm, and restoring and reconstructing the images in combination with the positional relationship of each fetus in three-dimensional space to generate a reconstructed image; Performing feature extraction processing according to the reconstructed image to obtain brain development feature data of each fetus, and performing difference analysis of relative development levels among multiple fetuses based on the brain development feature data to generate development difference data; Performing fusion processing according to the brain development characteristic data and the development difference data, and building a regression analysis model based on the fused data set to evaluate the actual brain age of each fetus and output a brain age prediction result; The image segmentation and motion artifact correction processing are performed according to the original image data to obtain a separated image, including: Performing preliminary segmentation processing based on the original image data, distinguishing brain structures of multiple fetuses by analyzing local gradient changes between the fetal brain and surrounding tissues, defining initial segmentation boundaries, and generating a preliminary brain contour image of each fetus; The preliminary brain contour image is segmented based on a preset Markov random field, and an optimized brain image is obtained by modeling the probability of spatial correlation between pixels and optimizing the boundaries of multiple adjacent tissues; Performing spatial consistency correction processing according to the optimized brain image, evaluating the consistency of the fetal brain structure in multiple scanned images by comparing similarity metrics of local blocks, and performing error adjustment to obtain an aligned brain image after error correction; Performing fetal brain boundary reconstruction processing according to the aligned brain image, preliminarily reconstructing the brain contour by using a convex segmentation technique, and optimizing the continuity and smoothness of the boundary by using a boundary constraint method to obtain a separated image; The mutually occluded areas in the separated images are reconstructed, the signal interference between adjacent fetuses is separated and compensated by a non-local means algorithm, and the images are restored and reconstructed in combination with the positional relationship of each fetus in three-dimensional space to generate a reconstructed image, including: Analyzing the local brightness difference between adjacent fetal brain signals according to the separated images, and identifying the occlusion area caused by signal intensity mutation and spatial discontinuity in combination with spatial continuity detection, and obtaining a marked adjacent occlusion area map; Decomposing local features according to the marked adjacent occlusion region map, analyzing sources of different signals in the occlusion region through gradient flow, distinguishing overlapping signals between the fetal brain and adjacent tissues, and obtaining a feature map of the occlusion region after signal separation processing; Performing tissue morphology compensation processing according to the characteristic map of the occluded area, using the brain morphology structure of the unoccluded area to perform multi-resolution morphology inference on the occluded area, gradually compensating for the missing morphological information of the occluded area, and obtaining a preliminary repaired image; Performing morphological restoration processing according to the preliminary restoration image, reconstructing the three-dimensional structure of the occluded area by using the three-dimensional morphological information of the unoccluded area and the known anatomical features of the fetal brain, and performing detail restoration under morphological constraints to obtain a final reconstructed image; Among them, feature extraction processing is performed according to the reconstructed image to obtain brain development feature data of each fetus, and difference analysis of relative development levels among multiple fetuses is performed based on the brain development feature data to generate development difference data, including: Performing adaptive morphological segmentation processing according to the reconstructed image, by analyzing the spatial limitations of multiple fetuses in the uterus and the asymmetry of brain structure morphology, dynamically adjusting the segmentation threshold to extract gray matter areas and white matter areas respectively, and calculating gray matter volume data and white matter volume data of each fetus; Performing developmental delay analysis based on the brain gray matter regions, establishing a model of cortical thickness changes during normal development to identify abnormal thickness regions, and correcting the abnormalities through dynamic compensation to generate cortical thickness distribution data for each fetus; Performing simulation processing according to the cortical thickness distribution data, simulating the developmental competition of multiple fetuses in the case of sharing maternal resources, and using fractal geometry to analyze the efficiency of resource allocation between multiple fetuses, evaluating the developmental complexity of brain sulci and gyri, and obtaining fractal dimension data of each fetus; Data fusion and association analysis are performed based on the gray matter volume data, the white matter volume data, the cortical thickness distribution data and the fractal dimension data, and developmental difference data is obtained by analyzing the developmental interactions between fetuses.

2. The method for predicting brain age of multiple fetuses according to claim 1, characterized in that: The brain development feature data and the development difference data are fused and processed, and a regression analysis model is constructed based on the fused data set to evaluate the actual brain age of each fetus and output a brain age prediction result, including: Based on the developmental difference data, simulating the competition process of each fetus in blood supply and nutrient allocation, quantitatively evaluating the specific impact of resource competition on the development of each fetus, and generating developmental competition impact data for each fetus; Analyzing changes in brain function priority caused by resource and physiological demand limitations during development according to the brain development characteristic data, determining the function priority of each brain region, and generating function priority data; Performing data fusion processing according to the developmental competition impact data and the functional priority data, comprehensively considering the comprehensive effects of resource competition and functional priority on development, and generating a comprehensive feature data set; The comprehensive feature data set is used to input into a preset regression model for analysis, the actual brain age of each fetus is predicted, and the brain age prediction result is output.

3. A system for predicting brain age of multiple fetuses, characterized in that: include: An acquisition module, used to acquire original image data, wherein the original image data is magnetic resonance image data collected during the pregnancy of a pregnant woman with multiple births; A segmentation module, used for performing image segmentation and motion artifact correction processing according to the original image data to obtain a separated image; A restoration module, used to reconstruct mutually occluded areas in the separated images, separate and compensate for signal interference between adjacent fetuses through a non-local means algorithm, and restore and reconstruct the images in combination with the positional relationship of each fetus in three-dimensional space to generate a reconstructed image; An extraction module, configured to perform feature extraction processing based on the reconstructed image to obtain brain development feature data of each fetus, and to perform difference analysis of relative development levels among multiple fetuses based on the brain development feature data to generate development difference data; A prediction module, used to perform fusion processing according to the brain development feature data and the development difference data, and to construct a regression analysis model based on the fused data set, to evaluate the actual brain age of each fetus, and to output a brain age prediction result; Wherein, the segmentation module includes: A first segmentation unit performs preliminary segmentation processing based on the original image data, analyzes the local gradient changes between the fetal brain and surrounding tissues to distinguish the brain structures of multiple fetuses, defines initial segmentation boundaries, and generates a preliminary brain contour image of each fetus; A second segmentation unit is configured to segment the preliminary brain contour image based on a preset Markov random field, and obtain an optimized brain image by modeling the probability of spatial correlation between pixels and optimizing the boundaries of multiple adjacent tissues; A first correction unit is used to perform spatial consistency correction processing according to the optimized brain image, evaluate the consistency of the fetal brain structure in multiple scanned images by comparing the similarity measurement of local blocks, and perform error adjustment to obtain an aligned brain image after error correction; A first processing unit is used to perform fetal brain boundary reconstruction processing according to the aligned brain image, preliminarily reconstruct the brain contour by using a convex segmentation technique, and optimize the continuity and smoothness of the boundary by using a boundary constraint method to obtain a separated image; Wherein, the repair module includes: A first analysis unit is used to analyze the local brightness difference between adjacent fetal brain signals according to the separated image, and identify the occlusion area caused by the signal intensity mutation and spatial discontinuity in combination with the spatial continuity detection, and obtain a marked adjacent occlusion area map; A first decomposition unit is used to perform local feature decomposition processing according to the marked adjacent occlusion region map, analyze the sources of different signals in the occlusion region through gradient flow, distinguish the overlapping signals between the fetal brain and adjacent tissues, and obtain the occlusion region feature map after signal separation processing; A first compensation unit is used to perform tissue morphology compensation processing according to the occluded area feature map, use the brain morphology structure of the unoccluded area to perform multi-resolution morphology inference on the occluded area, gradually compensate for the missing morphological information of the occluded area, and obtain a preliminary repaired image; A first restoration unit is used to perform morphological restoration processing according to the preliminary restoration image, reconstruct the three-dimensional structure of the occluded area by using the three-dimensional morphological information of the unoccluded area and the known anatomical features of the fetal brain, and perform detail restoration under morphological constraints to obtain a final reconstructed image; Wherein, the extraction module comprises: A third segmentation unit is used to perform adaptive morphological segmentation processing according to the reconstructed image, analyze the spatial limitations of multiple fetuses in the uterus and the asymmetry of brain structure morphology, dynamically adjust the segmentation threshold to extract gray matter areas and white matter areas respectively, and calculate the gray matter volume data and white matter volume data of each fetus; A second analysis unit is used to perform developmental delay analysis based on the brain gray matter area, establish a model of cortical thickness changes in normal development stages to identify abnormal thickness areas, and correct the abnormalities through dynamic compensation to generate cortical thickness distribution data for each fetus; A first simulation unit is used to perform simulation processing according to the cortical thickness distribution data, simulate the developmental competition of multiple fetuses under the condition of sharing maternal resources, analyze the resource allocation efficiency between multiple fetuses by using fractal geometry, evaluate the developmental complexity of brain sulci, and obtain fractal dimension data of each fetus; The third analysis unit is used to perform data fusion and association analysis based on the gray matter volume data, the white matter volume data, the cortical thickness distribution data and the fractal dimension data, and obtain developmental difference data by analyzing the developmental interactions between fetuses.

4. The system for predicting brain age of multiple fetuses according to claim 3, characterized in that: The prediction module comprises: A second simulation unit simulates the competition process of each fetus in blood supply and nutrient allocation based on the developmental difference data, quantitatively evaluates the specific impact of resource competition on the development of each fetus, and generates developmental competition impact data for each fetus; a fourth analysis unit, configured to analyze changes in brain function priorities caused by resource and physiological demand limitations during the development process according to the brain development characteristic data, determine the function priority of each brain region, and generate function priority data; The first fusion unit is used to perform data fusion processing based on the developmental competition impact data and the functional priority data, comprehensively consider the comprehensive effects of resource competition and functional priority on development, and generate a comprehensive feature data set; the first prediction unit is used to use the comprehensive feature data set, input it into a preset regression model for analysis, predict the actual brain age of each fetus, and output the brain age prediction result.

Images