Fetal magnetic resonance image brain region segmentation method and system, computer equipment and medium

Abstract

The invention provides a fetal magnetic resonance image brain region segmentation method and system, computer equipment and a medium, and belongs to the technical field of automatic segmentation of fetal brain magnetic resonance imaging. The method comprises the steps that firstly, 2D low-resolution fetal magnetic resonance images collected in multiple directions are processed, and 3D high-resolution fetal brain images are generated; secondly, performing selective interlayer labeling on the 3D image, and generating a complete segmentation label through a contour interpolation algorithm; and finally, performing mutual supervision learning by using a double-independent initialized segmentation network, including the steps of data enhancement consistency constraint, cross pseudo label supervision, feature comparison learning and the like. According to the fetal magnetic resonance image brain region segmentation method and system, the computer equipment and the medium, a high-performance divider can be trained only through a very few labels, the labeling cost is remarkably reduced, the segmentation efficiency is improved, the method and system are suitable for actual clinical fetal brain magnetic resonance research, and efficient technical support is provided for fetal brain development evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of automated segmentation of fetal brain magnetic resonance imaging (MRI), and in particular to a method, system, computer equipment and medium for segmenting brain regions in fetal MRI images. Background Art

[0002] Fetal MRI is a core tool for assessing fetal brain development, particularly when ultrasound examinations lack sufficient structural detail. This technology can identify congenital brain abnormalities, such as cortical malformations and ventricular dilatation, at an early stage, providing critical decision-making for clinical intervention planning and prognostic assessment. Precise segmentation of brain tissue (including gray matter, white matter, and cerebrospinal fluid) is essential for accurate quantitative analysis. This is also the foundation for constructing fetal brain maps and conducting developmental research. However, current manual segmentation relies on radiologists labeling each layer, which is not only time-consuming and laborious, but can also lead to subjective factors that affect the consistency of results. Automated segmentation methods based on deep learning hold great potential for the precise analysis of fetal brain magnetic resonance imaging (MRI). However, existing mainstream automated deep learning-based methods rely on supervised learning models, which require extensive labeled data. In the field of fetal brain MRI, obtaining labeled data presents numerous challenges, severely hindering the widespread application of these automated segmentation methods in real-world clinical data. Summary of the Invention

[0003] The purpose of the present invention is to provide a method, system, computer equipment and medium for fetal magnetic resonance image brain region segmentation. When used in actual clinical fetal brain magnetic resonance research, only a very small number of labels need to be marked according to the process to train a fetal brain region segmentor with excellent segmentation performance, avoiding the heavy reliance on labels of existing supervised learning networks when applied to actual clinical datasets.

[0004] To achieve the above object, the present invention provides a method for brain region segmentation in fetal magnetic resonance images, comprising the following steps: Step S1: Processing the 2D low-resolution fetal magnetic resonance images collected from multiple directions to generate a 3D high-resolution fetal brain image; Step S2: selectively labeling the 3D high-resolution fetal brain image generated in step S1 between layers, and generating complete segmentation labels using a contour interpolation algorithm; Step S3: Based on the 3D high-resolution fetal brain image of step S1 and the segmentation label of step S2, a dual independently initialized deformer segmentation network that can utilize a long-range attention mechanism is used for mutual learning.

[0005] Preferably, in step S1, processing the 2D low-resolution fetal magnetic resonance images acquired from multiple angles to generate a 3D high-resolution fetal brain image includes the following steps: Resampling the 2D low-resolution fetal magnetic resonance images acquired from multiple angles to reduce the resolution; A localization convolutional neural network consisting of three convolutional layers and one pooling layer is used to perform preliminary brain localization on the 2D low-resolution fetal MRI and generate a segmentation mask. Perform 3D morphological closing and opening operations on the segmentation mask in sequence; Select the largest connected component and fit the 3D bounding cube, then resample the 3D bounding cube to the original resolution to obtain the final positioning result; Based on the positioning results, a segmentation convolutional neural network with a multi-scale Dice loss function is used to extract the fetal brain. The segmentation convolutional neural network includes a three-layer convolutional encoder and a three-layer convolutional decoder: ; in, represents the multi-scale total loss function, represents the Dice loss function, S Indicates different zoom scales; express s The predicted label of the scale, express s The true label of the scale, represents the original predicted label, represents the original true label; Perform bias field correction on the acquired 2D low-resolution fetal magnetic resonance images; Registering the rectified image to the selected orientation by symmetric block matching; Based on the registration results, Nadaraya-Watson kernel regression is used to estimate the initial 3D high-resolution fetal brain image: ; in, Represents the coordinates of the point to be predicted, Represents known data point coordinates and values, represents the kernel function; Indicates the point to be predicted The kernel regression estimate at , represents the label of the known data point, Indicates the number of known data points; The final 3D high-resolution fetal brain image is obtained by iteratively solving the outlier-robust optimization problem: ; in, The first image represents the actual acquired 2D low-resolution fetal magnetic resonance image. k Layer data, represents the forward operator of the acquisition process, Indicates the 3D high-resolution fetal brain image that we want to solve. represents the regularization parameter, represents the difference function.

[0006] Preferably, in step S2, selective inter-layer annotation is performed on the 3D high-resolution fetal brain image generated in step S1, and a complete segmentation label is generated by a contour interpolation algorithm, comprising the following steps: Step S21: Based on the 3D high-resolution fetal brain image, select key layers according to the anatomical features of the brain region; Step S22: extract the contour boundaries of two adjacent marked key layers to obtain the original key layer contours; Step S23: Calculate the signed distance maps of the two layers based on the extracted contour boundaries; Step S24: For the middle layer between two adjacent labeled key layers, the two signed distance maps obtained in step S23 are fused according to the inter-layer position ratio; Step S25: extracting the zero distance boundary from the fused signed distance map as the middle layer contour; Step S26: Combine all generated intermediate layer contours with the original key layer contours, perform a closing operation on the combined complete contour sequence, and output the final segmentation label for training the segmentation network.

[0007] Preferably, in step S3, a mutually supervised learning is performed using a dual independently initialized deformer segmentation network that can utilize a long-range attention mechanism, comprising the following steps: Step S31: deploy two independently initialized deformer segmentation networks with the same structure and capable of utilizing a long-range attention mechanism for mutual learning; Step S32: Implement collaborative training on unlabeled 3D high-resolution fetal brain images: Step S321: Optimizing network robustness through data enhancement consistency constraints; Step S322: Implementing network mutual learning through cross-pseudo-label supervision; Step S323: Enhance the encoder's discriminative power through feature contrast learning; Step S33: Jointly optimize the loss of step S32 and output the model.

[0008] Preferably, the data enhancement consistency constraints include: Apply weak enhancement and strong enhancement operations to the same image respectively; Constrain the segmentation results of the dual network to be consistent with the weak enhancement and strong enhancement outputs.

[0009] Preferably, cross-pseudo-label supervision includes: Use the prediction results of the first network as the supervision label of the second network; Use the second network's prediction results as the supervision label of the first network; Cross entropy loss and Dice loss are used to calculate the supervision error.

[0010] Preferably, feature contrast learning includes: Extract feature vectors at the encoder output layer; Minimize the distance between features of similar anatomical structures and maximize the distance between features of different structures.

[0011] The present invention also provides a fetal magnetic resonance image brain region segmentation system, comprising: An image processing module is used to process 2D low-resolution fetal magnetic resonance images collected from multiple angles to generate 3D high-resolution fetal brain images; The annotation module is used to selectively annotate the generated 3D high-resolution fetal brain images between layers and generate complete segmentation labels through contour interpolation algorithm; A segmentation network module, which uses dual independently initialized deformer segmentation networks with long-range attention for mutually supervised learning based on 3D high-resolution fetal brain images and generated labeled data; A data augmentation module is used to enforce data augmentation consistency constraints on unlabeled 3D high-resolution fetal brain images; Pseudo-label supervision module, used to achieve network mutual learning through cross-pseudo-label supervision; Feature contrast learning module, used to enhance the encoder's discriminative power through feature contrast learning; Output module, used to output the final segmentation results.

[0012] The present invention also provides a computer device comprising a memory and a processor, wherein the memory is used to store instructions, and the processor is used to execute the instructions to implement the above-mentioned method for brain region segmentation in fetal magnetic resonance images.

[0013] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-mentioned method for brain region segmentation in fetal magnetic resonance images.

[0014] Therefore, the present invention adopts the above-mentioned fetal magnetic resonance image brain region segmentation method, system, computer equipment and medium, which has beneficial technical effects: when used in actual clinical fetal brain magnetic resonance research, only a very small number of labels need to be marked according to the process to train a fetal brain region segmentor with excellent segmentation performance, avoiding the existing supervised learning network's heavy reliance on labels when applied to actual clinical data sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of the method for brain region segmentation of fetal magnetic resonance images according to the present invention; Figure 2 The figure shows a comparison between the segmentation results obtained by the method of the present invention and the segmentation results of U-Net in a real clinical scenario with very little labeled data. DETAILED DESCRIPTION

[0016] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0017] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0018] Example 1 like Figure 1 As shown, the method for brain region segmentation of fetal magnetic resonance images includes the following steps: Step S1: Processing 2D low-resolution fetal magnetic resonance images (slice thickness 3-5 mm) acquired in sagittal, coronal, and axial positions to generate 3D high-resolution fetal brain images (slice thickness 0.5 mm), including the following steps: Resampling the 2D low-resolution fetal magnetic resonance images acquired in the sagittal, coronal, and axial planes to reduce resolution and reduce subsequent resource consumption; A localization convolutional neural network consisting of three convolutional layers and one pooling layer is used to perform preliminary brain localization on the 2D low-resolution fetal MRI, excluding a large amount of maternal tissue and generating a segmentation mask. Perform 3D morphological closing and opening operations on the segmentation mask in sequence; Select the largest connected component and fit the 3D bounding cube, then resample the 3D bounding cube to the original resolution to obtain the final positioning result; Based on the final positioning results, a segmentation convolutional neural network with a multi-scale Dice loss function is used to extract the fetal brain. The segmentation convolutional neural network includes a three-layer convolutional encoder and a three-layer convolutional decoder: ; in, represents the multi-scale total loss function, represents the Dice loss function, S Indicates different zoom scales; express s The predicted label of the scale, express s The true label of the scale, represents the original predicted label, represents the original true label.

[0019] Registering the rectified image to the selected orientation by symmetric block matching; Based on the registration results, Nadaraya-Watson kernel regression is used to estimate the initial 3D high-resolution fetal brain image: ; in, Represents the coordinates of the point to be predicted, Represents known data point coordinates and values, represents the kernel function; Indicates the point to be predicted The kernel regression estimate at , represents the label of the known data point, Indicates the number of known data points; The final 3D high-resolution fetal brain image is obtained by iteratively solving the outlier-robust optimization problem: ; in, The first image represents the actual acquired 2D low-resolution fetal magnetic resonance image. k Layer data, represents the forward operator of the acquisition process, Indicates 3D high-resolution fetal brain images, represents the regularization parameter, represents the difference function.

[0020] Step S2: selectively label the 3D high-resolution fetal brain image generated in step S1 between layers, and generate complete segmentation labels using a contour interpolation algorithm, including the following steps: Step S21: Based on the 3D high-resolution fetal brain image, a professional physician selects key layers according to the anatomical features of the brain region; Step S22: extract the contour boundaries of two adjacent marked key layers to obtain the original key layer contours; Step S23: Based on the extracted contour boundaries of adjacent slices (slice A and slice B), the signed distance maps (SDMs) of the two layers are calculated respectively: ; in, represents the contour boundary, d Represents the Euclidean distance. When it is outside the contour, SDM is negative. Indicates the horizontal and vertical coordinates of the corresponding points.

[0021] Step S24: For the middle layer between two adjacent labeled key layers, the two signed distance maps obtained in step S23 are fused according to the inter-layer position ratio; ; ; in, The ratio is t The interlayer coordinates of the middle layer at Represent the inter-layer coordinates of slice A and slice B respectively, Respectively represent the proportion t The signed distance diagram of slice A and slice B.

[0022] Step S25: extracting the zero distance boundary from the fused signed distance map as the middle layer contour; Step S26: Combine all generated intermediate layer contours with the original key layer contours, perform a closing operation on the combined complete contour sequence, and output the final segmentation label for training the segmentation network.

[0023] Step S3: Based on the 3D high-resolution fetal brain image from step S1 and the segmentation labels from step S2, a dual independently initialized deformer segmentation network that can utilize a long-range attention mechanism is used for mutual supervision learning, including the following steps: Step S31: deploy two independently initialized deformer segmentation networks with the same structure and capable of utilizing the long-range attention mechanism ( and ) to learn from each other; Step S32: Implement collaborative training on unlabeled 3D high-resolution fetal brain images: Step S321: Optimizing network robustness through data enhancement consistency constraints; The same input image is subjected to strong and weak enhancement respectively. Weak enhancement includes operations such as rotation and flipping to expand the diversity of the dataset. Strong enhancement includes intensity transformation and affine transformation. Constrain the dual network to ensure the consistency of segmentation results for weak enhancement and strong enhancement outputs; Step S322: Implementing network mutual learning through cross-pseudo-label supervision; The two networks generate pseudo labels for each other for supervised training, i.e. The network's prediction for this instance is The label of the network output, and vice versa, is constructed on this basis, namely: ; in, represents the cross entropy loss function, represents the pseudo-label generated by the output of the first network, represents the output prediction probability of the second network; represents the semi-supervised loss function, represents the pseudo-label generated by the output of the second network, represents the output prediction probability of the second network; Step S323: Enhance the encoder's discriminative power through feature contrast learning; A feature contrast loss function is applied at the encoder level to make similar features compact and inter-class features separate. The contrast loss function is in the form of Information Noise-Contrastive Estimation (InfoNCE): ; in, q represents the feature vector of the query sample, represents the positive sample feature, represents the negative sample set, Represents the similarity function dot product, represents the coefficient that controls the degree of distribution concentration, Indicates the negative sample labels, Indicates the number of negative samples.

[0024] Step S33: Jointly optimize the loss of step S32 and output the model.

[0025] The present invention will be further described below through specific examples.

[0026] The imaging device uses a 1.5T magnetic resonance scanning device with the following parameters: voxel size (resolution): 0.7 × 0.7 × 3 mm 3 , Field of View (FOV) 360, Echo Time (TE): 90-110ms, Recovery Time (TR): 1300-1400ms.

[0027] Eighty multi-directional 2D single-shot fast spin echo (SSFSE) images were collected. Brain extraction, brain localization, and iterative reconstruction were performed on this data set to produce 80 3D high-resolution fetal brain images. Two randomly selected data sets were then assigned to a professional physician for n-row brain region annotation. Contour filling was then used to obtain two semi-supervised labels. This two-labeled training data set was then fed into a semi-supervised network training program to produce the final automated fast segmentation model.

[0028] like Figure 2As shown in the figure, in the case of few labels, the method of the present invention is compared with the currently mainstream deep learning segmentation method U-Net (the left side is the mainstream method, and the right side is the method proposed by the present invention). It can be seen that the method of the present invention is significantly better than the traditional method in the case of few training labels.

[0029] Example 2 Fetal MRI brain region segmentation system, including: An image processing module is used to process 2D low-resolution fetal magnetic resonance images collected from multiple angles to generate 3D high-resolution fetal brain images; The annotation module is used to selectively annotate the generated 3D high-resolution fetal brain images between layers and generate complete segmentation labels through contour interpolation algorithm; A segmentation network module, which uses dual independently initialized deformer segmentation networks with long-range attention for mutually supervised learning based on 3D high-resolution fetal brain images and generated labeled data; A data augmentation module is used to enforce data augmentation consistency constraints on unlabeled 3D high-resolution fetal brain images; Pseudo-label supervision module, used to achieve network mutual learning through cross-pseudo-label supervision; Feature contrast learning module, used to enhance the encoder's discriminative power through feature contrast learning; Output module, used to output the final segmentation results.

[0030] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0031] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0032] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, and then editing, interpreting, or processing in another suitable manner as necessary, and then storing it in a computer memory.

[0033] It is worth noting that the contents not elaborated in detail in the present invention are all prior art and are well known to those skilled in the art.

[0034] Therefore, the present invention adopts the above-mentioned fetal magnetic resonance image brain region segmentation method, system, computer equipment and medium, and only requires very few labels to train a high-performance segmenter, which significantly reduces the labeling cost and improves the segmentation efficiency. It is suitable for actual clinical fetal brain magnetic resonance research and provides efficient technical support for fetal brain development assessment.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for brain region segmentation in fetal magnetic resonance images, characterized in that: The following steps are involved: Step S1: Processing the 2D low-resolution fetal magnetic resonance images collected from multiple directions to generate a 3D high-resolution fetal brain image; Step S2: selectively labeling the 3D high-resolution fetal brain image generated in step S1 between layers, and generating complete segmentation labels using a contour interpolation algorithm; Step S3: Based on the 3D high-resolution fetal brain image of step S1 and the segmentation label of step S2, a dual independently initialized deformer segmentation network that can utilize a long-range attention mechanism is used for mutual learning.

2. The method for brain region segmentation of fetal magnetic resonance images according to claim 1, characterized in that: In step S1, the 2D low-resolution fetal magnetic resonance images acquired from multiple directions are processed to generate a 3D high-resolution fetal brain image, including the following steps: Resampling the 2D low-resolution fetal magnetic resonance images acquired from multiple angles to reduce the resolution; A localization convolutional neural network consisting of three convolutional layers and one pooling layer is used to perform preliminary brain localization on the 2D low-resolution fetal MRI and generate a segmentation mask. Perform 3D morphological closing and opening operations on the segmentation mask in sequence; Select the largest connected component and fit the 3D bounding cube, then resample the 3D bounding cube to the original resolution to obtain the final positioning result; Based on the positioning results, a segmentation convolutional neural network with a multi-scale Dice loss function is used to extract the fetal brain. The segmentation convolutional neural network includes a three-layer convolutional encoder and a three-layer convolutional decoder: ； in, represents the multi-scale total loss function, represents the Dice loss function, S Indicates different zoom scales; express s The predicted labels of the scale, express s The true label of the scale, represents the original predicted label, represents the original true label; Perform bias field correction on the acquired 2D low-resolution fetal magnetic resonance images; Registering the rectified image to the selected orientation by symmetric block matching; Based on the registration results, Nadaraya-Watson kernel regression is used to estimate the initial 3D high-resolution fetal brain image: ； in, Represents the coordinates of the point to be predicted, Represents known data point coordinates and values, represents the kernel function; Indicates the point to be predicted The kernel regression estimate at , represents the label of the known data point, Indicates the number of known data points; The final 3D high-resolution fetal brain image is obtained by iteratively solving the outlier-robust optimization problem: ； in, The first image represents the actual acquired 2D low-resolution fetal magnetic resonance image. k Layer data, represents the forward operator of the acquisition process, Indicates the 3D high-resolution fetal brain image that we want to solve. represents the regularization parameter, represents the difference function.

3. The method for brain region segmentation of fetal magnetic resonance images according to claim 1, characterized in that: In step S2, the 3D high-resolution fetal brain image generated in step S1 is selectively annotated between layers, and a complete segmentation label is generated using a contour interpolation algorithm, including the following steps: Step S21: Based on the 3D high-resolution fetal brain image, select key layers according to the anatomical features of the brain region; Step S22: extract the contour boundaries of two adjacent marked key layers to obtain the original key layer contours; Step S23: Calculate the signed distance maps of the two layers based on the extracted contour boundaries; Step S24: For the middle layer between two adjacent labeled key layers, the two signed distance maps obtained in step S23 are fused according to the inter-layer position ratio; Step S25: extracting the zero distance boundary from the fused signed distance map as the middle layer contour; Step S26: Combine all generated intermediate layer contours with the original key layer contours, perform a closing operation on the combined complete contour sequence, and output the final segmentation label for training the segmentation network.

4. The method for brain region segmentation of fetal magnetic resonance images according to claim 1, characterized in that: In step S3, a mutually supervised learning is performed using a dual independently initialized deformer segmentation network that can utilize a long-range attention mechanism, including the following steps: Step S31: deploy two independently initialized deformer segmentation networks with the same structure and capable of utilizing a long-range attention mechanism for mutual learning; Step S32: Implement collaborative training on unlabeled 3D high-resolution fetal brain images: Step S321: Optimizing network robustness through data enhancement consistency constraints; Step S322: Implementing network mutual learning through cross-pseudo-label supervision; Step S323: Enhance the encoder's discriminative power through feature contrast learning; Step S33: Jointly optimize the loss of step S32 and output the model.

5. The method for brain region segmentation of fetal magnetic resonance images according to claim 4, characterized in that: Data enhancement consistency constraints include: Apply weak enhancement and strong enhancement operations to the same image respectively; Constrain the segmentation results of the dual network to be consistent with the weak enhancement and strong enhancement outputs.

6. The method for brain region segmentation in fetal magnetic resonance images according to claim 4, characterized in that: Cross-pseudo-label supervision includes: Use the prediction results of the first network as the supervision label of the second network; Use the second network's prediction results as the supervision label of the first network; Cross entropy loss and Dice loss are used to calculate the supervision error.

7. The method for brain region segmentation in fetal magnetic resonance images according to claim 4, characterized in that: Feature contrastive learning includes: Extract feature vectors at the encoder output layer; Minimize the distance between features of similar anatomical structures and maximize the distance between features of different structures.

8. Fetal MRI brain region segmentation system, characterized by: include: An image processing module is used to process 2D low-resolution fetal magnetic resonance images collected from multiple angles to generate 3D high-resolution fetal brain images; The annotation module is used to selectively annotate the generated 3D high-resolution fetal brain images between layers and generate complete segmentation labels through contour interpolation algorithm; A segmentation network module, which uses dual independently initialized deformer segmentation networks with long-range attention for mutually supervised learning based on 3D high-resolution fetal brain images and generated labeled data; A data augmentation module is used to enforce data augmentation consistency constraints on unlabeled 3D high-resolution fetal brain images; Pseudo-label supervision module, used to achieve network mutual learning through cross-pseudo-label supervision; Feature contrast learning module, used to enhance the encoder's discriminative power through feature contrast learning; Output module, used to output the final segmentation results.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory is used to store instructions, and the processor is used to execute the instructions to implement the method for brain region segmentation in a fetal magnetic resonance image according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for brain region segmentation of a fetal magnetic resonance image according to any one of claims 1 to 7 is implemented.

Images